Home

— User`s Manual —

1. X 2 2 File Object Series Figure 2 3 Objects and their interrelationship sequences Scalar Snapshot Series and File are objects expecting numerical values The characteristics of classes are inherited in the order of listing taking Scalar as a super class The Snapshot object is a set of Scalar objects equivalent to the multi dimensional arrangements for general purpose languages The Series object can be viewed as a series of Snapshots in time that is 1 dimensional arrangement of Snapshot objects The File object requires a file name for handling the specified data on the UNIX system Each object encapsulates data and processing methods Arithmetical operations are different for Scalar Series Snapshot String and File objects In case of the Scalar object the addition is performed by adding up only 1 point data as shown in Figure 2 4 e e Wa Figure 2 4 Addition between two Scalar objects In case of the Series object all values on the time axis must be added simultaneously as shown in Figure 2 5 gt he s gt Time gt Figure 2 5 Addition between two Series objects 5 kinds of object classes used in SATELLITE are explained in the pages that follow CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 16 Figure 2 6 Data organization in 1 dimensional Series object Series object Series is the object class in which a set of multi dime
2. CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS The limit number of available file calls is 10 32 Chapter 3 SYSTEM Module SYSTEM SYSTEM module is a gathering of basic functions for handling data It includes the functions for extracting a subset of data finding a maximum or minimum of a sequence modifying data format displaying header information of data files etc They are illustrated in the following subsections 3 1 HELP displaying a command manual Display the on line reference of SATELLITE commands usage help com_name com_name stands for a command name It needs to be put in double quotation marks For example the explanation of HELP command is displayed by the following SATELLITE tom rose 63 help help 3 2 HEADER displaying or modifying the header informa tion of a data file This command is for confirmation or alteration of a data file header information such as data format index etc usage header file_name mode file_name stands for a file name and mode is the integer that selects the mode 0 display 1 modify The following example displays information about the data file test dat on a display JSATELLITE tom rose 64 header test dat 0 3 3 WAIT interrupting the execution of a program It pauses the batch processing until CR key is pressed usage wait 33 CHAPTER 3 SYSTEM MODULE SYSTEM 34 Figure 3 1 An exampl
3. Pre correction amount n lt J Pre correction amount n 1 Pre correction amount n 2 History n History n 1 History n 2 Weight data n 2 Pre correction amount n 2 Figure 6 7 Weight history file format in Overwrite mode Error Error Error History 1 Unit 1 Unit2 Unitn Total Error History 2 Error Error Error Total Error History n Error Error ee Eror Total Error Figure 6 8 Format of error history file Record direction mode result file does not exist If the file exists the mode of overwriting is confirmed The message that the overwriting has been performed is displayed if y is pressed The command will be terminated and the file name must be reset when n is chosen 6 3 6 Tracing connection weights and errors By combining the commands of BPS with the GPM s ones the trace of the internal parameters of MLP can be performed The data sets which are necessary for tracing are read from the weight history file the error history file and the test result file generated during the learning and testing of MLP using the following commands errload Data are read from the error history file wgtload Data are read from the weight history file actload Data are read from the test result file As previously mentioned the WOPEN command is used for opening graphic windows By using GPM commands such as CONT GRAPH GSOLM and MAP it is possible to display the data in
4. 2 2 output fir coef The source signal is smoothed by using the FIR command First the coefficient of the filter coef is set 2 3 output fir coef input The original signal input is set By the above procedure the smoothed data is stored in the Series object output The FIR command carries out the processing shown in Figure 4 6 Therefore the number of the filter coefficients the order of the filter must be the odd number it is assumed that this value is equal to 2n 1 Besides the original data form the n 1 th point is used for filtering because the data from the beginning to the n th point is not possible to deal with precisely Similarly the final data the ones to the n 1 th point from behind is ignored CHAPTER 4 INTERACTIVE SIGNAL PROCESSING PACKAGE ISPP 58 Input Series Time Start End Figure 4 6 Filtering by the FIR command a1 2 5 are the filter coefficients FIR filter When the FIR filter of the low pass high pass or band pass type is designed by the window function method the filtering by FIR is carried out after the filter coefficient is obtained using the FIRMAKE command The example of a low pass filter is shown below 1 sam 1024 The sampling frequency is set 2 1 coef The Series object for storing the filter coefficients is set 2 2 coef firmake 1 The filter is set to be the low pass type 1 Low pass 2 High pass 3 Band pass 2 3 coef fir
5. 01 30 31 CHAPTER 3 SYSTEM MODULE SYSTEM 40 Figure 3 13 An example of using MERGE on 2 dimensional Series object 5 01 32 33 6 01 34 35 SATELLITE tom rose 351 Similarly Snapshot objects 3 10 FILL filling data with a specified value This command fills a part of an object with a particular value usage y fill x start end value x is the original object y is the final object and start and end are the start and end points for filling the value All range in x specified by start and end is filled with the same value value If we want to fill a part of 1 dimensional Series object a with 20 as shown in Figure 3 14 we proceed as follows SATELLITE tom rose 351 a 177 SATELLITE tom rose 361 b fi11 a 3 5 20 SATELLITE tom rose 37 b 0 1 2 3 20 20 5 20 7 SATELLITE tom rose 38 The example of filling 2 dimensional Series object follows see also Figure 3 15 SATELLITE tom rose 381 a 1714 SATELLITE tom rose 39 b reform a 7 2 SATELLITE tom rose 40 c f i11 b 3 0 5 0 30 SATELLITE tom rose 411 c Lo 01 1 2 1 01 3 4 2 01 5 6 3 01 30 8 4 01 30 10 5 07 30 12 6 01 13 14 SATELLITE tom rose 421 Similarly the operation can be performed on Snapshot objects 3 11 ZERO filling data with 0 This command fills a part of a
6. OOF ep Oe ON tes er et a type sentence module sentence input sentence output sentence observable sentence constant sentence parameter sentence function sentence Although it is possible to omit some sentences their order cannot be changed See the previous section for the details on each sentence Put end word at the end of the module description Example gap type module type GAP module G input VOP 0 1 0 output Ig parameter GL 5 0 function Ig GL VOP POSOUT end Network description The modules used in the model are defined and the properties of each component are described using mathematical expressions The sentences used in the network type and their order are shown in the following Do OY A type sentence module sentence cell sentence synapse sentence gap sentence connection sentence Although it is possible to omit some sentences their order cannot be changed See the previous section for the details on each sentence Put end word at the end of the module description CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 100 FOR sentences can be used in order to incorporate loops The format of the FOR sentence is shown in the following for expri expr2 expr3 4 sentences The expressions such as substitution relative operator and so on can take the place of expr1 expr2 and expr expri is an expression for initializing the loop
7. Table 7 3 Operators for conditional expressions Operation Operator parity disparity lt gt comparison lt gt lt gt logical OR OR logical AND AND negation NOT if V gt 10 IK GK V VK elsef IK 0 14 Comment The part contained in and is a comment sentence It is possible to use it anywhere and exceed two or more lines Description of cell type modules The cell can be described using a membrane model based on the ionic currents The sentences used in the cell type module and their order are as follows type sentence module sentence exinput sentence input sentence output sentence observable sentence constant sentence parameter sentence SOE OR EV Ba OY i Coes i function sentence Although it is possible to omit some sentences their order cannot be changed See the previous section for the details on each sentence The end word is required at the end of the module description Example cell type module type CELL module CL exinput lex input lg output V CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 98 observable IK 11 Ig nk constant VK 12 0 Vl 10 6 parameter Cm 1 0 GK 36 0 Gl 0 3 function if V 10 an 0 01 10 V exp 10 V 10 1 else an 0 01 10 bn 0 125 exp V 80 dnK an 1 nK bn nK nk integral 0 3177 dnK IK GK pow n
8. in SATELLITE From the standpoint of data analysis visualization of data is much more important than numerical evaluation The GPM module provides various graphic functions for making charts useful for writing articles or presentation The GPM module consists of about 30 commands which are divided into the following two categories e Commands for drawing graphic charts e Parameter setting commands The main commands are described in Table 5 1 The commands for drawing can also be classified as follows e Commands for displaying 1 dimensional objects e Commands for displaying 2 dimensional objects such as contour maps bird s eye pictures etc Many parameters such as line type width color etc are needed in order to draw pictures They can be set up by the commands such as LTYPE LWIDTH or COLOR Even if one do not know how to use GPM commands exactly it is possible to make beautiful charts by using the online message function see 2 9 1 The parameters related to drawing are initialized to default values at the time of starting SATELLITE 5 2 Drawing and Printing WOPEN command is used for opening a window for drawing figures or charts Conversely the command for closing it is WCLOSE Many windows can be opened CHWIN specifies the target window It is not allowed to draw charts on two or more windows simultaneously WE is the command for erasing graphics in the window After making the charts by GPM commands they can be pre
9. and the definitions of user modules commands aliases sampling frequency the functions used often and the variables used in the user setup file This system is terminated by typing either close exit or Ctrl D shown as follows SATELLITE home tom 1 close SATELLITE home tom 1 exit SATELLITE home tom 1 Right after terminating the user clean file clean sl is executed Then the history is saved to the file history sl after execution of the closing commands of system modules release of the system common area shared memory destruction of system parameter area temporary directory dispatch of the end signals to all child processes etc are performed The options for starting SATELLITE are as follows read a program from a standard input terminal rc do not read the system rc file setup do not read the user setup file clean do not read the user clean file CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 10 SATELLITE Language titan lamdifasihometkunita DEN hone kunita 63 3 T Figure 2 2 Title of the window while using SATELLITE log specify a directory name for the error log file work specify a directory name for the work domain system parameter area The work directory SLxxxx is deleted after termination temp same as work Moreover if we have another file to read automatically besides the user setup file we can specify it as follows sl setu
10. c 0 01 4 0 SATELLITE tom rose 221 Similarly the operation can be performed on Snapshot objects There is a similar command for obtaining the position of the minimum MINPOS Figure 3 26 An example of using MAXPOS on 1 dimensional Series object 2 Figure 3 27 An example of using MAXPOS on 2 dimensional Series object 3 17 MAX getting the maximum of data This command obtains the maximum value of an object usage y max x x is the object and y is the maximum value of it In case of 1 dimensional Series object the example follows SATELLITE tom rose 231 a 3 7 5 1 6 2 4 SATELLITE tom rose 24 c max a CHAPTER 3 SYSTEM MODULE SYSTEM 48 SATELLITE tom rose 25 c 7 SATELLITE tom rose 26 In case of a 2 dimensional Series object the performance is similar SATELLITE tom rose 26 a SATELLITE tom rose 27 b SATELLITE tom rose 28 c 7 13 1 3 12 6 11 4 14 2 reform a 5 2 max b SATELLITE tom rose 291 c 14 SATELLITE tom rose 30 Similarly the operation can be performed on Snapshot objects There is a similar command for obtaining the minimum MIN 3 18 FIND finding the value close to the specified one in data This command obtains the nearest value to the specified one in an object usage ip find x val num x is an object ip is the returned value of the po
11. lt LHC 1 gt G 0 lt LHC 1 gt gt BI 0 lt 8SYBI 0 lt R 0 gt gt lt HNABI 0 lt LHC 0 gt gt FOR n 1 nc 2 R n lt GABA n lt LHC n gt gt LHC n lt GLU n lt R n gt gt lt G n 1 lt LHC n 1 gt G n BI n lt SYBI n lt R 0 gt gt lt NABI n lt LHC 0 gt gt NEXT n R nc 1 lt GABA nc 1 lt LHC nc 1 gt gt LHC nc 1 lt GLU ne 1 lt R ne 1 gt gt lt G nc 2 lt LHC nc 2 gt gt BI nc 1 lt SYBI nc 1 lt R nc 1 gt gt lt HABI nc 1 lt LHC ne 1 gt gt end Light Condition Horizontal Cell Layer Bipolar Cell Layer Figure 1 5 Simulation of a realistic neural etwork Example of NCS CHAPTER 1 WHAT IS SATELLITE Output Layer linear 31 4th Layer sigmoid 10 3rd Layer linear 3 2nd Layer sigmoid 10 E Input Layer Bias Uni eo aan A P linear 81 Figure 1 6 Simulation of artificial neural network Example of BPS Chapter 2 SATELLITE Shell and its functions 2 1 Introduction Signal processing techniques simulations using mathematical models etc are effective for analysis of the organizations and biological systems Various software systems such as Mathematica LabVIEW and AVS have been provided However if we use these software products the whole efficiency may fall remarkably because of data conversion to other systems SATELLITE enables to perform the consistent processing even if we use the completely differe
12. sentence is executed if the result of expr2 is true Relative operators shown in Table 7 3 can be used in expr2 expr3 is evaluated at the end of every iteration In the FOR sentence the parentheses are necessary Example Hodgkin Huxley s cell model type NETWORK module SQUID cell HH 11 gap G 10 connection HH 0 lt G O lt HH 1 for n 1 n lt 10 n HH n lt G n 1 lt HH n 1 G n lt HH n 1 HH 10 lt G 9 lt HH 9 end The description of networks must be at the beginning of the model description file 7 2 4 Example Hodgkin Huxley model In this section the programming by the NCS language is explained using a real model It is the network model connecting the Hodgkin Huxley H H models Table 7 1 as in Figure 7 4 The sodium current of the H H model is shown in the following equation a Jna m h V Ena 7 1 I an l m fmm 7 2 a E a bm 4exp 7 4 ap l h By h 7 5 El pe 0 07exp 5 7 6 E AA 7 7 exp 30 V 10 1 Gna Sodium membrane conductance 120 mS cm Ena Sodium reversal potential 115 mV CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 101 By using the NCS language Eq 7 3 and Eq 7 4 are described as follows am 0 1 25 V exp 25 V 10 1 bm 4 exp V 18 The value V 25 mV However a has to be rewritten as follows if V 25 am 0 1 25 V exp 25
13. teach data and initial values of connection weights from files It writes the history of connection weights and square errors during learning It is also possible to store the sum of errors at each iteration by giving a Series or Snapshot object as an argument to LEARN The real time buffer monitoring function BM can be used For example series x or snapshot x l1c bm x learn x lc stands for the number of iterations If the sum of errors does not have to be stored set learn 0 The connection weight history during learning is stored to the weight history file and the square error values to the error history file There are 2 kinds of storing modes for the weight history file Append mode in which the history is added in order of storing and Overwrite mode in which the history is overwritten The MLP Figure 6 3 is used to illustrate usage of LEARN MLP consists of 3 layers The number of units is 2 2 and 1 for the input hidden and output layers respectively The hidden layer and the output layer have the bias terms In Figure 6 3 the circles and the squares stand for units and bias units respectively The number on each weight corresponds to the order of storing in the file The weights are stored as shown in Figure 6 4 Weight 1 Weight 2 Weight 3 Weight 9 Figure 6 4 Format for storing connection weights For the interruption restart of learning the revised values of the connection weights 1 step
14. 1 3 5 7 JSATELLITE tom rose 71 Selecting 2 dimensional Series object at some intervals is shown in Figure 3 22 The commands are as follows JSATELLITE tom rose 71 a 1721 SATELLITE tom rose 721 b reform a 7 3 SATELLITE tom rose 731 b CHAPTER 3 SYSTEM MODULE SYSTEM 45 Da N SIN gt A Y N NV NN Figure 3 22 An example of using MABI on 2 dimensional Series object o 01 1 2 1 01 2 01 7 3 01 10 11 4 01 13 14 5 01 16 17 6 01 19 20 SATELLITE tom rose SATELLITE tom rose 7 3 gt 3 2 Lo 01 1 3 1 01 10 12 2 01 19 21 SATELLITE tom rose 3 6 9 12 15 18 21 741 c mabi b 3 2 751 c 2 761 Similarly the operation can be performed on Snapshot objects 3 15 GET getting a value at the specified position of data This function reads a value at the particular position of an object usage y get x position x is an object and y is the value at position A simple example is shown as follows see also Figure 3 23 SATELLITE tom rose 771 a SATELLITE tom rose 78 b 177 get a 3 SATELLITE tom rose 791 b CHAPTER 3 SYSTEM MODULE SYSTEM 46 Figure 3 24 An example of using GET on 2 dimensional Series object 4 SATELLITE tom rose 80 In case of 2 dimensional Series ob
15. 2 When the name of this file is set to be hhmodel s1 the execution of the simulation can be performed as follows SATELLITE tom rose 131 inline hhmodel s1 1 series V lin INa0 2 nassign hhmodel assign model file 3 nppl run the preprosessor 4 nlink 02 making simulation program 5 ntime 10 0 001 0 01 1 set simulation time 6 nstim HH 0 P 1 0 100 3 999 set external input variable 7 nout Iin HH 0 2 set output variable 8 nout V HH 0 1 9 nout INa0 HH 0 3 INa 10 ninteg R set integral method 11 ncal run simulation program List 2 Example of the simulation batch file CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 111 7 3 7 Display and analysis of simulation results The calculation results are stored in the buffers specified by the NOUT command The other system modules of SATELLITE can carry out the graphical representation and analysis For example the simulation results from hhmodel s1 are displayed as a graph using the system module GPM see Chapter 4 as follows SATELLITE tom rose 14 wopen 1 A4 0 0 SATELLITE tom rose 15 graph V T 0 0 0 0 0 SATELLITE tom rose 161 axis 1 1 XY XY 3 5 0 0 0 0 0 T T T T T T T T 1 0 0 200 0 400 0 600 0 800 0 1000 0 Figure 7 9 Example of a simulation result
