人工智能/神经网络

开发平台：
Visual C++

ART1.C：源码内容
							/******************************************************************************
                      ===========================
        Network:      Adaptive Resonance Theory 1
                      ===========================
        Application:  Brain Modeling
                      Stability-Plasticity Demonstration
        Author:       Karsten Kutza
        Date:         27.6.96
        Reference:    G.A. Carpenter, S. Grossberg
                      A Massively Parallel Architecture
                      for a Self-Organizing Neural Pattern Recognition Machine
                      Computer Vision, Graphics, and Image Processing, 37,
                      pp. 54-115, 1987
 ******************************************************************************/
/******************************************************************************
                            D E C L A R A T I O N S
 ******************************************************************************/
#include <stdlib.h>
#include <stdio.h>
#include <math.h>
typedef int           BOOL;
typedef char          CHAR;
typedef int           INT;
typedef double        REAL;
#define FALSE         0
#define TRUE          1
#define NOT           !
#define AND           &&
#define OR            ||
#define MIN_REAL      -HUGE_VAL
#define MAX_REAL      +HUGE_VAL
typedef struct {                     /* A LAYER OF A NET:                     */
        INT           Units;         /* - number of units in this layer       */
        BOOL*         Output;        /* - output of ith unit                  */
        REAL**        Weight;        /* - connection weights to ith unit      */
        BOOL*         Inhibited;     /* - inhibition status of ith F2 unit    */
} LAYER;
typedef struct {                     /* A NET:                                */
        LAYER*        F1;            /* - F1 layer                            */
        LAYER*        F2;            /* - F2 layer                            */
        INT           Winner;        /* - last winner in F2 layer             */
        REAL          A1;            /* - A parameter for F1 layer            */
        REAL          B1;            /* - B parameter for F1 layer            */
        REAL          C1;            /* - C parameter for F1 layer            */
        REAL          D1;            /* - D parameter for F1 layer            */
        REAL          L;             /* - L parameter for net                 */
        REAL          Rho;           /* - vigilance parameter                 */
} NET;
/******************************************************************************
               A P P L I C A T I O N - S P E C I F I C   C O D E
 ******************************************************************************/
#define N             5
#define M             10
#define NO_WINNER     M
#define NUM_DATA      30
CHAR                  Pattern[NUM_DATA][N] = { "   O ",
                                               "  O O",
                                               "    O",
                                               "  O O",
                                               "    O",
                                               "  O O",
                                               "    O",
                                               " OO O",
                                               " OO  ",
                                               " OO O",
                                               " OO  ",
                                               "OOO  ",
                                               "OO   ",
                                               "O    ",
                                               "OO   ",
                                               "OOO  ",
                                               "OOOO ",
                                               "OOOOO",
                                               "O    ",
                                               " O   ",
                                               "  O  ",
                                               "   O ",
                                               "    O",
                                               "  O O",
                                               " OO O",
                                               " OO  ",
                                               "OOO  ",
                                               "OO   ",
                                               "OOOO ",
                                               "OOOOO"  };
BOOL                  Input [NUM_DATA][N];
BOOL                  Output[NUM_DATA][M];
FILE*                 f;
void InitializeApplication(NET* Net)
{
  INT n,i,j;
  for (n=0; n<NUM_DATA; n++) {
    for (i=0; i<N; i++) {
      Input[n][i] = (Pattern[n][i] == 'O');
    }
  }
  Net->A1  = 1;
  Net->B1  = 1.5;
  Net->C1  = 5;
  Net->D1  = 0.9;
  Net->L   = 3;
  Net->Rho = 0.9;
  for (i=0; i<Net->F1->Units; i++) {
    for (j=0; j<Net->F2->Units; j++) {
      Net->F1->Weight[i][j] = (Net->B1 - 1) / Net->D1 + 0.2;
      Net->F2->Weight[j][i] = Net->L / (Net->L - 1 + N) - 0.1;
    }
  }
  f = fopen("ART1.txt", "w");
}
void WriteInput(NET* Net, BOOL* Input)
{
  INT i;
   
  for (i=0; i