16. 2 mNa0 0 05293 GNa jE 120 CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 109 GK 36 Gl 0 3 Module name G GL 5 Change of the delay We can change the delay by the NDELAY command The format of NDELAY is shown in the following Description ndelay mdl var dt init Arguments 1 mdl Module name 2 var Internal variable s name 3 dt Delay time 2 init Initial value of output The delay of the internal variable VOP of a module with the name G is changed by the following SATELLITE tom rose 65 ndelay G VOP 0 25 If the delay is added to the electrical chemical synapse module it is impossible to use the integration with adaptive calculation steps which is the default of integration in NCS Therefore we have to choose another integration algorithm by NINTEG command The delay condition can be displayed by setting D as the argument of the NSCLIST command as follows SATELLITE tom rose 66 nsclist D DELAY DELAY INFORMATION Data No Input Name Delay Time Initial Output 1 GC VOP 0 25 0 Selection of the numerical integration algorithm The algorithm with adaptive calculation steps is used for numerical integration NCS provides also Runge Kutta and Euler methods They can be used by the NINTEG command The format is shown in the following Description ninteg type mcell relerr Arguments 1 type Integration algorithm F With adaptive ca
17. 83 peteteteretatatetetetesetetetetaretatatetatetcteretereteretatatatetetetcteterererstetateteteseseteteteteretatetatetetetetcteteterstateteteteteteteteteterstatad SYSTEM MODULE SYSTEM 4 31 External Functions Install Ok SYSTEM MODULE EA 31 External Functions Install Ok SYSTEM MODULE 41 External Functions Install Ok SYSTEM MODULE PS2 25 External Functions Install Ok SYSTEM MODULE Gk 23 External Functions Install Ok SYSTEM MODULE EZEN 3 21 External Functions Install Ok Figure 2 1 X Window after starting SATELLITE ls 5 where stands for CR key The X Window after starting SATELLITE is shown in Figure 2 1 Right after starting the rc file usr local satellite lib satellite rc sl which is prepared by the system is automatically read at first and the setup file 7 setup sl which is set by each user is read the next Since these files are processed in the state of echo off messages of UNIX commands are not displayed on a terminal except the standard output errors To display the messages it is necessary to use the standard output errors or redirect the output as echo Welcome to SATELLITE WORLD gt dev tty Fundamentally we can write anything to the system rc file and the user setup file as long as it is syntactically correct However the starting will become slow if we call external executions frequently Generally we put the definitions of system modules in the system rc file
18. CIRCUIT SIMULATOR NCS 108 The conditions on the output can be displayed by setting 0 as the argument of the NSCLIST command JSATELLITE tom rose 61 nsclist 0 OUTPUT Variable Num index OUTPUT VARIABLE Tin 84 O EX INPUT OF HHC0 v 83 0 OUTPUT OF HHC0 INa0 87 O INa OF HH 0O Change of the parameter values The parameter values can be changed by the NPARA command The format of NPARA is shown in the following Description npara mdl var num Arguments 1 mdl Module name 2 var Parameter name 3 num Parameter value to set For execution of the NPARA command it is required to define the variable var of the module md1 in the description of the model file by the parameter sentence SATELLITE tom rose 62 npara HH Cm 1 2 The values that are changed by NPARA command are available until the simulation condition file is initialized by command such as NPP It is possible to display the present parameter values by the NLIST command SATELLITE tom rose 631 nlist HH Parameter List Module name HH Cm 1 2 mNa0 0 05293 GNa 120 GK 36 Gl 0 3 NLIST has the module name as its argument and the parameter values of the module are displayed When ALL is set as the argument the parameter values of all modules are shown as follows SATELLITE tom rose 64 nlist ALL Parameter List Module name HH Cm 1
19. HH 16 exinput lex 17 input lg 18 output A 19 observable INa IK 11 Ig 20 constant VNa 115 0 VK 12 0 Vl 10 6 21 parameter Cm 1 0 mNaO 0 05293 GNa 120 0 GK 36 0 22 Gl 0 3 hNa0 0 5961 nKO 0 3177 VO 0 23 function 24 if V 25 sodium current 25 am 0 1 25 V exp 25 V 10 1 26 else 27 am 0 1 10 28 bm 4 exp V 18 29 dmNa am 1 mNa bmx mNa 30 mNa integral mNa0 dmNa 31 ah 0 07 exp V 20 32 bh 1 exp 30 V 10 1 33 dhNa ah 1 hNa bh hNa 34 hNa integral hNa0 dhNa 35 INa GNa pow mNa 3 0 hNa V VNa 36 if V 10 potassium current 37 an 0 01 10 V exp 10 V 10 1 38 else 39 an 0 01 10 40 bn 0 125 exp V 80 41 dnK an 1 nK bn nK 42 nK integral nK0 dnK 43 IK GK pow nK 4 0 V VK 44 Il Gl V V1 leakage current 45 Iall Iex INa IK Il Ig 46 dV Iall Cm 47 V integral VO dV 48 end 49 x G module 50 type GAP 51 module G 52 input VOP 0 1 0 53 output lg 54 parameter GL 5 0 55 function 56 Ig GL VOP POSOUT 57 end List 1 Description of the Hodgkin Huxley model by the NCS Language 91 CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 92 e PULSE Description y pulse a b c d e Arguments 1 a Starting time for input 2 b Initial input value 3 C Pulse height 4 d Time width 5 e Time period e
20. IU SATTcETTef 0 z 0 0 0 0 JEJE sstel1ite 2 9 2701 wopen 1 Ad 0 0 e TE ST F op D Figure 1 3 SATELLITE interactive terminal CHAPTER 1 WHAT IS SATELLITE 4 GPM Graphic Package Module From the standpoint of data analysis visualization of data is much more important than numerical evaluation GPM module provides various graphic functions for making charts contour maps bird s eye pictures etc The images can also be printed 1 3 Platform Support SATELLITE runs on the following platforms Operating system from SunOS 4 1 2 from Solaris 2 5 from HP UX 9 05 from HP UX 10 01 from DEC OSF 1 V3 0 from Digital UNIX V3 2c from FreeBSD 2 1 0R from Linux 2 0 0 Window system from X Window Ver 11 R4 from OSF Motif Ver 1 1 Language to code C Language CHAPTER 1 WHAT IS SATELLITE 1 4 Examples ya a i RSX RAS PERE aR IEE Baws E me PEN A vaste eee Ee amy a is aag MIOS Figure 1 4 Biological signals during micro gravity Example of ISPP CHAPTER 1 10 0 50 into WHAT IS SATELLITE Photoreceptor a Anon terminal Bipolar Cell Inhibitory Synapse Excitatory Synapse Neural Circuit Simulator NETHORK rhb R 100 LHC 100 BI 100 GLU 100 GABA 100 S BI 100 NABI 100 G 100 parameter nc 100 connection R 0 lt GABA O lt LHC 0 gt gt LHC 0 lt GLU 0 lt R 0 gt gt lt G 0
21. data The String object is used for labeling e g in the case of drawing a chart outputting a message from a program etc The concatenation deletion repetition and separation are performed by sending operators and respectively A character sequence should be marked by double quotation marks For example concatenating a character sequence with another one is performed as follows SATELLITE tom rose 29 test dat test dat SATELLITE tom rose 30 Moreover we use to delete a character sequence SATELLITE tom rose 307 test dat dat test SATELLITE tom rose 311 For repetition of a character sequence is used SATELLITE tom rose 31 ABC 4 ABCABCABCABC SATELLITE tom rose 32 In order to separate a character sequence is used SATELLITE tom rose 32 A BC D EFG H IJK 0 A BC D EFG H 5 IJK SATELLITE tom rose 33 where 0 and 5 represent the index of data for displaying two or more elements CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 20 2 4 2 Class definition Class definition of variables in SATELLITE does not restrain the types permanently but generates the objects whose contents are flexible Definition of Series objects with no index specifies the 1 dimensional time series SATELLITE tom rose 35 series x The Series object 64 x
22. following see also Figure 3 10 CHAPTER 3 SYSTEM MODULE SYSTEM T 0 0 2 0 4 0 6 0 8 0 10 0 Figure 3 7 A graphic using the sampling frequency 1000Hz default T T T T T T T 1 0 0 200 0 400 0 600 0 800 0 1000 0 Figure 3 8 A graphic using the sampling frequency 10Hz 37 CHAPTER 3 SYSTEM MODULE SYSTEM 38 Figure 3 10 An example of using CUT on 2 dimensional Series object SATELLITE tom rose 851 a 1714 SATELLITE tom rose 86 b reform a 7 2 c c SATELLITE tom rose 87 cut b 3 0 5 0 SATELLITE tom rose 88 o 071 7 1 01 9 2 01 11 SATELLITE tom rose 891 Similarly selection can be performed on Snapshot objects 3 8 PUT replacing old data with new one This command replaces a part of an object with another one usage z put x y index x is the original object y is the object to insert into x z is the object after replacement and index is the position where to put y The size of the object z is the same as that of x For example as shown in Figure 3 11 replacement of a part of a 1 dimensional Series object a by b can be performed as follows SATELLITE tom rose 891 a 177 SATELLITE tom rose 901 b 11717 SATELLITE tom rose 91 c put a b 3 SATELLITE tom rose 92 c 0 1 2 3 11 12 51 13 14 SATELLITE tom rose 93 The next example i
23. generating the uniform random numbers CHAPTER 4 INTERACTIVE SIGNAL PROCESSING PACKAGE ISPP 53 20 20 0 200 400 600 800 1000 Time msec Figure 4 1 Waveform of the synthetic signal 4 Synthesis of signals The signals obtained by the above procedures are synthesized as follows 5 1 data The Series object that stores the synthetic signal is set 5 2 data acut bcut nois Signals are synthesized Mixture of two sine waves and normal random numbers is stored in the Series object data The waveform of the synthetic signal is shown in Figure 4 1 Preprocessing of data The methods of DC removal and window processing are shown below 1 Removal of DC The DCCUT command is used for removing the DC of data 1 1 datal The Series object that stores the data after removing DC is set 1 2 datal dccut data The original object is set By the above procedure the data with removed DC is stored in the Series object data1 The signal waveform is shown in Figure 4 2 2 Window processing We use the WINDOW command for the window processing 2 1 data2 The Series object that stores the data after the window processing 2 2 data2 window datal CHAPTER 4 INTERACTIVE SIGNAL PROCESSING PACKAGE ISPP 54 20 20 0 200 400 600 800 1000 Time msec Figure 4 2 Signal waveform after the removal of DC 20 20 0 200 400 600 800 1000 Time msec Figure 4 3 Signal waveform aft
24. loops They perform the statement stmt1 repeatedly until the condition expr1 is true In case of WHILE sequence the evaluation of expr1 is performed before the execution of stmt1 including its effects On the other hand the statement in case of DO WHILE is processed after execution of stmt1 WHILE sequence while expri stmt1 DO WHILE sequence do stmt1 while expri Processing of While x is smaller than n add x to s is described by the WHILE sequence as follows CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 25 SATELLITE tom rose 89 while x lt n s 8s x gt n Hb SATELLITE tom rose 907 The same example by the DO WHILE sequence is as follows SATELLITE tom rose 907 do s 8s x gt n while x lt n SATELLITE tom rose 91 2 7 3 FOR sequence In FOR sequence the first expression expr1 is evaluated only once that is during the initialization of a loop FOR sequence is terminated if expr2 is false which is evaluated before each iteration Expression expr3 is used for the re initialization of a loop after repetition FOR sequence for expri expr2 expr3 stmt1 For example processing of Add x to s n times is described by the FOR sequence as follows SATELLITE tom rose 91 for i 1 i lt n i s 8s x gt FE SATELLITE tom rose 921 BRAKE forces termination of a loop CONT
25. object class as argument in character format and converts it to the class The return value is the read object In case of the objects that consist of two or more elements like Series the elements are separated by commas For example CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 30 SATELLITE tom rose 77 y read series 1 2 3 4 5 6 7 8 The numerical values are stored in y as follows SATELLITE tom rose 78 y 0 1 2 3 4 5 51 6 7 8 ISATELLITE 1 tom rose 791 2 8 5 Data stream handling The redirection of data displayed on terminal to variables or UNIX commands can be performed by the data stream handling in SATELLITE It is similar to a pipe in UNIX The function UNIX is used for interfacing UNIX with SATELLITE It hands over a UNIX command to the shell In SATELLITE the input data is converted into the String object Then it can be substituted to a variable Data from the standard output can also be handed over to an UNIX command using the lt lt operator The example of the collective operation for all files listed by the ls command of UNIX in a current directory is shown as follows Function UNIX is used JSATELLITE tom rose 791 files unix ls dat SATELLITE tom rose 807 files 01 fnamal dat fname2 dat fname3 dat 3 fname4 dat fname5 dat SATELLITE tom rose 81 for i 0 i lt length files i Operation of files i SATELLITE tom ros
26. rose 611 b rotate a 3 SATELLITE tom rose 621 b 0 4 5 6 7 1 51 2 3 ISATELLITE tom rose 631 The following example is for 2 dimensional Series object as shown in Figure 3 20 SATELLITE tom rose 631 a 1714 SATELLITE tom rose 64 b reform a 7 2 SATELLITE tom rose 65 b Lo 01 1 2 1 01 3 4 2 01 5 6 3 01 7 8 4 01 9 10 5 01 11 12 6 01 13 14 SATELLITE tom rose 66 c rotate b 3 0 SATELLITE tom rose 67 c Lo 01 7 8 1 01 9 10 2 01 11 12 CHAPTER 3 SYSTEM MODULE SYSTEM 44 Figure 3 20 An example of using ROTATE on 2 dimensional Series object 3 01 13 14 4 01 1 2 5 01 3 4 6 01 5 6 SATELLITE tom rose 68 Similarly the operation can be performed on Snapshot objects 3 14 MABI selecting the subsequence of data This command selects from an object a subsequence of data specified by interval usage y mabi x step x is the original object y is the resulting object and step is the interval at interval of step 1 points There is no difference between x and y in the case where step 0 or 1 The following example and Figure 3 21 shows MABI function on 1 dimensional Series object SATELLITE tom rose 681 a 177 SATELLITE tom rose 69 b mabi a 2 SATELLITE tom rose 70 b 7 gt 4 0
27. the cell module used in a model is defined by this The module name and the number of components are required In order to set two or more cell modules we use commas to link them Example Definition of the cell module with 11 components with name HH cell HH 11 4 synapse In the network description the chemical synapse module used in a model is defined by this The module name and the number of components are required In order to set two or more cell modules we use commas to link them Example Definition of the chemical synapse module with 5 components with name SYN synapse SYN 5 5 gap In the network description the electrical synapse module used in a model is defined by this The module name and the number of components are required In order to set two or more cell modules we use commas to link them Example Definition of the electric synapse module with 10 components with name G gap G 10 6 exinput It defines the name of an external input variable in a cell module The number of external input variables is just one Example CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 94 7 10 Definition of Iex as an external input variable exinput lex input It defines the name of an input variable in a module The number and description order of input variables in the cell module should correspond to the descriptions in the connection sentences The number of input
28. tom rose 291 x 3 4 SATELLITE tom rose 307 The next example shows the operation on a multi dimensional object The object class is defined as follows see 2 4 2 for further details SATELLITE tom rose 301 series y 2 2 A value of y 0 1 is assigned e g SATELLITE tom rose 311 y 0 1 x To display the value of y 0 1 type as follows see also Figure 2 7 SATELLITE tom rose 321 y 0 1 o 1 2 3 CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 17 Figure 2 8 Data organization in 2 dimensional Series object Example 2 5 6 7 SATELLITE tom rose 33 To obtain the spatial data of certain time type see also Figure 2 8 SATELLITE tom rose 331 y 3 0 11 0 4 0 0 0 0 SATELLITE tom rose 341 Snapshot object Snapshot is the object class similar to matrix Figure 2 3 It is for dealing with static data sets and used as a subset of a Series object or a matrix Only on Snapshot objects with the same size can be performed operations For the mixed operation with a Scalar object the same operation is repeatedly performed between each element of the Snapshot and the Scalar Some examples of operations on Snapshot objects are shown below Figure 2 9 First an object class is defined as follows see 2 4 2 for further details SATELLITE tom rose 34 snapshot z 2 2 SATELLITE tom rose 35
29. 1 z 0 17 0 0 o 0 0 0 SATELLITE tom rose 36 A value is assigned to the item of this object as follows SATELLITE tom rose 371 z 0 1 4 SATELLITE tom rose 381 z gt Figure 2 9 Data organization in 2 dimensional Snapshot object CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 18 D Figure 2 10 A Scalar object o 1 0 4 o 0 0 0 SATELLITE tom rose 391 We can get the value of certain item as follows SATELLITE tom rose 391 z 0 1 4 SATELLITE tom rose 407 Scalar object Scalar is the object class for numbers such as variables to control sequences elements in Series objects etc Figure 2 10 It is expressed as the double precision number SATELLITE tom rose 41 k 0 8 SATELLITE tom rose 421 k 0 8 SATELLITE tom rose 431 File object This class is used for saving data in a file on a hard disk Data can be loaded from files and stored to files Therefore we can deal with it as with other objects without taking care of the format or the data type Moreover mixed operations with the Series object are also possible The File object has almost the same structure as Series and can store two or more sets of multi dimensional data Series Snapshot in the direction of the Record see Figure 2 11 Record corresponds to the time of the Series object and has flexible length
30. 3 FOR sequence e 95 2 8 Functions and procedures ececonarrss so 95 2 8 1 The scope of variables and constants and arguments in functions and procedures 25 289 Internal functions a Oe Ge e baa ee aR e daa Aoa 26 Fre Teer deined hinction erea a E a NO ee eee a A E eee ee E Ske 27 2 84 Input and output lt lt eee 29 2 8 5 Data stream handling ooo oo ee eee eee 30 2 9 Programming 264 wee ee RN I ee eee te Pe ew 3 2 9 1 Online message coocoo rra ee ee eee 3 2 9 2 Loading a program from a file e reee 31 3 SYSTEM Module SYSTEM 33 3 1 HELP displaying a command manual e e e ee 33 3 2 HEADER displaying or modifying the header information of a data file 33 3 3 WAIT interrupting the execution of a program ttt ts 33 3 4 REFORM changing the size or index of data o s r rrr 34 3 5 BM data monitoring ooo ek e a ee es 35 3 6 SAM sampling frequency setting lt ee eee 35 3 7 CUT selecting a subset of data o erre 36 CONTENTS 3 8 PUT replacing old data with new one est 3 9 MERGE merging two data sets 06 6 66 e ee 3 10 FILL filling data with a specified value s o s c s o o ooro r 3 11 ZERO filling data with 0 lt lt o 4 1 3 12 REVERSE reversing the order of data tt Tie as a p e enan E aie duka 3 13 ROTATE rot
31. 34 Learning sad p E e E A SVE de e 6 35 MEP testing Leica dd A ew A A A Pe ee a 6 3 6 Tracing connection weights and errors 6 3 7 Internal representation analysis of MLP tt rro Neural Circuit Simulator NCS 7 1 Introduction ttt 7 1 1 Basic specifications 4 6 aaa a 7 1 2 Concept of modularization ttt tt te 72 NOS Language 24440249 vias al a to bee eee e ee led TO Reserved Words we tooo E PU ti IA Ya ite ele EAN ee 7 2 2 Library functions eee e aa 7 2 3 Description of modules o 66 a 7 2 4 Example Hodgkin Huxley model error Ted gt How tos NOSE A a E AA A to id ce ras 7 3 1 Preparation of a model file sir ORAR A A ee ee 7 3 2 Registration of a model file lt lt lt lt o ooooooococ o 7 3 3 Preparation of an execution and a simulation condition file 7 3 4 Setting simulation conditions e eoe da e LASATA ANE GE B e E ETa e aoa moh Weacition of simulation a LEE PE a ee E ee ee ee TRE West bath dle eso ack A AA Ba eee ee fa 7 3 7 Display and analysis of simulation results s tt es Chapter 1 What is SATELLITE 1 1 Concept of SATELLITE It is generally agreed that the biological system is one of the most complex and sophisticated mechanisms on earth However in this moment since there are few systematic theories for approaching such systems trial and error studies based on knowledge of phys
32. 64 is defined by as follows SATELLITE tom rose 36 series y 64 64 Definition of Snapshot objects is performed as follows SATELLITE tom rose 37 snapshot a 10 b 20 20 In the case of Snapshot we cannot omit the size Scalar objects are defined as follows SATELLITE tom rose 38 scalar i j K Scalar objects are not allowed to have the size that is each object deal with only one value Finally definition of String objects is performed as follows SATELLITE tom rose 39 string str mstr 10 It is possible for String objects to specify their size 2 4 3 Conversions between two or more object classes The object class type of variables in SATELLITE is determined at the time of substitution It is the same as the size of the class on the right side of the equality work The above mentioned definition method is used only for receiving values as arguments of a function assigning values to multi dimensional objects etc We do not need to define the class in the case where it is determined by the assignment as follows SATELLITE tom rose 50 a 1 We cannot use undefined variables for the arguments of functions or procedures For example FF TC in the ISPP module is one of such commands SATELLITE tom rose 51 fftc P x y u v where P is a flag for specifying the calculation method x and y are input series and u and v are output series of the FETC command In this case u and v sho
33. 8 SATELLITE tom rose 59 plusten num 18 SATELLITE tom rose 601 Moreover functions can be called recursively The function fac for obtaining x is described as follows SATELLITE tom rose 607 func fac x if x lt 0 return 1 else return x fac x 1 2 SATELLITE tom rose 61 The next example is the procedure that performs Substitute n for the argument x and n 1 for the argument y At first we have to define x and y before calling the procedure as mentioned in 2 4 3 SATELLITE tom rose 61 scalar x y SATELLITE tom rose 62 proc plusone n x y 2 ty n 1 pie a ll SATELLITE tom rose 63 CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 28 The following is an example of calling this procedure SATELLITE tom rose 62 plusone 14 x y SATELLITE tom rose 63 x 14 SATELLITE tom rose 64 y 15 SATELLITE tom rose 65 The variables used in a function and procedure are local ones They are effective only in the function or procedure unless EXTERNAL is used In order to use global variables it is required to define every time in the function or the procedure The following is the same operation as the above mentioned example except for using EXTERNAL definition of x and y SATELLITE tom rose 65 proc subplusone n external x y gt yes 1 El SATELLITE tom rose 66 1 Anoth
34. 81 weight data n pre correction n weight data n 1 weight data n 2 pre correction n 2 history n history n 1 history n 2 weight data n weight data n 1 weight data n 2 pre correction n 2 Figure 6 6 Weight history file format in Append mode Then the sum of them is given by the following equation M _ 2 B gt es i 1 The initial value of error is the difference between the teaching data and the output of MLP with the initial weight values Learning is terminated if the sum of errors is less than the set tolerance or the number of iterations reaches the maximum At that time the following message is displayed Learning is done 6 3 5 MLP testing MLP is tested by the REC command REC reads the test data and the connection weights from files and writes the output results to a file It is possible to show the activity of units by change of a square size or color in the display of the structure of MLP In that case it is required to open a graphic window by the WOPEN command Using SIZE ORIGIN and COLOR in the GPM module we can set the size position and color of the connection weights of MLP respectively Figure 6 10 is the picture while executing REC The MLP structure and the test parameters are displayed during the execution The message that a new file is created is displayed when designated test CHAPTER 6 BACK PROPAGATION SIMULATOR BPS 82 Weight data n Weight data n 2
35. A the range can be adjusted by setting the second and fourth arguments in SCALE to D before the GRAPH command twelfth line In the fifteenth line the command LTYPE changes the line type The dashed line y 0 is drawn by DRAW in the sixteenth line Example 3 Displaying time series One of the merits of SATELLITE for analysis of biological data is efficient time series manipulation Here the example of a Gaussian noise sequence time series x nrand 1000 1 0 1 Generate a Gaussian noise sequence wopen 1 A4 0 1 sam 10000 Set the sampling frequency size 80 80 origin 20 200 title 1 time msec value scale N A N A graph x T 0 0 0 0 0 Draw time series axis 1 1 XY XY 3 5 0 0 0 0 0 frame label I 20 70 5 0 0 example4 Display labels CHAPTER 5 GRAPHIC PACKAGE MODULE GPM 70 example4 value T T T T T 0 0 30 0 60 0 90 0 time msec Figure 5 3 A Gaussian noise sequence The result of display is shown in Figure 5 3 We generated 1000 Gaussian random values The SAM command in the third line sets the sampling frequency to 10000Hz in order to consider them as a time series with the range 0 1sec the default is 1000Hz The argument related to the X axis in the eighth line is T It means that the horizontal axis corresponds to time If it is set to D then X axis corresponds to data points In the eleventh line the
36. AL CIRCUIT SIMULATOR NCS 90 Table 7 1 The Hodgkin Huxley model Membrane potential V membrane potential mV dV Cm a I Ina Ixn Iz Cm membrane capacity uF cm I total membrane current uA cm Sodium current Ina Gna m h V Ena dm g 2 w 3 am 1 m Bm m Ina sodium current uA cm 0 1 25 V Am E ARA e ea exp 25 V 10 1 Bm 4exp E JNa Membrane conductance 120 mS cm dh Z h Bn h a ap A an 0 07exp Ena reversal potential 115 mV B E 1 h exp 30 V 10 1 Potassium current Ik Gk nv Ex Ik potassium current uA cm dn T 2 w an l n Bnn 9K membrane conductance 36 mS cm 01 10 an ee ACs as oo Ex reversal potential 12 mV exp 10 V 10 1 V n 0 125 B exp 30 Leak current L membrane conductance 0 3 mS cm Ir OL V EL EL reversal potential 10 6 mV Table 7 2 Mathematical library functions of C language that can be used in NCS Function Definition exp e pow 7 sin sin x cos cos x tan tan x CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 1 Hodgkin Huxley s cell model 2 type NETWORK 3 module SQUID 4 cell HH 11 5 gap G 10 6 connection 7 HH 0 lt G 0 lt HH 1 8 for n 1 n lt 9 n 9 HH n lt G n 1 lt HH n 1 G n lt HH n 1 10 11 HH 10 lt G 9 lt HH 9 12 end 13 HH module 14 type CELL 15 module
37. Each data stored in a record must have the same size Moreover since the number of dimensions and indexes of the File object depends on that of the object stored in the first place the object with the different number of dimensions and indexes is stored after conversion Storing is performed by assigining a data element to File object For example y a Series or Snapshot object is stored to the record 0 of data dat SATELLITE tom rose 261 data dat 0 y In case of loading data we just type the name of a File object in an editing line SATELLITE will automatically treat it as the Series object For example data in the record 0 of data dat is loaded to x SATELLITE tom rose 271 x data dat 0 All records of data dat can be loaded to y as follows SATELLITE tom rose 281 y data dat Both x and y are Series objects and their dimension and index numbers depend on data dat For example when 2 dimensional data is stored in a record x is 2 dimensional Series object and y is 3 dimensional one CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 19 File type Data type Header part Owner 256byte Comment Data part Record 0 Date Dimension Record 7 i iden Record n 1 Record n Figure 2 11 Data file structure String class File names are basic and important information for managing data They usually include the attributes or serial numbers of
38. GHT command sets the following The name of the initial weight file the initial values of connection weights are read from this file the name of the weight history file the histories of connection weights the interval for storing weights and the mode for storing weights there are two kinds of storing methods append A or overwrite 0 It is also possible to set the generated weight file In this case the final history written to the file is used for setting of initial values If it is not necessary to store the history the mode should be overwrite The details on weight history file format and the mode for storing are described later The ERROR command is for setting the name of the error history file storing the histories of the sum of square error and the interval the direction record direction R or data point direction D and the mode for storing error When the direction for storing error is set to the record direction the error of each output unit and the sum of them are stored However if the direction is set to data point only the sum of error is stored The details on the error history file are described later CHAPTER 6 BACK PROPAGATION SIMULATOR BPS 77 In the TEACH command the following are set Input data file name teach data file name the number of patterns the beginning and end points of the input and teach data When the numbers are set from the Oth to the Oth all patterns are used for learni
39. INUE returns a loop to its starting point 2 8 Functions and procedures 2 8 1 The scope of variables and constants and arguments in functions and procedures The variables in SATELLITE are effective only in the function or the procedure where they are defined that is it is not allowed to refer to those variables in another function or procedure In order to compare external variables in a function or a procedure we need to use the reserved word EXTERNAL Internal constants and the constants are defined by CONST They are available in functions or procedures after their definitions Although we can define constants in a function or a procedure locally they become effective after processing Since all arguments of the functions and procedures in SATELLITE are handed over as variables the results obtained by operations on arguments inside return to the root CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 26 2 8 2 Internal functions SATELLITE has defined some internal functions for mathematical calculations or system management The mathematical function library apply to all objects All internal functions have the same priority Mathematical and system functions are shown in the following list cf Command Reference Manual List of mathematical functions abs x x acos x cos a asin x sin x atan x tan g atan2 x y tan x y same as atan x y cos x cosg exp x e exp2 x 2 int x the integer part of
40. K 4 V VK Potassium Current Il Gl V V1 Leak Current Iall lex IK 11 Ig dV Iall Cm V integral 0 dV Membrane Potential end Description of chemical synapse type modules The chemical synapse type module is the module with one input and one output It can combines two or more cell type modules In the description there is no difference between the chemical synapse type and the electrical synapse type The sentences used in the chemical synapse type module and their order are shown in the following type sentence module sentence input sentence output sentence observable sentence constant sentence parameter sentence initial sentence SOOO SES ire COP IRD r function sentence Although it is possible to omit some sentences their order cannot be changed See the previous section for the details on each sentence The end word is required at the end of the module description Example synapse type module type SYNAPSE module GABA input PO 0 1 0 output Tr parameter FB 0 01 CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 99 function Tr PO FB PO end Description of electrical synapse type modules The electrical synapse type module is the module with one input and one output It can combines two or more cell type modules The sentences used in the electrical synapse type module and their order are shown in the following
41. LABEL command displays labels The specified coordinates are relative values from the origin defined by ORIGIN The FONT command can set the font type of characters displayed by LABEL 5 3 2 Displaying 2 dimensional objects Example 4 displaying 2 dimensional random values The following examples show a bird s eye view a contour map and a color map of 2 dimensional Series object wopen 1 A4 0 1 size 80 80 x nrand 128 1 0 1 Generate Gaussian noise sequence y reform x 16 8 Convert 1 D Series to 2 D origin 20 200 gsolm y 0 3 0 4 0 0 0 4 0 1 1 X 1 0 Draw bird s eye picture origin 20 100 cont y 5 X 1 0 Draw contour map origin 120 200 map y X 1 0 1 Draw color map Type 1 origin 120 100 map y X 0 0 1 Draw color map Type 2 The result of the above command sequence is shown in Figure 5 4 CHAPTER 5 GRAPHIC PACKAGE MODULE GPM 71 ao l OS 11 MITE 2y TEREE ESE ES LEE I e gt MMS Y O O O e S CERS O LER F 5 57 EE O LESI E EN NE Figure 5 4 Various kinds of displays for 2 dimensional Gaussian noise sequence The command converting a 1 dimensional Series object to 2 dimensional one is REFORM fourth line It changes 128 point object to 16x8 object In the sixth line the object is displayed as a bird s eye picture in which the hidden line elimination is done The contour map is drawn in the eighth line The tenth and tw
42. NDEX command For example if we define a series object as SATELLITE tom rose 58 a 1710 then the index of the object can be obtained by JSATELLITE tom rose 59 index a 10 SATELLITE tom rose 60 In case of multi dimensional data such as b 10 50 the information is displayed as follows SATELLITE tom rose 607 snapshot b 10 50 SATELLITE tom rose 611 index b 0 10 50 SATELLITE tom rose 62 7 2 5 Expressions and operators Expression relates not only simple arithmetical operations but also substitutions functions etc The results of the evaluation of expressions are displayed automatically except for substitutions Although the notation of operators of SATELLITE is different from its internal functions one they are internally treated equally The operator and the internal function appeared in an expression is sent to the linked object and the first argument object respectively as a message Therefore even if two operators or internal functions are the same their performance may be different and depending on the object class Operators include arithmetical operators relational operators logical operators increment and decrement CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 22 Table 2 2 Priority table for operators in order of the high priority O tl ell A right AS left de left left left gt gt lt lt left amp amp
43. RAMP Decription y ramp a b c Arguments 1 a Starting time for input 2 b Initial input value 3 c slope e INTEGRAL Description y integral A P Definition y t rae y 0 A t e SIGMOID Description y sigmoid x il Definition y 1 exp z Figure 7 6 Ramp e RCASB Description y rcasb V a b c d e f g a expid VW c d V e Definition oe eee apf V 0 g 7 2 3 Description of modules Each module is described using the sentences explained in the next subsection The combination of these sentences depends on types The details of sentences and different type description methods are shown in the following Sentence Each sentence should be started with the definition that specifies what is described and completed with a semicolon The contents are divided by colons as follows Definition contentA contentB contentZ 1 type It defines the type of the module There are 3 types CELL Cell type SYNAPSE Chemical synapse type GAP Electrical synapse type The combination of them type NETWORK Example CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 93 Definition of the network type type NETWORK Definition of the cell type type CELL 2 module It defines the module name Use letters of alphabet for the beginning of a module name Example Setting the module name HH module HH 3 cell In the network description
44. Sampling intervals 4 itv Storing intervals last is the time when the simulation ends cal is the time interval used for numerical calculation Since the error convergence speed and the execution time of the simulation depend on this value it should be set to the appropriate value str is the time interval for monitoring the calculation results itv is the time interval for storing the execution results to the buffer designated by the NOUT command The simulation time is dramatically prolonged when this value is too small The unit of each argument is msec For example to set the simulation time to 10 msec the interval of calculation to 0 001 msec the interval of sampling to 0 01 msec and the interval of storing to 1 msec the following is required CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 106 Figure 7 7 Pulse function Figure 7 8 Ramp function SATELLITE tom rose 53 ntime 10 0 001 0 01 1 The simulation time conditions can be displayed by setting T as the argument of the NSCLIST command SATELLITE tom rose 54 nsclist T TIMER Last Time 10 Calc Step 0 001 Store Step 0 01 BufferStep 1 External input conditions Conditions for the external input are set by the NSTIM command NSTIM has the format that requires the module name the component number and the input waveform The followings are prepared for the input waveform Pulse function ramp function optional functio
45. Series gt y Tine Series y EEN 1 1 2e 16 0 1 1 50 1 50 Figure 3 5 Window for monitoring y 0 Figure 3 6 Window for monitoring y 1 frequency is a sampling frequency For example define a Series object a as follows SATELLITE tom rose 771 a 1710 The chart of a by using WOPEN GRAPH and AXIS commands is shown in Figure 3 7 In this case the default sampling frequency is 1000Hz Figure 3 8 displays the chart of a after changing the sampling frequency as follows SATELLITE tom rose 78 sam 100 The sampling frequency set by SAM is referred in commands of ISPP module or GRAPH command of GPM module 3 7 CUT selecting a subset of data This command allows selection of specified subset of data contained in an object usage y cut x start end x is the original object y is the object picked out start is the starting point and end is the end point The following example selects a part of a 1 dimensional Series object a as shown in Figure 3 9 SATELLITE tom rose 791 a 177 SATELLITE tom rose 80 a 0 1 2 3 4 5 6 7 ISATELLITE tom rose 81 b cut a 3 5 SATELLITE tom rose 82 b o 4 5 6 SATELLITE tom rose 83 One cannot obtain the proper result if the start point is replaced with the end one as follows SATELLITE tom rose 81 b cut a 5 3 One can also select a part of the 2 dimensional Series object b by the
46. System Analysis Total Environment for Laboratory Language and InTeractive Execution Biological and Physiological Engineering Laboratory Department of Information and Computer Sciences Toyohashi University of Technology Toyohashi 441 8580 JAPAN USER S MANUAL Contents 1 What is SATELLITE 1 1 1 Concept of SATELLITE s 4464 i a oe ee eH a tea wae a nee bee iG 1 ES SATELLITE Modules 2 n 3 ae ee a ee ev A ae 2 1 3 Platform Support k a p co RR 4 1 4 Examples lt 5o oester aare eee 5 2 SATELLITE Shell and its functions 8 91 Introduction see cn Os Bale eee ee ne BO dk a a a ae 8 2 2 How to start SATELLITE e t ee 8 23 Operation ceo cis dd ee be ay ee 10 2 3 1 Prompts and a window title 6 66 6 e ee ee 10 232 Widiting 24 a aor wigs E we a ke a eS SS 11 23 3 Preprocessor 64 6445 ee aai ee Ala aTa a eee eee 14 2 3 4 Arithmetical operation lt lt oo ee 14 2 4 Data handling A ie A A AA A A as 14 2 4 1 Objects and classes o mr 14 AD Olas detona y do AT dd o Slit ata da 20 2 4 3 Conversions between two or more object classes tt ts 20 2 4 4 Type of object Hoke Se Rh AA Be ee ee ees 21 2 5 Expressions and operators a 21 26 Internal constanta e A eee eS PA re wae eta Ae ye 22 2 7 Control sequence a a a ee a ea 23 2 7 1 IF sequence Foe ye A e ee oe a eee ee AES 23 2 7 2 WHILE and DO WHILE sequences t tt tt a moe Eug e k A ai 24 2 7
47. V 10 1 Jelset am 0 1 10 The gating variable m of the sodium channel is expressed by a differential equation It can be described by the following dmNa am 1 mNa bm mNa mNa integral mNa0 dmNa Similarly h can be described as follows from Eq 7 5 Eq 7 6 and Eq 7 7 ah 0 07 exp V 20 bh 1 exp 30 V 10 1 dhNa ah 1 hNa bh hNa hNa integral hNa0 dhNa From Eq 7 1 the sodium current Iya is given by the following INa GNa pow mNa 3 0 hNa V VNa The potassium current is described by following equations Ik Ox n V Ex 7 8 d anll n 6 n 7 9 0 01 10 V os et SR 7 10 E exp 10 V 10 1 ey V Ba 0 125exp z5 7 11 gx Potassium membrane conductance 36 mS cm Eg Potassium reversal potential 121 mV From Eq 7 9 Eq 7 10 and Eq 7 11 the gating variable n of potassium channel is described in the same way as the sodium current if V 10 an 0 01 10 V exp 10 V 10 1 else an 0 01 10 F bn 0 125 exp V 80 dnK an 1 nK bn nK nK integral nKO dnK CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 102 From Eq 7 8 the potassium current Iw is given by the following IK GK pow nK 4 0 V VK The leak current is defined by the following equation I GL V E 7 12 gz Leak current membrane c
48. a complex number Calculation of poles from AR coefficients Calculation of the gain of a complex number Calculation of histogram and the Gaussian density function value from data Calculation of power spectrum and phase of data Window processing for data CHAPTER 4 INTERACTIVE SIGNAL PROCESSING PACKAGE ISPP 52 2 Selecting a part of data of 1024 point from the Oth point to the 1023rd point of a sinusoidal wave 3 1 acut beut Series objects which store picked data are set 3 2 acut bcut cut a cut b Original objects are set 3 3 acut cut a 0 bcut cut b 0 Each starting point is set 3 4 acut bcut cut a 0 1023 cut b 0 1023 Each ending point is set By the above procedure the selected data are stored in the Series objects acut and bcut 3 Generation of random numbers We use NRAND command to generate the normal random numbers 4 1 nois The Series object that stores generated random numbers is set 4 2 nois nrand 1024 The number of datum point to generate is set 4 3 nois nrand 1024 1 The initial value to generate random numbers is set This must be an odd number 4 4 nois nrand 1024 1 0 The mean value of random numbers is set 4 5 nois nrand 1024 1 0 1 The variance of the random number is set By the above procedure 1024 point standard normal random number data are stored in the Series object nois Furthermore the URAND command is used for
49. age md1 should be the end of the name of the model file Using the NE command an editor is executed for the model file which is registered by the NASSIGN command see 7 3 2 The default is vi editor It can be changed by setting the environmental variable EDITOR of UNIX 7 3 2 Registration of a model file Starting the preprocessor by the NPP command see 87 3 3 and execution file by the NLINK command are done for the model file registered in the work area of SATELLITE The model must be registered in order to carry out the simulation using the NCS The registration is made by the NASSIGN command or the NPP command see 7 3 3 The NASSIGN command takes the model file name as its argument and registers the model file with the given name In the model file name there is no need to add md1 SATELLITE tom rose 50 nassign hhmodel In the above example hhmodel md1 is registered as a model file CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 105 7 3 3 Preparation of an execution and a simulation condition file After the execution of the NPP command the preprocessor is started and the model file described in the NCS language is converted into the source file of C language including the simulation condition file group SATELLITE tom rose 511 nppO The NPP command can take the model file name as its argument even if it is not registered by NASSIGN There is no necessity to add mdl in this c
50. al part of the original object is set 3 3 fftc P data2 rei The imaginary part of the original object is set 3 4 fftc P data2 rei Rout The Series object that stores the real part of the data after Fourier transform is set 3 5 fftc P data2 rei Rout lout The Series object that stores the imaginary part of the data after Fourier transform is set By the above procedure the processed data is stored in the Series objects Rout Iout The signal waveform after Fourier transform is shown in Figure 4 4 CHAPTER 4 INTERACTIVE SIGNAL PROCESSING PACKAGE ISPP 56 10 20 10 i 0 20 40 60 80 100 Frequency Hz Figure 4 5 Power spectrum 2 Power spectra The POWER command is used for adding two squared objects Using this command we can obtain the power spectrum of the original time series from the real part and the imaginary part of the data after the Fourier transform 4 1 pw The Series object that stores the obtained power spectrum is set 4 2 pw power Rout The real part of the data after Fourier transform is set 4 3 pw power Rout Iout The imaginary part of the data after Fourier transform is set The power spectrum data is stored in the Series object pw and shown in Figure 4 5 3 Phase property The PHASE command is used for obtaining the phase of the original time series from the real part and the imaginary part of the data after the Fourier transform 5 1 phs The Series obje
51. ase hhmodel mdl is registered as a model file and the preprocessor is started by the following SATELLITE tom rose 511 npp hhmodel After executing NPP the execution file is made by the NLINK command for compiling the source file of C language and linking it with the library in NCS SATELLITE tom rose 52 nlink 02 The argument of NLINK 02 is the optimization level for compilation The cc command of UNIX is used for compiling and linking Refer to the manual of UNIX for further details 7 3 4 Setting simulation conditions Several conditions necessary for carrying out the simulation are required to set Some of them are already set in the model file at the point when the NPP command is executed The commands for the simulation conditions described in the following overwrite the simulation condi tion file which is generated by NPP That is NPP must be executed before the execution of simulation condition commands If the simulation condition commands are not executed before NPP the following error message is displayed sl Error lt NCS nout gt No 1 Improper Model File Name in near line lt n gt The conditions are initialized when NPP is executed again Simulation time conditions Simulation time conditions are set by the NTIME command The format of NTIME is as follows Description ntime last cal str itv Arguments 1 last Simulation time 2 cal Calculation intervals 3 str
52. ating data eee eee 3 14 MABI selecting the subsequence of data eet e 3 15 GET getting a value at the specified position of data est tet tees 3 16 MAXPOS getting the position of the maximum in data eee res 3 17 MAX getting the maximum of data e tee ee eee 3 18 FIND finding the value close to the specified one in data tts Interactive Signal Processing Package ISPP 4 1 The command system of ISPP lt lt ooo eee eee 45 Examples tonser eee esa ee ee DARK A eee wade bl tai transfor 8 SS RP Soh he od oe Banca VAL Eo amp 4 ADD Witenes xaos elite See Be oe WE a ee IAS E A 4 2 3 Matrix operation lt lt lt ooo ee e Graphic Package Module GPM 54 Introduction ta AS ine A hh te PA te ad 5 2 Drawing and Printing ee es 5 3 Examples 2S es Sh Oe a we oe Ea ae ee ee Se Rie eee Oe als 5 3 1 Displaying 1 dimensional objects o rorroross 5 3 2 Displaying 2 dimensional Objects Back Propagation Simulator BPS GT Introduchion Bci tt wal E AR E ARS AA E ee eS Ee 6 27 The filetypes used in BPS cdas ce ot AAA Ad a DA 63 Bss cumple a a A Aa A di a 6 3 1 Preparation of input teach and test data files etree 6 3 2 Setting learning parameters ee 1 6 3 3 Initialization of weights lt gt e e 0
53. ayed for all input parameters If a default parameter is acceptable we just press the CR key to move to the next parameter The syntax of every SATELLITE command is checked However preprocessor can compensate for simple mistakes Parameters can be edited freely since they are stored in editting buffers 2 9 2 Loading a program from a file To make the interpreter load a program from a file the function INLINE is used Its argument is the file name it treats the file as the standard input Each line of the program is processed one after another similarly as in the case of the input from a terminal It is also possible to process the program of another file from the one called In fact the INLINE is also processed in a syntactic mode and connects the standard input of the interpreter to the file The limitation is given by the number of files that can be opened If an error occurs while loading a file the subsequent lines are not executed The example of a program file with name testsum sl is shown as follows psum 0 for i 1 i lt 10 i sum sum i printf sum d n sum Using INLINE the interpreter reads the above file SATELLITE tom rose 14 inline testsum sl sum 55 SATELLITE tom rose 151 When required to read a file in another file use INLINE in the file as follows sum 0 for i 1 i lt 10 i sum sum i printf sum 7 d n sum inline testsum2 sl
54. backward char B interrupt C delete char D DEL end of file D listing up files D end of lines E forward char F gt backward delete char H BS newline J kill line K newline M down history N l up history P 1 tty start output Q tty stop output S keyword completion W filename completion TAB ESC ESC command completion TAB ESC ESC 2 3 2 Editing The micro line editor for deletion or insertion of characters offers comfortable environment for interactive programming from a terminal This editor has internal buffers for editing The contents in the buffers are usually consistent with the character sequences which a user inputs and displayed on the editing line back from the prompt Line editing SATELLITE has a GNU Emacs like micro line editor The editing line is always in insert mode and we can move the cursor position by Ctrl F gt Ctrl B Ctrl A and Ctrl E keys Moreover Ctrl D DEL Ctrl H BS and Ctrl K can perform deletion of characters For example when we input the character sequence shown as follows SATELLITE tom rose 63 n O the cursor is at the right hand side of 0 now By pressing Ctrl H 0 is eliminated and the cursor is moved left On the other hand the cursor is moved left by Ctrl B without eliminating 0 The cursor moves to the head of the sentence that is to the position of n by pressing Ctrl A The list of key bin
55. before are CHAPTER 6 BACK PROPAGATION SIMULATOR BPS 80 stored at the end of the connection weight data file as shown in Figure 6 5 In the initial weight file all of the values are 0 In Figure 6 5 m records from record n or m records from record n m 1 correspond to 1 data block The WINIT command creates 2 data blocks that consist of the initial connection weight values and the 0 values The LEARN command reads the last 2 data blocks in the initial weight file and carries out the processing In this way it is possible to perform the learning process by continuing from the point where the learning was interrupted or terminated Record n Connection weight data e e Record n m Record n m 1 pre correction amount Record n 2m 1 Figure 6 5 Format of a weight history file Append mode and Overwrite mode are explained next The Append mode is the mode where the weight values and previous revisions are stored in 1 data block after the last history as shown in Figure 6 6 By utilizing this storing method it is possible to put the latest data at the end of the file and leave the history from the start point to the end point of learning The Overwrite mode is the mode where the weight data are overwritten to the initial values as shown in Figure 6 7 This mode is useful in the cases where the scale of the MLP s structure is large and there are many weights or it is not required to observe the history of the connectio
56. buffers CHAPTER 6 BACK PROPAGATION SIMULATOR BPS 83 El I M LEARNING ITERATION 200 UARE S ERRO 9 998863e 01 FFERENCE 1 222071e 06 bin kano xpt000 MMENT TM LEARNING SCALE W AMP our 300 9 996896e 01 3 237101e 06 I M LEARNING ERAT ION UARE S ERRO FFERENCE MMENT nnn ITERATION QUARE S ERRO IFFERENCE OMMENT 400 9 988988e 01 1 743440e 05 I M LEARNING ITERATION SQUARE S ERRO DIFFERENCE COMMENT 500 9 852359e 01 6 543T1Te 04 I M LEARNING fou 1 ITERATION SQUARE S ERRO 600 1 44620Te 08 Figure 6 9 Execution of LEARN 6 3 7 Internal representation analysis of MLP Even if the ISPP commands are used we can carry out the analysis of the internal state of MLP The following commands have been implemented into BPS commands for it RVMAP SIGMOID ERRFUNC and COR rvmap The inverse projection operation is carried out for the connection weights and the results are stored to 2 dimensional objects They can be displayed by using GPM commands sigmoid This command tests MLP with the test parameters The total input value and activation value of the unit are displayed in the quadrature axis and the vertical line respectively Furthermore the activation values of some inputs can be plotted on the curve It is also possible to show the histogram of input values errfunc Two connection weights between optional units are chosen and the
57. cd home tom TeX RE home tom TeX RETINA1 home tom TeX RETINA2 JSATELLITE tom rose 91 cd home tom TeX In the special case the list of all files and subdirectories which are consistent with the character sequences including wild cards in the current directory can be displayed as follows SATELLITE tom rose 107 u where stands for a blank Calling UNIX commands When the token not registered as reserved word or variable name appears in the head of the sentence the system leaves the processing to the UNIX shell We can deal with UNIX commands in the same way as the UNIX shell When the variable with the same name as UNIX command is already registered we can avoid duplication by attaching the backslash 1 to the head of the commands CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 14 2 3 3 Preprocessor The character sequence edited by the simple line editor is handed over to the preprocessor It mainly performs 1 history substitution 2 alias substitution and 3 parameter passing to SATELLITE commands History substitution refers to the last history items Istr refers to the newest history item which starts with str In both cases the head of the sentence is recognized as a history item and the replacement can be performed without destroying the character sequences before and after it Alias substitution If aliases are already defined just the first token of the sentence is replaced That
58. ct that stores the obtained phase is set 5 2 phs phase Rout The real part of the data after Fourier transform is set 5 3 phs phase Rout lout The imaginary part of the data after Fourier transform is set 5 4 phs phase Rout Iout D The type of the output phase D degree 0 radian is set 5 5 phs phase Rout Iout D U U is set if we want to perform the phase rehydration but 0 if not By the above procedure the phase data is stored in the Series object phs CHAPTER 4 INTERACTIVE SIGNAL PROCESSING PACKAGE ISPP 57 Method for obtaining the power spectrum and the phase of the original data Using the SPCF command we can obtain both the power spectrum and the phase from an input time series The procedure is shown below 1 series pw phs The Series objects that store the power spectrum and the phase are defined 2 1 spcf data2 The original time series is set 2 2 spcf data2 pw The Series object that stores the obtained power spectrum is set 2 3 spcf data2 pw phs The Series object that stores the obtained phase is set The power spectrum and the phase data are stored in the Series objects pw and phs by the above 4 2 2 Filtering MA filter Using the moving average method the procedure for smoothing the source signal is shown as follows 1 coef 1 5 1 5 1 5 1 5 1 5 The coefficient vector of the filter is set 2 1 output The object that stores the smoothed signal is set
59. d CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 104 7 3 How to use NCS This system is implemented in SATELLITE Batch processing using the batch file or interactive processing are possible Refer to the SATELLITE language command reference manual for the details of commands The procedure for doing simulations using NCS is as follows 1 Preparation of a model file The file that describes a model using the NCS language should be made See 7 2 for further details of the description method by the NCS language 2 Registration of a model file The name of a model description file for simulation should be defined After the definition NCS carries out the processing 3 Preparation of an execution and a simulation condition file After starting the preprocessor and linking the registered model file the execution file and the simulation condition file group are created 4 Setting simulation conditions The followings are set for the simulation The simulation time the external input variables output variables etc 5 Execution of the simulation The simulation is executed 6 Display and analysis of the simulation results The results obtained by the simulation are graphically displayed By referring to the simulation using the model description shown in Listing 1 the detail of this procedure are described in the next subsections 7 3 1 Preparation of a model file We can describe a model file with the specifications of the NCS langu
60. d electric synapses gap junctions The elements constituting the neural circuit are electric synapse chemical synapse and neuron All neural circuits are made of the combination of these elements and each of them has the specific function in the neural circuit In the model description by NCS they are regarded as type Neural circuit is classified into 3 types cell type neuron synapse type chemical synapse and gap type electric synapse By using the notation of the set theory it can be described as follows Elements in a neural network cell type synapse type gap type Although the type is a cluster of all elements with the identical function it can be subdivided into modules In case of retina for example there are some cell type modules namely the cone photore ceptor CONE which receives the light and the horizontal cell HC which controls and modifies visual information Their characteristics are different each other Every element with different characteristics is divided and defined as a module in NCS The cell type is cell type HC module CONE module and similarly other types The expression for the HC module is as follows HC module HC 0 HC 1 The element in the module is called component It corresponds to the substance of the elements which are parts of the neural network The component is indicated by index which is attached to the module name as follows A modu
61. data files 6 2 In order to learn MLP input and teach data files must be made The record direction corresponds to the patterns and the data point direction to the input or output units as shown in Figure 6 1 The test result files generated by the REC command also take this form There is no limitation in the number of patterns and the number of units The number of input units and output units are 2 and 1 respectively The number of patterns is 4 in case of the XOR problem The type of data is given as Series type or Snapshot type Suppose that the names of objects for the input and teach data are in and out respectively The substitution of each object is as follows CHAPTER 6 BACK PROPAGATION SIMULATOR BPS 74 pattern 1 data 11 data 12 data 13 eee data In J 3 Q Pattenm2 data21 data 22 data 23 los data 2n Es Gn z pattern 3 data 31 data 32 ii 33 4 98 aata 3n c o e e pattern m Figure 6 1 Format of input data teach data test data and test result files Series objects Snapshot objects series in 2 out 1 snapshot in 4 2 out 4 in 0 0 0 in 0 0 0 in 0 1 0 in 1 0 1 in 1 0 0 in 1 1 1 in 2 1 0 in 2 0 1 in 21 1 0 in 3 1 1 in 3 0 1 in 3 1 1 out 0 0 out 0 0 out 1 1 out 1 1 out 2 1 out 2 1 out 3 0 out 3 0 Input and teac
62. data points is handed over to sig The result is the Series object with 21 elements We can easily make programs dealing with time series using mathematical formulas only In the above mentioned example the result is obtained just as we intended in cases where the argument is a Scalar Snapshot Series or File object If the argument t is a String object an error message is returned SATELLITE tom rose 751 sig test sl string not supported method ISATELLITE 1 tom rose 76 2 8 4 Input and output There are some external functions and commands for displaying objects Using PRINT we only have to arrange the objects to display separated by commas In SATELLITE the message is displayed on line according to specific format e g SATELLITE tom rose 701 x 3 SATELLITE tom rose 71 print x x An x 3 SATELLITE tom rose 72 print 1 2 3 4 5 An 0 1 2 3 4 5 SATELLITE tom rose 73 The function PRINTF is also available We can specify the precision of displayed elements Although the usage is similar to the printf function of C language it is internally different SATELLITE tom rose 731 x 3 SATELLITE tom rose 74 printf x d n x x 3 SATELLITE tom rose 75 printf 9 4f n 1 2 3 4 5 0 1 0000 2 0000 3 0000 4 0000 5 0000 JSATELLITE tom rose 76 The READ function reads an object from a terminal It receives an
63. ds is shown in Table 2 1 History The inputs from a terminal are recorded in the history buffers By pressing Ctrl P the history buffers are traced back and the history is copied to the editing buffers Ctrl N performs the history search in ascending order We can freely edit and execute commands from the history buffers For example CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 12 SATELLITE tom rose 63 n SATELLITE tom rose 64 j SATELLITE tom rose 65 o gt o e The following can be displayed by pressing Ctrl P SATELLITE tom rose 65 SATELLITE tom rose 641 j 0 Again the following can be displayed by pressing Ctrl P SATELLITE tom rose 64 j 0 SATELLITE tom rose 63 n O Moreover the following can be displayed by pressing Ctrl N o E 0 SATELLITE tom rose 631 n SATELLITE tom rose 641 j When a character sequence is already in the editing buffer only the history lines whose heads match the character sequence are called For example Oe o 2 SATELLITE tom rose 631 n SATELLITE tom rose 64 j JSATELLITE tom rose 65 n By pressing Ctrl P the following is displayed JSATELLITE tom rose 65 n SATELLITE tom rose 63 n 0 Completion of file names and commands If TAB key is pressed after inputting characters the help commands will be uniquely identified by the head of the edi
64. e 82 The following is the example in which the data generated in SATELLITE is stored into a text file SATELLITE tom rose 82 t 2 PI 071024 1024 SATELLITE tom rose 83 unix cat gt data txt lt lt sin t The next example is the reverse operation that is from a text file to an object SATELLITE tom rose 84 s unix cat data txt The String object s is converted to the Series object t by the following SATELLITE tom rose 851 t 0 s SATELLITE tom rose 86 t 0 0 000 0 006 0 012 0 018 0 024 5 0 030 0 036 0 042 0 049 0 055 Omitted 10151 0 05 0 04 0 04 0 03 0 03 1020 0 02 0 01 0 01 0 00 0 00 SATELLITE tom rose 87 CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 31 2 9 Programming 2 9 1 Online message One of the special features of SATELLITE is that 1t allows us to deal with parameters interactively while displaying their explanation The parameters are separated by comma Here the example is shown e g for function GRAPH SATELLITE tom rose 131 graph SATELLITE tom rose 131 graph x Desire Y AXIS DATA Object or TFD It is required to input the object of Y axis If we input volt for example the following is displayed SATELLITE tom rose 13 graph volt SATELLITE tom rose 131 graph volt T Seve X AXIS DATA Object or TFD Next the object of X axis follows Messages are displ
65. e of using REFORM 1 Figure 3 2 An example of using REFORM 2 3 4 REFORM changing the size or index of data Command for the modification of an object format usage y reform x index x stands for an input object y for output and index for the index of y As shown in Figure 3 1 for example we can change a 1 dimensional Series object a to a 2 dimensional Series object b by the following SATELLITE tom rose 65 a 1714 SATELLITE tom rose 66 a 01 f 2 3 4 5 5 6 7 8 9 10 10 11 12 13 14 JSATELLITE tom rose 67 b reform a 7 2 SATELLITE tom rose 68 b Lo 01 1 2 1 01 3 4 2 0 5 6 3 01 7 8 4 01 9 10 5 01 i 12 6 01 13 14 SATELLITE tom rose 69 Conversion of 2 dimensional Series object b to 3 dimensional Series object c shown in Figure 3 2 is performed as follows If the specified index size is bigger than the input object s one Os are filled in the tail of data SATELLITE tom rose 691 c reform b 3 2 4 SATELLITE tom rose 707 c o 0 01 1 2 3 4 0 11 01 5 6 7 8 1 o 01 9 10 11 12 1 1 01 13 14 0 0 2 0 01 0 0 0 0 2 1 01 0 0 0 0 SATELLITE tom rose 711 Similarly can be reformatted Snapshot objects CHAPTER 3 SYSTEM MODULE SYSTEM 35 x EE DE a 1 100 Figure 3 3 Buffer monitor F
66. e symbol table of the interpreter can also be completed by pressing Ctrl W For example if we want to complete the reserved word or the variable name that starts with i SATELLITE tom rose 891 i SATELLITE tom rose 89 if By pressing Ctrl W again the 2nd candidate is displayed as follows SATELLITE tom rose 89 if SATELLITE tom rose 89 inline Listing files File names can be listed by Ctrl D halfway This function is helpful for checking the file names while typing a program or using UNIX commands such as cd cp mv etc For example SATELLITE tom rose 8 cd home tom TeX As shown above we can get the subdirectory names under home tom Tex by pressing Ctrl D without interrupting the input of character sequences SATELLITE tom rose 8 cd home tom TeX RETINA1 RETINA2 work1 tex work2 tex JSATELLITE tom rose 8 cd home tom TeX Character is appended to the end of directory names to executable file names to symbolic links to sockets to FIFOs pipe with a name to character devices and ff to block devices respectively After displaying the list the command inputted halfway is redisplayed We can also obtain the list of the files that start with certain characters In the following example all of file and subdirectory names that start with RE will be displayed JSATELLITE tom rose 9
67. elfth lines are displaying color maps There are two sorts of displays for color maps The first normalized numerical values according to min max range correspond to the rectangle size The second normalized numerical values correspond to colors Chapter 6 Back Propagation Simulator BPS 6 1 Introduction BPS is one of the system modules of SATELLITE It consists of the functions and procedures for simu lating a multi layered perceptron model MLP It is possible to use the error back propagation method BP and its five accelerated modifications as the learning algorithms A little background about MLP and its learning algorithm is required for using BPS module Followings are the features of BPS e Using the SATELLITE interactive programming environment it is possible to define the structure of MLP easily Setting changing the parameters connection weights of MLP and execution of simulations can also be done easily e Using the INLINE command it is possible to batch process setting of parameters and the network structure learning testing the trace of internal weight representation etc e Using the buffer monitoring function BM it is possible to monitor the real time change of the error during learning It is also possible to display the simulation results easily by the GPM module in SATELLITE e Using the ISPP module of SATELLITE we can carry out the multilateral and detailed analysis of MLP 6 2 The file typ
68. er example follows SATELLITE tom rose 66 func glplusone gn external x y subplusone gn z x y return Z SATELLITE tom rose 67 scalar x y z SATELLITE tom rose 68 glplusone 4 9 SATELLITE tom rose 691 x 4 SATELLITE tom rose 701 y 5 SATELLITE tom rose 711 z 0 SATELLITE tom rose 721 Since functions never check their arguments classes the ones having multi defined operators and mathe matical functions are performed exactly regardless of the object class of arguments multi state functions except the class the operators cannot deal with The example of a sigmoidal function is shown At first it is defined as follows SATELLITE tom rose 72 func sig t return 1 1 exp t SATELLITE tom rose 73 CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 29 When the argument t is a Scalar object the return value of the function is also the Scalar object SATELLITE tom rose 731 sig 0 0 5 SATELLITE tom rose 741 In the case where t is a Series object we have SATELLITE tom rose 741 sig 10710 0 4 54e 05 0 0001234 0 0003354 0 0009111 0 002473 51 0 006693 0 01799 0 04743 0 1192 0 2689 107 0 5 0 7311 0 8808 0 9526 0 982 15 0 9933 0 9975 0 9991 0 9997 0 9999 207 1 SATELLITE tom rose 75 uu Thus the series from 10 to 10 obtained by the operator 21
69. er the window processing The original object is set 2 3 data2 window datal 1 The type of the window 1 Humming window 2 Hanning window 3 Blackman window 4 Triangle window is set 2 4 data2 window data1 1 0 1 is set if we want to correct data so that both integrated values of data before and after the window processing become equal but 0 if not By the above procedure the data after the window processing is performed is stored in the Series object data2 The signal waveform is shown in Figure 4 3 Fourier transform 1 Fourier transform Using the FFTC command it is possible to carry out Fourier transform and inverse Fourier transform for complex number data Since the FFT algorithm is used for the Fourier transform the number of data must be the power of 2 CHAPTER 4 INTERACTIVE SIGNAL PROCESSING PACKAGE ISPP 55 0 1000 1000 0 20 40 60 80 100 Frequency Hz Figure 4 4 The signal waveform after Fourier transform black line the real parts of data gray line the imaginary parts of data 1 series Rout lout The Series object that stores the real part and the imaginary part of an output time series is defined 2 rei 0 1023x0 In the case where the imaginary part of the original signal does not exist 1024 zeros are stored in the Series object rei 3 1 fftc P The flag for calculation P Fourier transform I Inverse Fourier transform is set 3 2 fftc P data2 The re
70. es used in BPS In BPS the exchange of data during learning or testing of MLP is carried out through files There are seven file types as shown in Table 6 1 Although all file formats are in conformity to SATELLITE ones each type is different see below for further details There is another type of files the parameter file ASCIT file which is for preserving the parameters of network structure learning conditions the management of data etc 6 3 BPS use example In the following the use of BPS is explained on the concrete examples of the MLP simulation The example is XOR Exclusive OR problem There are two input variables and one output variable Table 72 CHAPTER 6 BACK PROPAGATION SIMULATOR BPS 73 Table 6 1 File types for the BPS module File ype Input data file Input data for learning Teaching data file Teaching data for learning it ae lgo winit Initial weight File Initial weight values Lene a weight weight setrec The weight values learn wgtload during learning errfunc rvmap sigmoid error learn Error history file The error d lea Test data file Input data for testing Test result file Output results setrec ures E for testing actload x Using the file for input xx Using the file for output Weight history file Table 6 2 The XOR problem Inputs Output 0 0 0 0 1 1 1 0 1 1 1 0 6 3 1 Preparation of input teach and test
71. ext we calculate the filter coefficients from zr zi pr and pi using the IRCOEF command 4 series a b The Series objects that store the coefficients of the denominator and numerator of the transfer function are defined 5 1 iircoef zr zi pi pr The zero points and poles of the transfer function are set 5 2 iircoef zr zi pr pi a The Series object that stores the coefficients of the denominator of the transfer function is set 5 3 iircoef zr zi pr pi a b The Series object that stores the coefficients of the numerator of the transfer function is set By the above procedure the obtained filter coefficients can be used in both FIR and IIR Finally filtering is carried out as follows 6 temp fir b input The numerator of the transfer function is calculated 7 output iir a temp gain The denominator of the transfer function is calculated and the result of the filtering is obtained by gain multiplication The signal after filtering is stored in the Series object output CHAPTER 4 INTERACTIVE SIGNAL PROCESSING PACKAGE ISPP 61 Input Series Time LE Output Series a 0 Time Start End Figure 4 7 Filtering by the IIR command a1 2 5 are the filter coefficients Example The design of the 13th order IIR type low pass filter with the cut off frequency 100Hz and with the Butterworth characteristics is provided The procedure is shown below 1 sam 512 The sampli
72. h data are stored in files as follows in dat in out dat out In this example in dat and out dat are input data and teach data files respectively The test data file is required when performing the test of MLP If one wants to observe the MLP s output on the same input data used during learning the input data file can be used as a test data file 6 3 2 Setting learning parameters Some parameters must be set before the learning and testing of MLP are executed There are four types of parameters e MLP s structure parameters e Parameters for generating the initial values of connection weights e Learning parameters e Testing parameters CHAPTER 6 BACK PROPAGATION SIMULATOR BPS 79 It is possible to store the parameters using the BPSAVE command to a parameter file The parame ters can be read from the parameter file by the BPLOAD command Since the parameter file is the ASCII type another simulation under the different conditions can be carried out easily by changing the parameters using an editor The content of the parameter file is shown in the following Contents of the learning parameter file number of layers x number of cells in each layer x status of activation functions and bias units x weight initialization algorithm x initial weight file name stored by WINIT x seed of random number generator for weight initialization maximum for initial weights minimum for initial weight
73. h a specified element of the data Data interpolation interp akima spline Interpolation of 2 dimensional data Interpolation of 1 dimensional data by the Akima s method Interpolation of 1 dimensional data by the natural cubic spline Arithmetic operation average integ det eigen inv mul trans nmeq Calculation of the arithmetic mean of data Calculation of the sum of data Calculation of the determinant Calculation of eigenvalues and eigenvectors Calculation of the inverse matrix Calculation of the product of two matrices Calculation of the transposed matrix Solving the normal equation Table 4 2 ISPP commands 2 Data analysis bpbtw burg cep fftc fftn fir firmake hil hpbtw icep iir iircoef levin lpbtw phase pole power rank spcf window Design of IIR type band pass filter with the Butterworth property Calculation of power spectra by the Burg method Calculation of complex cepstrum Complex Fourier transform Complex Fourier transform for multi dimensional data Filtering by FIR type filter Design of FIR type filter Hilbert transform Design of IIR type high pass filter with the Butterworth property Inverse cepstrum analysis Filtering by I R type filter Calculation of the coefficients of IIR type filter from zero points and poles Calculation of power spectra by the Levinson Durbin s algorithm Design of IIR type low pass filter with Butterworth property Calculation of the phase of
74. he 12th line As mentioned in 7 1 2 three types of module descriptions are provided electric synapse chemical synapse and cell In the listing lines from the 14th to the 52nd line correspond to the cell type module and lines from the 54th to the 62nd line describe the electric synapse type module Details of reserved words NCS library functions sentences and special descriptions used in the NCS language are explained next 7 2 1 Reserved words The following 5 words are defined as the reserved words by the system 1 TIME The simulation time It can be used for all modules CN The component number It can be used for all modules PRECN The cell module that is the input value to the chemical electrical synapse is called presynaptic cell module of the synapse PRECN holds its component number It can be used for chemical and electrical synapse modules POSTCN The cell module that is the output value from chemical electrical synapse is called postsynaptic cell module of the synapse POSTCN holds its component number It can be used for chemical and electrical synapse modules POSOUT The output value from the postsynaptic cell module It can be used for chemical and electrical synapse modules 7 2 2 Library functions The following 5 kinds of functions have been implemented Mathematical library functions of C language can be also used as shown in Table 7 2 CHAPTER 7 NEUR
75. he color instead of number Mix capital letters with small letters for describing colors is not allowed Example 2 displaying two sinusoidal curves with different amplitude and frequency wopen 1 A4 0 1 origin 40 40 Set the origin of the coordinate axes size 80 80 Set the size of the chart title 1 time f t Set the labels of X axis and Y axis t 07100 100 yi sin PI 5 t y2 0 5 sin 2 PI 5 t scale N F N F 0 0 1 0 1 2 1 2 lwidth 1 2 Set the width of the lines graph y1 t 0 0 0 0 0 lwidth 2 2 graph y2 t 0 0 0 0 0 axis 1 1 XY XY 4 0 0 0 0 0 lwidth 1 2 ltype 1 2 Set the dashed line type draw Y 0 Draw a line such that Y 0 ltype 1 1 frame CHAPTER 5 GRAPHIC PACKAGE MODULE GPM 69 f t time Figure 5 2 Two sinusoidal curves with different amplitude and frequency The result is shown in Figure 5 2 The origin of the coordinate axes second line the size of the chart third line the labels of X and Y axis fourth line and the width of lines ninth eleventh and fourteenth lines were set The argument values of ORIGIN command should be the absolute coordinate values from the bottom left corner of a window They can be displayed by moving a mouse cursor in the window In order to draw two curves in one chart the range of drawing should be fixed by setting the second and fourth arguments in SCALE to F If they are set to
76. ial equations and the solutions are obtained by the numerical calculations Simulation programs are coded by general purposive programming languages The Neural Circuit Simulator NCS has been developed as a software system supporting research In NCS characteristics or connecting states of cells are described using the NCS language which is the exclusive model description language It is possible to efficiently carry out the simulations under various conditions without rewriting the model description Followings are the features of NCS e It is possible to perform the large scale neural circuit model simulation based on a physiological knowledge e It is possible to handle not only the model with faithful physiological evidence but also the general continuous system model e Programming except the mathematical model construction is unnecessary e Simulations are possible without recompiling if the conditions are changed 7 1 1 Basic specifications The NCS system structure is shown in Figure 7 1 The NCS consists of three components NCS prepro cessor NCS library and a command group for setting conditions NCS preprocessor converts a neural circuit model described in the NCS language into a simulation program and a simulation condition file group coded in C language The simulation program compiled by the C compiler becomes a simulation execution file by linking the NCS library The NCS library is aggregate of the basic programs for exec
77. ibute is the internal variable its name must be set as the argument The format of NOUT is shown in the following Description nout buff mdl com type val Arguments 1 buff Buffer name for storing output values 2 mdl Module name 3 com Component number 4 type Attribute of output 1 Output value 2 Input value 3 Internal variable s value 5 val Variable name if type 3 Output value is the value of the variable designated in the output sentence in the description of the module with the name md1 The value of the variable in the exinput sentence is input value For example to set the input value of the Oth component of the module with the name HH as an output to the buffer Iin is done as follows SATELLITE tom rose 58 nout Iin HH 0 2 To set the output value of the Oth component of the module HH as an output to the buffer V is carried out as follows SATELLITE tom rose 59 nout V HH 0 1 To set the value of the internal variable INa of the Oth component of the module HH as an output to the buffer INa0 is done by the following SATELLITE tom rose 607 nout INa0 HH 0 3 INa The buffer which stores output values must be defined as Series type before the execution of the NOUT command For example the buffers Iin V INa0 should be defined as follows JSATELLITE tom rose 57 series lin V INa0 CHAPTER 7 NEURAL
78. igure 3 4 Window for setting up a range to draw 3 5 BM data monitoring This command displays a window for monitoring objects simultaneously while processing other com mands usage bm x x stands for an object to monitor Example SATELLITE tom rose 721 bm x The example of the buffer monitor is in Figure 3 3 In this example the object x has already been defined by the following SATELLITE tom rose 701 t 0799 SATELLITE tom rose 71 x sin 2 PI t 100 Another window can be opened by clicking on the SCALE button One can adjust scaling of a chart Figure 3 4 shows the window An example of 2 dimensional Series objects is given First we convert 1 dimensional Series object x to 2 dimensional Series object y by REFORM as follows SATELLITE tom rose 731 y reform x 2 50 If we want to monitor y we can proceed similarly as in the previous example SATELLITE tom rose 74 bm y The buffer monitor window of this example is shown in Figure 3 5 By clicking on the button P the chart is changed as shown in Figure 3 6 Figure 3 5 is the chart of y 0 and Figure 3 6 of y 1 That is Figure 3 5 corresponds to the chart from x 0 to x 49 and Figure 3 6 from x 50 to x 99 3 6 SAM sampling frequency setting This command defines a sampling frequency usage sam frequency CHAPTER 3 SYSTEM MODULE SYSTEM 36 Pune
79. ilt in functions Procedures operators data Protocol External Functions ISPP Digital signal SL UTIL NCS lonic current model Utilities simulation BPS NPE Neural network Nonlinear paramete Figure 1 2 A modular scheme of SATELLITE system 1 2 SATELLITE Modules SATELLITE organizes analysis techniques for various systems by grouping its functions into modules according to the purpose or method There are several modules containing basic tools for system analysis such as digital signal processing numerical simulation model parameter estimation etc as listed below Details are described in the subsequent chapters SYSTEM module is a gathering of basic functions for handling data It includes the functions such as picking up data finding a maximum or minimum of a sequence modifying data format displaying header information of data files etc ISPP Interactive Signal Processing Package is a module for data analysis based on signal processing and statistical theories They are extremely important for modeling and extracting the char acteristics from experimental data Built in commands can be applied to not only the time series but also the multi dimensional data see also Figure 1 4 NCS Neural Circuit Simulator isa neural modeling and simulation environment In this system special description language is utilized to describe the neuronal properties and the network structure Thi
80. ilter is shown below 1 sam 1024 The sampling frequency is set 2 series zr zi pr pi The Series objects for storing the zero points and poles of the transfer function are defined 3 1 gain The Scalar object for storing the gain of the designed filter is set 3 2 gain 1pbtw 100 The cut off frequency is set 3 3 gain lpbtw 100 13 The order of the filter is set This value must be odd number the maximum is 101 3 4 gain lpbtw 100 13 zr zi The Series objects for storing the zero points real part and imaginary part of the transfer function are set 3 5 gain lpbtw 100 13 zr zi pr pi CHAPTER 4 INTERACTIVE SIGNAL PROCESSING PACKAGE ISPP 60 The Series objects for storing the poles real part and imaginary part of the transfer function are set By the above procedure it is possible to obtain the zero points zr and zi the poles pr and pi and the gain gain of the transfer function of the 13th order IIR type low pass filter with the cut off frequency 100Hz with the Butterworth characteristics The high pass filter can be set as follows gain hpbtw 400 13 zr zi pr pi In this example the 13th order IIR type high pass filter with the cut off frequency 400Hz is designed In case of band pass type two cut off frequencies must be set When the cut off frequencies are 100Hz and 400Hz and the order of the filter is 13 for example the following is set gain bpbtw 100 400 13 zr zi pr pi N
81. inear FUNCTION has to be executed after the number of layers is set by LAYER In the XOR problem the number of output units is 1 and the number of input units is 2 The hidden layer is composed of 2 units with sigmoidal activation functions The activation function of the output unit is linear and each unit in the hidden and output layers has the threshold Then setting parameters for the MLP is carried out as follows see also the Command Reference Manual layer 3 2 2 1 function LN SA LA Weight initialization parameters The initial weight file which stores the initial values of connection weights is generated by the WINIT command WINIT requires several parameters They specify the algorithm for generating initial values the name of the initial weight file details are described later the seed for generating random numbers and the maximum and minimum of initial values There are two methods for generating initial weight value using random numbers generated from a given seed R and the Jia s algorithm J When the Jia s algorithm is used the bias unit must be added in each layer These parameters are set using the WALGO command In the XOR problem for example the parameters for the initial values of connection weights are set by the following walgo R initwf 1 1 0 1 0 winit Learning parameters The WEIGHT ERROR TEACH LALGO LEND and DISP commands are used to set the parameters for learning The WEI
82. iology psychology etc has to continue Environment to support and realize the ideas of scientists could be so important to advance the research We assert that the establishment of basic platform for data analysis and model simulation could be relevant for analyzing the complex systems such as neural systems The basic concept of system analysis forms the cycle data analysis modeling computer simulation evaluation and experimental testing as shown in Figure 1 1 SATELLITE System Analysis Total Environment for Laboratory Language and InTeractive Execution has been developed considering this scheme ISPP NCS NPE BPS Signal Processing Modeling Parameter Estimation measurement Simulation Evaluation GPM Figure 1 1 A general flow chart of biological system analysis SATELLITE consists of the SATELLITE shell which provides interactive and C like language process ing system and several modules which together cover more than 200 commands and signal processing numerical simulation etc See also Figure 1 2 The most important facility of SATELLITE shell is an interactive operating environment User can execute command sequence from the text file batch pro cessing in case of the complex and large scale simulations see also Figure 1 3 One can also visualize data and print it CHAPTER 1 WHAT IS SATELLITE 2 Program Analysis Algorithm Extensity Objects Functions bu
83. ire to store data or data files One can analyze data multilaterally using the signal processing or statistical techniques 4 1 The command system of ISPP The commands of ISPP are classified into the categories shown in Table 4 1 and 4 2 By combining the commands with fundamental functions it is possible to carry out complicated analysis 4 2 Examples to use The fundamental use of ISPP is described by referring to Fourier transform filtering and matrix oper ation 4 2 1 Fourier transform Fourier transform of a signal is shown Signal is synthesized two sinusoidal waves overlapped by noise Generation of data 1 Generation of a sinusoidal wave 1 t 071999 2000 Data is stored in the Series object t 2 a 5 sin 2 PI 20 t 3 b 3 sin 2 PI 50 t 3 The sine waves with DC in which their frequencies and amplitude values are different from each other are stored in the Series objects a and b 50 CHAPTER 4 INTERACTIVE SIGNAL PROCESSING PACKAGE ISPP 51 Table 4 1 ISPP commands 1 Data generation Data generation by AR model Generation of 2 dimensional Gaussian distribution function Generation of random data with optional probability distribution Generation of multi dimensional Gaussian random data Generation of Gaussian random data Generation of uniform random data Data operation DC removal from data Data normalization Shifting the whole data so that the specified value is consistent wit
84. is similar to the C shell of UNIX Parameter passing One of the strongest points of SATELLITE is that several parameters required in each function can be passed interactively The details are described in 2 9 1 2 3 4 Arithmetical operation To perform arithmetical operations using SATELLITE we can input them directly For example doing multiplication 3 x 6 SATELLITE tom rose 13 3 6 18 SATELLITE tom rose 141 Similarly dividing as follows SATELLITE tom rose 14 3 6 0 5 SATELLITE tom rose 151 2 4 Data handling 2 4 1 Objects and classes Data obtained from biological systems or numerical simulations is usually a multi dimensional series We rarely pay attention to one value but rather deal with a set SATELLITE deal with such a time series as a single data class object and provides a data structure namely Series object which can treat the differences between the temporal changes and the spatial changes of the multi dimensional data efficiently see also Figure 2 3 There are 4 other kinds of object classes than the Series class Snapshot String Scalar and File classes These classes are divided with respect to values they deal with numerical values and character CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 15 Time i eg X 2 dimensional Snapshot 2 dimensional Series
85. ject as shown in Figure 3 24 we have JSATELLITE tom rose 801 a 1714 SATELLITE 1 tom rose 811 b reform a 7 2 SATELLITE tom rose 821 c get b 3 0 SATELLITE tom rose 831 c 7 SATELLITE tom rose 84 Similarly the operation can be performed on Snapshot objects 3 16 MAXPOS getting the position of the maximum in data This command obtains the position of the maximum in a object usage y maxpos x num x is an object y is the Series object and num is the number of positions to consider For example as shown in Figure 3 25 if we need to obtain the position of the maximum in the 1 dimensional Series object a then SATELLITE tom rose 84 a 3 7 5 1 6 2 4 SATELLITE tom rose 85 c maxpos a 1 SATELLITE tom rose 86 c 1 SATELLITE tom rose 87 To get the positions of the 1st and 2nd maxima in a as shown in Figure 3 26 is done by the following SATELLITE tom rose 87 c maxpos a 2 SATELLITE tom rose 88 c LO 01 1 CHAPTER 3 SYSTEM MODULE SYSTEM 47 1 01 4 SATELLITE tom rose 891 As shown in Figure 3 27 to get the position of the maximum in 2 dimensional Series object b proceed as follows SATELLITE tom rose 181 a 7 13 1 3 12 6 11 4 14 2 SATELLITE tom rose 191 b reform a 5 2 SATELLITE tom rose 20 c maxpos b 1 SATELLITE tom rose 21
86. lculate the sinusoidal function scale N A N A Set the range of drawing graph y t 0 0 0 0 0 Draw the chart frame Draw frame axis 1 1 XY XY 3 5 0 0 0 0 0 Draw the coordinate axes All the beginning it is required to open a drawing window The first line command WOPEN does it The last argument should be set to 1 if we want to print the picture otherwise 0 is the default value CHAPTER 5 GRAPHIC PACKAGE MODULE GPM 68 Table 5 2 Colors for drawing magenta magenta SSA o The type and range of the coordinate axes is defined in the fourth line The axis type can be N linear or L logarithmic We choose N for both X and Y axis The argument A in SCALE means that the range is set automatically This is default if we do not use the SCALE command In order to specify the range we need to set the argument to F and set the minimum and maximum values for X and or Y axis as the fifth and sixth arguments If we omit those SATELLITE presses us to set see 2 9 1 The chart is drawn by the commands in the fifth line Sixth and Seventh lines display the frame and the coordinate axes In this example the color of the chart is white by default To specify the color use COLOR command before GRAPH command There are 8 possible colors to display for both charts and frames as shown in Table 5 2 Although the numbers 0 to 7 are usually used for specification of colors one can write the name of t
87. lculation steps R Runge Kutta method E Euler method 2 mcell Maximum number of cells 3 relerr Relative accuracy of integration If the delay condition is added to the electrical chemical synapse module it is impossible to use the integration with adaptive calculation steps which is the default of integration in NCS Therefore we have to choose another integration algorithm in this case that is Runge Kutta or Euler SATELLITE tom rose 67 ninteg R CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 110 7 3 5 Execution of simulation The simulation is executed by the NCAL command after the simulation conditions are set SATELLITE tom rose 11 ncal The following message is displayed when NCAL is executed NCS Ver 6 8 3 SIMULATION PROGRAM on Sun gt gt NOW CALCULATING WAIT FOR A TIME PLEASE lt lt 0 0 done 0 0 stands for the percentage of the completed calculation If it reaches 100 the following message is displayed NCS Ver 6 8 3 SIMULATION PROGRAM on Sun gt gt THE CALCULATION HAS FINISHED 1 lt lt 100 0 done SATELLITE tom rose 121 The simulation finished and the shell is waiting for another command 7 3 6 Use of batch file The above procedure can be described in a batch file We can execute it by using the INLINE command inline batch_filename The example of the batch file for the simulation using the model file hhmodel md1 is shown in List
88. le name and a suffix specify one element 7 2 NCS Language Language called the NCS language is used to describe a model User can construct various models by describing the model for each module and changing its description if needed Moreover some special descriptions are used in order to deal with large scale models The NCS language has been composed of reserved words NCS library functions sentences and special descriptions For example the neural circuit model Figure 7 4 in which the Hodgkin Huxley H H model Table 7 1 is connected in series by resistances can be described as shown in Listing 1 From the 2nd to the 12th line A network description It is certainly necessary for model files The connected form in Figure 7 4 is defined From the 14th to the 47th line The description of the H H model The membrane potential V is the input of the gap current Jj The simultaneous differential equa tions in Table 7 1 are described in the function sentence From the 49th to the 56th line The description of electric synapses The current outputs in proportion to the membrane potential CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 89 G 0 G 1 G 9 Figure 7 4 Example of a neural circuit model The descriptions by the NCS language can be divided roughly into the module description using differ ential equations from the 14th to the 62nd line and the network description which shows the connected modules from the 2nd to t
89. le type when the NCS program is converted into C language through the NCS preprocessor Therefore it would happen that the desirable values are not substituted if we used an old compiler of C language CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 95 11 12 parameter It defines parameter variables The parameters defined here are registered in a simulation condition file and can be changed using the NPARA command Use the double notation for parameters even if they are integers Example Definitions of Cm mNa0 GNa GK GL as parameters with values 1 0 0 05293 120 36 0 3 respectively parameter Cm 1 0 mNa0 0 05293 GNa 120 0 GK 36 0 Gl 0 3 connection It defines the relationship of between two or more components It is called a relative expression lt in the relative expression stands for the composition in which the the output of the right hand side is input of the left hand side Each input is in parentheses For example mdl i lt inputi input2 input1 and input2 are substituted for the first and the second inputs of the i th component of the module md1 As mentioned before the number of input variables in the input sentence should be equal to the relative expression s one That is two input variables have to be defined for the description of mdl as follows input Vi V2 In this case the outputs from input1 and input2 are substituted for V1 and V2 respec
90. left I left right Notice In the above table right means that the operator is combined with the right hand side object and left means that the operator is combined with the left hand side object 19 operators the substitution operator the operator for generating a sequence with margin 1 the operator for connecting two or more series etc Followings are the examples of usage SATELLITE tom rose 561 x 37 1 SATELLITE tom rose 57 x 0 3 2 1 SATELLITE tom rose 581 y 173 SATELLITE tom rose 591 y 0 1 2 3 ISATELLITE tom rose 607 z x 0 y SATELLITE tom rose 611 z 0 3 2 al 51 2 3 SATELLITE tom rose 6217 Operators are interpreted by following the priority shown in Table 2 5 The following example demon strates for comparison operators SATELLITE tom rose 631 z gt 0 z 0 0 0 0 0 1 5 2 3 SATELLITE tom rose 64 Objects are destroyed after performing operations If we want to keep the results of operations we have to assign them to variables The variable mentioned here can be regarded as a simple container for objects without restricting the type of data Therefore even if object names are the same there is some possibility that their contents become different after substitution Memory management of objects is done by garbage collecting me
91. make 1 11 The order of the filter is set The value is the odd number used by the FIR command 2 4 coef firmake 1 11 100 The cut off frequency is set 2 5 coef firmake 1 11 100 3 CHAPTER 4 INTERACTIVE SIGNAL PROCESSING PACKAGE ISPP 59 The type of the window function is set 0 Rectangle window 1 Hanning window 2 Humming window 3 Blackman window 4 Kayser window By the above procedure the coefficients of the 11th order FIR type low pass filter with the cut off frequency 100Hz is stored in the Series object coef Similarly can be obtained high pass type filter coef firmake 2 11 400 3 In this example the 11th order FIR type high pass filter with the cut off frequency 400Hz is designed Band pass filter with the cut off frequencies 100Hz and 400Hz and the order 11 for example is obtained as follows coef firmake 3 11 100 400 3 Then filtering can be performed by the FIR command with the coefficients coef as shown below 3 output fir coef input IIR filter In order to design an IIR filter of the low pass high pass or band pass type with the Butterworth characteristics it is first necessary to obtain zero points poles and gain of the transfer function using the LPBTW HPBTW or BPBTW command respectively Then filtering is carried out by using FIR and IIR after the zero points and poles are converted into the filter coefficients by the IRCOEF command The example of a low pass f
92. mp Figure 4 9 Impulse response obtained by FIR command filtering has from 2 x filter order 1 points In this example since 2 x 13 1 25 it is possible to obtain the accurate filtering result by defining delay as a value larger than 25 e g 50 9 gain 1pbtw 100 13 zr zi pr pi 10 iircoef zr zi pr pi a b The filter is designed by obtaining the transfer function 11 firtemp fir b d_impulse 12 firinp cut firtemp delayt 1 datptdelay 1 By calculating the numerator part of the transfer function and removing Os in d_impulse the impulse response is shifted delay points 13 output iir a firinp gain The part of the denominator of the transfer function is calculated and the impulse response of the designed filter multiplied the gain is obtained Figure 4 10 CHAPTER 4 INTERACTIVE SIGNAL PROCESSING PACKAGE ISPP 63 foe pe 3 2 4 F ol is 0 20 40 60 80 100 120 140 160 Data Point Figure 4 10 Impulse response of the designed filter mn o0 So E o E A e o T T T T T T T T T T 0 50 100 150 200 250 Frequency Hz Figure 4 11 Amplitude chart of the designed filter 14 spcf output u v By executing Fourier transform of the impulse response output and calculating power spectrum we obtain the amplitude chart of the designed filter is obtained Figure 4 11 We can confirms the Butterworth characteristics can be confirmed 4 2 3 Matrix operation Matrix o
93. n from a file and a function from a buffer The format is shown in the following Description nstim mdul com type pil p2 p3 p4 p5 Arguments 1 mdul Module name 2 com Component number 3 type Function type for input P Pulse function R Ramp function F Input from a file B Input from a buffer 4 pl p5 parameters depending on type type pl p2 p3 p4 p5 P Start time Initial value Height Width Period R Start time Initial value Steepness F File name Buffer no B Buffer name The module with the name specified by this command must have the description of the external input using the exinput sentence Setting to input pulse function of the Oth component of the module with the name HH is carried out by the following CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 107 SATELLITE tom rose 55 nstim HH 0 P 1 0 100 3 999 The conditions on the external input can be displayed by setting S as the argument of the NSCLIST command SATELLITE tom rose 56 nsclist S EXINPUT EXTERNAL INPUTS Data No Component Function Data No 1 HH 0 1 No 1 Function start_tm init_out height width period lt PULSE gt 11 o 100 3 999 Output conditions The output conditions are set by the NOUT command NOUT has the buffer name which stores output values the module name the component number and the attributes of output as the arguments If the output attr
94. n object with 0 CHAPTER 3 SYSTEM MODULE SYSTEM 41 Figure 3 15 An example of using FILL on 2 dimensional Series object usage y zero x start end x is the original object y is the filled object and start and end are the start and end points The range in x specified by start and end is filled with 0 The following example is similar to the FILL s one as shown in Figure 3 16 except the specified value 20 is replaced with 0 SATELLITE tom rose 421 a 177 SATELLITE tom rose 43 b zero a 3 5 SATELLITE tom rose 44 b 0 1 2 3 0 0 5 0 7 ISATELLITE tom rose 451 In the case of 2 dimensional Series object as shown in Figure 3 17 the following example is quite similar to previous one ISATELLITE tom rose 451 a 1714 SATELLITE tom rose 461 b reform a 7 2 SATELLITE tom rose 471 c zero b 3 0 5 0 SATELLITE tom rose 48 c 0 01 1 2 1 01 3 4 2 01 5 6 3 01 0 8 4 01 0 10 5 01 0 12 6 01 13 14 SATELLITE tom rose 491 Similarly the operation can be performed on Snapshot objects 3 12 REVERSE reversing the order of data This command reverses the order of data in an object CHAPTER 3 SYSTEM MODULE SYSTEM 42 Figure 3 17 An example of using ZERO on 2 dimensional Series object usage y reverse x x is the original object and y is the rever
95. n weights There are 2 kinds of methods for storing the error history record direction and data point direction Moreover Append mode and Overwrite mode are provided Record direction storing is the method for storing error of each unit in the output layer and totals as shown in 6 8 The data length of 1 block is equal to number of output units 1 Figure 6 9 shows the execution of LEARN The MLP s structure and the learning parameters are displayed The message that new files are created is displayed if the specified weight history file and error history file do not exist When either of them exists the notifying message is displayed and it is storing mode confirmed Append or Overwrite In the case the mode is Overwrite the mode is reconfirmed by pressing y If n is pressed the command is terminated and the file name must be reset In the case the mode is Append no message is displayed but the mode is set by pressing y The input n means that the mode is changed to Overwrite with the message on display When the learning is in progress the number of iterations the sum of square errors the difference from previous error value and comments are displayed Assume that the number of output units is N Then the squared error e for each pattern is as follows N ei E 05 j 1 where t is teaching data and 0 is output data of MLP Assume that the number of all patterns is M CHAPTER 6 BACK PROPAGATION SIMULATOR BPS
96. nection Input the output from the 1st component of the cell module HH to the Oth component of HH through the Oth component of the synapse module G HH O lt G O lt HH 1i Input the sum of the output from the 4th component of the cell module HH through the 4th component of the synapse module G and the output from the 6th component of HH through the 5th component of G to the 5th component of HH HH 5 lt G 4 lt HH 4 G 5 lt HH 6 2 inputs to the 10th component of the cell module HH The 1st input is the output from the 9th component of HH through the 9th component of the synapse module G The 2nd input is the output from the Oth component of CA through the Oth component of the synapse module SYN HH 10 lt G 9 lt HH 9 SYN O lt CA O 13 function Characteristics of the module are described by using mathematical expressions The values of parameters should be preset before they are used The mathematical expressions are divided by semicolons IF sentences can be used in function sentences if condition sentencel else sentence gt If the expression condition is true then sentence is executed Otherwise sentence is performed Relative operators for conditional expressions are shown in Table 7 3 Examples Descriptions of INa and IK function INa GNa V VWa CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 97
97. ng The LALGO command sets the mode for learning on line learning P or batch learning S the learning algorithm six methods steepest descent method conjugate gradient method etc the learning rate and the necessary parameters for each algorithm In the LEND command the maximum number of iterations and the tolerance are set as the termi nation conditions for learning In the DISP command the followings are set The interval to display the number of iterations and the square error value and a comment sentence The example of setting the learning parameters for the XOR problem is shown weight initwf wgt 200 A error err 200 D teach in out 0 0 lalgo S 6 0 005 0 6 0 0003 0 75 0 6 lend 0 0 5000 disp 200 I M LEARNING Testing parameters The parameters for testing MLP are the weight file name the weight values of MLP after learning the weight history number to use the test data file name the pattern numbers to use the input layer and output layer numbers and the test result file name If the pattern numbers are set from the Oth to the Oth all patterns are used for testing These parameters are set by the SETREC command setrec wgt 0 in 0 0 0 2 res 6 3 3 Initialization of weights An initial weight file is necessary in order to carry out learning However there is no necessity of executing the WINIT command when the history stored in the weight history file generated d
98. ng frequency is set 2 series zr zi pr pi 3 series a b 4 series u v The Series objects used in LPBTW ITRCOEF and SPCF are defined 5 delay 50 6 datp 511 7 impulse 1 1 datp 0 8 d_impulse 0 delay 0 impulse The impulse signal and the signal that contains zeros for delay are generated Since the output points within the filter order can not be calculated by the FIR command the signal d_impulse is created as the union of delay Os and impulse Figure 4 8 a as shown in Figure 4 8 b The impulse response is obtained by shifting the delay points Figure 4 9 b after filtering of d_impulse is performed Figure 4 9 a FIR command uses the future input in order to obtain the present output as it was shown in Figure 4 6 Since the causality is not satisfied the IIRCOEF command outputs the coefficients by joining filter order 1 Os to the coefficients of the numerator of the transfer function Therefore the data for CHAPTER 4 INTERACTIVE SIGNAL PROCESSING PACKAGE ISPP 62 2 2 o oO q 3 E an 3 4 a EO lt lt 2 e o Y T T T T T 6 T T T T T 0 50 100 150 200 0 50 100 150 200 Data Point Data Point a inpulse b inpulse Figure 4 8 Impulse signal o o S S 5J 5J 2 o oO 3 E EE a S a S eo Boy lt lt 2 2 o T T T T T o T T T T T T 0 50 100 150 200 0 50 100 150 200 Data Point Data Point a firtemp b firi
99. nsional data is lined to the direction of a time axis Figure 2 3 The Series object that includes a single value is the same as 1 dimensional array in general purpose languages Figure 2 6 The operation between two or more Series classes is possible only when each size of a data set Snapshot is the same That is we cannot deal with a 2 dimensional Series object and a 1 dimensional Series object together Moreover when the length of the direction of the time axis is different the operation is performed within the limits of the shorter one and the remainder is copied as it is A mixed operation between a Series object and a Scalar object can be performed e g the multiplication of the Scalar object and each element in the Series objects The characteristics of Series objects are the implicit calculations repeated to each element and the operation functions for time series data by the operators and such as selection filling etc Here some examples of operations on Series objects are shown below Data from 1 to 7 are stored in a 1 dimensional Series object by the following command see also Figure 2 6 SATELLITE tom rose 271 x 177 2 Here is the operator for generating an arithmetical series with a margin 1 see 2 5 for further details SATELLITE tom rose 281 x o 1 2 3 4 5 51 6 7 SATELLITE tom rose 291 We can check the 3rd element as follows SATELLITE
100. nt appli cation software systems It places simulators and signal processing packages as its external functions and organizes them along with API Application Program Interface specification The merit of SATELLITE is that several different data sets such as multi dimensional time series matrices and so on can be processed to make analyzing the biological system easier The processing system of SATELLITE is an interpreter Programs are translated into the intermediate stack code The stack machine code is executed by stack machines Therefore the repetition procedures or functions such as for command and while command are performed at slightly higher speed The internal composition is shown as follows Simple line editor Preprocessor Lexical and syntax analysis Stack machine code Stack machine execution If the program is syntactically correct the processing system will translate it into the stack machine code and execute it Frequently one may want to use an editor check a file name change a current directory use UNIX commands etc If the token that appears at the beginning of a sentence is not defined and is not substitution SATELLITE passes such commands to Bourne shell of UNIX 2 2 How to start SATELLITE SATELLITE is started by typing sl shown as follows CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 9 Welcome to The SATELLITE World Copyright C 1992 BPEL Toyohashi University of Technology Version 2
101. onductance 0 3 mS cm E Leak current reversal potential 10 6 mV From the above equation the description of the leak current is as follows Il Gl V V1 The current which crosses a membrane I is given by the following equation dV ma I Iva Ix Iz 7 13 dt From Figure 7 4 the total current J is the sum of currents from the neighboring cells through the electrical synapse J and the injection current Iez PS Len ie apy 7 14 Eq 7 13 can be transformed as follows dV I D a E wl dt Cm ey From the above the description in the NCS language is given by the following Iall lex INa IK Il lg dV Iall Cm V integral VO dV The module with the above characteristics is described next Assume that the type is the cell type and the name is HH type CELL module HH The external input is the injection current lex the input is the current from the adjoining cell J and the output is the membrane potential V exinput lex input lg output V The following are observed Sodium current Iya potassium current Ix leak current Iz and the current from the adjoining cell Ig observable INa IK 11 Ig The reversal potential of each ion current Eya Ex and Ez is defined as a constant constant VNa 115 0 VK 12 0 V1 10 6 CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 103 The followings are defined as parameters membrane capacitance Cm the initial val
102. p2 sl 6699 Not providing option means that reading from the standard input terminal is not performed and the system closes right after termination However if the files are read except the rc file the setup file and the clean file the system state is echo on Then the command messages are displayed 2 3 Operation 2 3 1 Prompts and a window title The interpreter shows prompt As shown below the prompt of SATELLITE displays the current directory name and the line number Only last two parts of a current directory name are displayed because of the length of the prompt When a path name is not complete appears before the path name For example SATELLITE home tom 62 cd work SATELLITE tom work 63 Moreover when a program exceeds 1 line we are urged by the prompt For example if we input SATELLITE tom work 63 n 0 SATELLITE tom work 64 for i 0 i lt 10 i the following is displayed In such case process is completed by inputting the following n ent i e In the case of X Window terminal emulator Xterm Kterm DECterm etc the host name and the complete path name of the directory are displayed at the window title see Figure 2 2 That helps us compensate the imperfect information displayed at the prompt CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 11 Table 2 1 Key binds of the line editor for SATELLITE beginning of line A
103. perations are one of the features of ISPP Here as a practical example we obtain the solution of a system of n linear equations 21101 Zin n Yi System of equations can be written in the matrix form as follows Til Tjin 91 y 4 1 X0 Y 4 2 The solution of Eq 4 1 is as follows under the assumption that the matrix X is regular 0 X Y 4 3 CHAPTER 4 INTERACTIVE SIGNAL PROCESSING PACKAGE ISPP 64 The procedure to calculate Eq 4 3 by ISPP is shown in the following 1 11 10 0 laa SN RR amp 2 1 3 16 03 In the ISPP Module there are several matrix operation commands such as MUL command to get the product of two matrices INV to calculate the inverse matrix etc 1 tempx 1 1 1 1 2 1 2 1 3 2 x reform tempx 3 3 Matrix X is created 3 tempy 10 7 16 4 y reform tempy 3 1 In the same way matrix Y is formed 5 ix inv x The inverse matrix of X is calculated and stored in ix 6 theta mul ix y By obtaining the product of X 1 and Y the solution is obtained Moreover the procedure from step 5 and 6 can be realized by setting commands as the arguments of other commands theta mul inv x y The result of the above operation is the solution of a system of linear equations Chapter 5 Graphic Package Module GPM 5 1 Introduction GPM is the module for visualization of the data processed or analyzed by modules ISPP BPS NCS etc
104. s x initial weight file name loaded by LEARN x weight history file name x interval to store weight history x mode to store weight history x error history file name x interval to store error history x direction to store error history mode to store error history x input data file name for learning x teaching data file name for learning x first pattern number for learning x last pattern number for learning x learning mode x learning algorithm x learning rate x momentum x increasing factor for learning rate x reduction factor for learning rate x threshold for Vogl s method factor for Ochiai s method minimum error to stop learning maximum steps to stop learning x interval to display comments comment x weight file name for testing x weight history number for testing x test data file name x first pattern number for testing x last pattern number for testing x input layer number for testing CHAPTER 6 BACK PROPAGATION SIMULATOR BPS 76 output layer number for testing x test result file name MLP structure parameters The structure parameters for MLP namely the number of layers the number of units in each layer and the type of activation functions must be set in all cases of generating the initial values of connection weights learning testing tracing etc The parameters are set by the LAYER and FUNCTION commands The activation functions of units in the input layer the Oth layer are l
105. s language offers an environment in which the large scale physiological model can be described easily in NCS see also Figure 1 5 BPS Back Propagation Simulator is developed to examine neural network characteristics and capabilities Function for tracing weight change offers precious data for analysis of learning process local minima and internal network representation see also Figure 1 6 CHAPTER 1 WHAT IS SATELLITE 3 aERRREREEBLBBII LLL LLELELELELELELLLLELELERERERERERREEREBRBRBIETETLILELELELELE Welcome to The SATELLITE World Copyright C 1992 2002 BPEL Toyohashi University of Technology interactive operating system Version 2 94 y UN O O OO OOO O O OLO OOOO O O O O OO OOO OLO O OOO OOD O O O OOOO DODO DODO SYSTEM MODULE SYSTEM MODULE SYSTEM MODULE SYSTEM MODULE Ms 7 SYSTEM MODULE EE NC r SYSTEM MODULE SYSTEM MODULE SYSTEM MODULE 31 External Functions Install Ok 31 External Functions Install 4 41 External Functions Insta ysk 25 External Functions Insj411 Ok 19 External Functions tall Ok 24 External Functions Install Ok 24 External Function Install Ok 34 External Funct igfts Install Ok Fungtions Install Ok O te oo te pe C like descriptions Eat sstellite 2 9x 260 2 data dat y ER satellite 2 9x 261 2 ls da a a satellite 2 9x 262 2 for i 05 i lt 10 itt 4 if x i lt 5 lt xili 14 Telsel e Ja on line message
106. s the case of 2 dimensional Series object replacement as shown in Figure 3 12 SATELLITE tom rose 231 a 1714 SATELLITE tom rose 24 b 31736 SATELLITE tom rose 25 ar reform a 7 2 SATELLITE tom rose 26 br reform b 3 2 CHAPTER 3 SYSTEM MODULE SYSTEM 39 Figure 3 12 An example of using PUT on 2 dimensional Series object SATELLITE tom rose 27 c put ar br 2 0 SATELLITE tom rose 281 c Lo 01 1 2 1 01 3 4 2 01 30 31 3 01 32 33 4 01 34 35 5 01 11 12 6 01 13 14 SATELLITE tom rose 291 Similarly the operation can be performed on Snapshot objects 3 9 MERGE merging two data sets This command merges two objects together usage z merge x y x and y are the objects to link and z is the final object in which y is attached at the end x The subindex of x must be equal to the y s one in the case of multi dimensional objects The example of merging 2 dimensional Series objects ar and br is shown below and depicted Figure 3 13 SATELLITE tom rose 291 a 178 SATELLITE tom rose 30 b 31736 SATELLITE tom rose 311 ar reform a 4 2 SATELLITE tom rose 32 br reform b 3 2 SATELLITE tom rose 33 c merge ar br SATELLITE tom rose 34 c 0 0 1 2 1 Lol 3 4 2 0 5 6 3 0 7 8 4
107. sed one For example the reversed 1 dimensional Series object a is obtained by the following SATELLITE tom rose 491 a 177 SATELLITE tom rose 50 b reverse a SATELLITE tom rose 51 a 0 1 2 3 4 5 51 6 7 SATELLITE tom rose 52 b 0 7 6 5 4 3 5 2 1 ISATELLITE 1 tom rose 53 The case of 2 dimensional Series object see also Figure 3 18 SATELLITE tom rose 541 a 1714 SATELLITE tom rose 55 b reform a 7 2 SATELLITE tom rose 56 b Lo 01 1 2 1 01 3 4 2 01 5 6 3 01 7 8 4 01 9 10 5 01 11 12 6 01 13 14 SATELLITE tom rose 57 c reverse b SATELLITE tom rose 58 c 0 01 14 13 1 01 12 11 2 01 10 3 01 4 01 CHAPTER 3 SYSTEM MODULE SYSTEM 43 Figure 3 18 An example of using REVERSE on 2 dimensional Series object 5 01 4 3 6 01 2 1 SATELLITE tom rose 591 Similarly the operation can be performed on Snapshot objects 3 13 ROTATE rotating data This command moves the head pointer of an object to the specified position usage y rotate x index x is the original object y is the rotated object and index is the position of the front An example is shown in Figure 3 19 SATELLITE tom rose 591 a 177 SATELLITE tom rose 60 a 0 1 2 3 4 5 51 6 7 SATELLITE tom
108. served as files or printed In order to print graphics in a window proceed as follows gpm2ps GPMDVIFILE1 gt filename ps lpr Pxxx filename ps 65 CHAPTER 5 GRAPHIC PACKAGE MODULE GPM Table 5 1 Commands in GPM module Related to X windows wopen Open a window wclose Close a window we newpage chwin raph axis dran Tine Tabel gsoln color Factor Font type width scale size title Set labels of the coordinate axes origin Set the origin of the coordinate axes 66 CHAPTER 5 GRAPHIC PACKAGE MODULE GPM 67 0 0 0 2 0 4 0 6 0 8 1 0 Figure 5 1 A sinusoidal curve GPMDVIFILE1 is the middle file generated by GPM and xxx is a printer name GPM2PS command converts the file GPMDVIFILE1 to the PostScript PS file filename ps Encapsulated PostScript EPS file for LaTeX PowerPoint or tgif can be made as follows gpm2eps GPMDVIFILE1 gt filename eps 5 3 Examples The followings are examples showing the use of GPM commands See also the Command Reference for further details 5 3 1 Displaying 1 dimensional objects Example 1 displaying a sinusoidal curve Program from the line editor and also file can be processed by using the INLINE command see 2 9 2 Figure 5 1 shows the chart drawn by the following wopen 1 A4 0 1 Open a window A4 size t 07100 100 Substitute numerical values from O to 1 for the Series object t y sin 2 PI t Ca
109. sition of the nearest value val is the value to locate and num is the number of values to find As shown in Figure 3 28 for example we can obtain the first and second nearest values to the specified value 5 8 in a 1 dimensional Series object by the following SATELLITE tom rose 301 a SATELLITE tom rose 311 c DATA 6 POINT 5 DATA 5 POINT 4 177 2 find a 5 8 2 JSATELLITE tom rose 321 c o 071 5 1 01 4 SATELLITE tom rose 331 The following example demonstrates FIND on 2 dimensional Series object see also Figure 3 29 ISATELLITE tom rose 331 a SATELLITE tom rose 34 b SATELLITE tom rose 351 c DATA 7 POINT 3 0 DATA 6 POINT 2 1 1714 reform a 7 2 find b 6 8 2 JSATELLITE tom rose 36 c 0 01 3 0 1 01 2 1 SATELLITE tom rose 37 Similarly the operation can be performed on Snapshot objects CHAPTER 3 SYSTEM MODULE SYSTEM Figure 3 29 An example of using FIND on 2 dimensional Series object 49 Chapter 4 Interactive Signal Processing Package ISPP ISPP is the core module of SATEILITE A lot of processing functions are represented by commands which cover the methodology of the digital signal processing such as the preprocessing by window functions FFT spectrum analysis by linear prediction model filtering cepstrum analysis etc All commands requ
110. thod 2 6 Internal constants SATELLITE has defined several internal constants in order to ease programming or operating internal functions There are three kinds of internal constants CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 23 e Floating point constants such as 3 0 1 0e 5 etc and character sequence constants such as Wel come to SATELLITE World e Mathematical constants 180 a DEG 57 2957 The base of log E 2 7182 Euler s constant GAMMA 0 5772 Golden ratio PHI v5 1 2 1 6180 m PI 3 1415 e User defined constants defined by the command CONST For example SATELLITE tom rose 56 const Degree PI 180 SATELLITE treats the internal constants and user defined constants equally Moreover we can change the internal constant to the user defined one by the command CONST CONST can deal with the expression in which its right hand side is a formula or an object like Series It is evaluated right after it is defined The difference between variables and constants is just in permission of substituting objects 2 7 Control sequence As in C language we can use IF WHILE DO WHILE For as control sequences and for grouping statements together e if expri stmt1 e if expri stmti else stmt2 while expri stmt1 e do stmti while expri e for expri expr2 expr3 stmt1 expri expr2 and expr3 are general expressions including substitutions or functions s
111. ting buffer A command name will be completed and the full name will be displayed on the terminal and the editing buffer The first candidate is shown if the command cannot be specified uniquely The next candidate is called by pressing TAB key again For example suppose that there are six files in the current directory namely reportl tex report2 tex report3 tex workl tex work2 tex and work3 tex The file name or the directory name that starts with wo is searched and displayed from the current directory shown as follows SATELLITE tom rose 88 wo SATELLITE tom rose 88 work1 tex The 2nd candidate is displayed by pressing TAB key again as follows SATELLITE tom rose 88 work1 tex SATELLITE tom rose 88 work2 tex The candidates are searched in paths and order described by the environment variable PATH If the search cycle is completed the editing buffer is cleared After that if TAB key is pressed again the first candidate will be called again If there is no candidate there is nothing to display CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 13 Completion of file names UNIX commands and directory names can be performed in the arbitrary position of the editing buffer The keywords for discrimination between both cases are the blank just before the cursor the equaling character and the character sequence divided by a double quotation mark Reserved words or variable names in th
112. tively If the module does not have any inputs that is there is no input sentence the relative expression for such a module is given as follows mdl il lt QO The input of the cell module must be inputted through the components of the chemical or electrical synapse modules For example the description that the output from the j th component of the module mdl is substituted for the first input of its i th component through the k th component of the chemical or electric synapse is md1 1 lt syn k lt mdl j input2 It is also possible to designate the result of adding the outputs of components of modules to the input of the component of each module In this case we use to link them For example the second input of the i th component of the cell module mdl is the sum of the output of the n th component of mdl through the m th component of the synaptic module syn and the output of the l th component of md1 through the o th component of syn mdl i lt inputi syn m lt mdl n syn o lt mdl1 1 The component of a cell module which appears in the right hand side of a relative expression should be presented in the most left hand side of the different relative expression Note that we CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 96 never use the components of chemical or electrical synapse modules They necessarily have the components of cell modules on the input and output sides Examples con
113. tmt1 and stmt2 are single statements A set of statements in parentheses is also regarded as the single statement AND operator amp amp OR operator and other relation operators can be used in expressions If the result of evaluation of an expression is equal to zero it is treated as false or else true In the case where two or more results are obtained by a logical operation such as a comparison between two Series objects if all of them are not equal to zero it is regarded as true 2 7 1 IF sequence If the result of the conditional expression expri is true the first statement stmt1 is performed If the condition expr1 is evaluated as false the next statement stmt2 is executed instead of stmt1 CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS IF sequence 1 if expri stmt1 IF sequence 2 if expri stmt1 else stmt2 For example processing of If x is smaller than n then add x to s is described as follows SATELLITE tom rose 861 if x lt n e 8 8s x e hea aed SATELLITE tom rose 87 Processing of If x is smaller than n then add zx to s or else subtract x from s is described as follows SATELLITE tom rose 87 if x lt n e s s x gt else s s x e e SATELLITE tom rose 88 2 7 2 WHILE and DO WHILE sequences 24 WHILE and DO WHILE sequences controll the
114. ue of an integration constant membrane conductance of each ion Jya JK JL parameter Cm 1 0 mNa0 Gl 0 3 hNa0 0 05293 GNa 120 GK 36 0 5961 nKO 0 3177 VO 0 Above mentioned descriptions are used for function sentences The description of the electrical synapse type module for connecting two cells is shown in the following the name is G type GAP module G Its input is the voltage and its output is the current The initial value of the voltage is 0 with the delay time 0 1 s input VOP 0 1 0 output Ig The conductance of the electric synapse is defined as a parameter parameter GL 5 0 From Figure 7 4 the output current I is given by following equation I gx Vie Va 7 16 where V and Vz are the voltages at both ends of the electric synapse Thus the function sentence is described as follows function Ig GL VOP POSOUT Finally consider the network description The name of this module is SQUID type NETWORK module SQUID As shown in Figure 7 4 11 HH cells and 10 G synapses are necessary They are defined as follows cell HH 11 gap G 10 Relationship is described using the FOR sentence connection HH 0 lt G O lt HH O for n 1 n lt 10 n HH n lt G n 1 lt H n 1 G n lt HH n 1 HH 10 lt G 9 lt HH 9 The description using the NCS language shown in Listing 1 is complete
115. uld be defined before calling FFTC function Conversion between two object classes is automatically performed In this way we can change an object class to another one In case of the operation between String and Scalar objects for example the Scalar value is converted to a character sequence The resulting class is String e g SATELLITE tom rose 521 test 3 test3 SATELLITE tom rose 53 3 1415926 3 14159 SATELLITE tom rose 541 CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 21 The reason why the result of 3 1415926 becomes 3 14159 is due to round off for displaying A numerical value is obtained after changing String into Scalar as follows SATELLITE tom rose 54 3 3 1415926 6 1415926 SATELLITE tom rose 55 0 1 08e 2 0 0108 SATELLITE tom rose 56 Similarly other object conversions can be performed such as Series String String Series etc Conversion between more than two objects can be also performed 2 4 4 Type of object To obtain the object class of the variable whose class is unknown the TYPEOF command is used Suppose that x is Series object and y is Snapshot object Then we can get type of each class of these variables by the following SATELLITE tom rose 561 typeof x series SATELLITE tom rose 57 typeof y snapshot SATELLITE tom rose 58 In order to get the index of a object we use the I
116. uring previous learning is utilized as the initial weight files The example of the WINIT command execution is shown in Figure 6 2 The followings are displayed The MLP s structure and the parameter values for generating the initial values of connection weights If the initial weight file with given name does not exist the message that a new file has been created is displayed The following confirmation message is shown File filename already exists Overwrite y n If y the following message is displayed File filename has been overwritten If n is chosen the error message is shown In this case the file name has to be reset CHAPTER 6 BACK PROPAGATION SIMULATOR BPS 78 Sal hone kunita 26212 winit structure of network gt gt Humber of layer 3 layer num of cells condition 2 linear non bias 2 sigmoid append bias 1 linear append bias lt lt Initialize condition gt gt Init algorithm rangom Initialize data file name initut bhw Seed Max Height Min Weight 1 1 1 File inituf bhu is created AAN hone kunita 265 Figure 6 2 Execution of WINIT CHAPTER 6 BACK PROPAGATION SIMULATOR BPS 79 Threshold Input layer Threshold 1 237 4 5 6 Hidden layer 7 8 9 Output layer Ko Figure 6 3 The example of an MLP s structure 6 3 4 Learning Learning of MLP is carried out by the LEARN command LEARN reads input data
117. uting the simulation including the processing before the simulation starts numerical integration routines etc The simulation condition file group is the assembly of the files containing information on the execution of the simulation such as outside stimulation conditions model parameters signal delay information etc The simulation is carried out using the condition file group and the execution file 7 1 2 Concept of modularization The relationships between the type module and component in NCS are shown in Figure 7 2 and Figure 7 3 85 CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 86 Model Description NCS Language NCS Preprocessor NCS Library Simulation Program Create Compiler Linker oli Executive File Figure 7 1 Composition of NCS CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 87 Neural Network Module Horizontal Cell HC GABA Glutamate GLU HC 0 HC 1 Sena CONE 1 GABA O GABA 1 GLU 0 GLU 1 GAP 0 GAP 1 v NDS Neural Network Component Figure 7 2 Neural circuit elements and the model structure in NCS CONE module component si AAA CONE 0 CONE 1 CONE 2 1 CELL type e type os o are ae ee GABA 0 HC module Figure 7 3 Correspondence to the real neural circuit CHAPTER 7 NEURAL CIRCUIT SIMULATOR NCS 88 A neural circuit consists of large number of neurons combined by chemical synapses an
118. values of the weights change a little when the other weight values are fixed Then the data are presented to MLP the square errors of the outputs are calculated Since they are stored to the object it is possible to display them using GPM commands cor The command for obtaining the correlation matrix of objects Data are read from the weight history file to the object using the WGTLOAD command CHAPTER 6 BACK PROPAGATION SIMULATOR BPS O SATELLITE Languages dec500 amd fas home nagasaua SL_MAN Zangua DES sl _MAN Zaneyo 57 rec 0 1 display mode 1 4 1 minimum value 0 maximum value 2 1 FREE Portrait lt lt structure of network gt gt Number of layer 3 layer num of cells condition linear non bias 1 2 sigmoid append bias 2 1 linear append bias lt lt Testing condition gt gt Weight file for test wet bhw Weight history number 0 Input data file for test Input pattern number Input layer 0 Output layer 2 Result file for test test_out brc ERENT SL_MAN Zangyo 5812 test_in dat all Figure 6 10 Example of REC execution Chapter 7 Neural Circuit Simulator NCS 7 1 Introduction The information processing mechanism in the brain and nervous system of human is described mathemat ically based on the results of physiological experiments Generally mathematical models are described by nonlinear multi dimensional simultaneous different
119. variables in the chemical or electrical synapse module is one In order to set two or more cell modules use commas to link them Moreover it is possible to add delay information to the input sentence for the chemical and electrical synapse The delay information is described by the following input name_of_a_variable delay_time initial_value initial_value is output in the interval from 0 to delay_time Example Definition of Ig as an input variable input lg Definition of VOP as an input variable with the delay 0 1 and the initial value 0 0 input VOP 0 1 O output It defines the name of an output variable in a module The number of output variables is one Example Definition of V as an output variable output V observable It defines the name of the variable of which value is to be observed used in the function sentence Specify the output by the NOUT command The value of the defined variable can be observed while the simulation is running Example Definitions of INa IK 11 Ig as the variables to observe observable INa IK Il Ig constant It defines the name and value of a constant used in the function sentence Use the double notation for constants even if they are integers Example Definition of VNa VK Vl as constants with the values 115 12 10 6 respectively constant VNa 115 0 VK 12 0 V1 10 6 1The variables become doub
120. x rounding off decimal fractions mod x y the remainder of x y same as x y log x log x a natural logarithm log2 x log x log10 x logijo pow x y x same as x y sgn x the sign of x sin x sinz sqrt x yz tan x tanz List of system functions abort Termination of a program by force alias x y Alias operation history x History operation index x Acquisition of an object s index inline x Execution of a program from a file length x Acquisition of the number of data points elements CHAPTER 2 SATELLITE SHELL AND ITS FUNCTIONS 27 printf x Indication of objects read x Reading an object strlen x Acquisition of the length of a character sequence typeof x Acquisition of an object class undef x Elimination of a variable unix x Execution of UNIX command write_type Specification of a file type for writing GarCo Garbage collection Symbols Indication of a variable name in symbol table Note x and y are the function arguments stands for the arguments in which the number of them is variable 2 8 3 User defined function We can define functions and procedures of our own For example the function plusten that performs Add 10 to the argument n is given as follows SATELLITE tom rose 57 func plusten n return n 10 ei SATELLITE tom rose 58 The following is an example of calling this function SATELLITE tom rose 58 num

— User`s Manual —

Contents

Download Pdf Manuals

Related Search

Related Contents

&mdash; User`s Manual &mdash;

Contents

Download Pdf Manuals

Related Search

Related Contents

— User`s Manual —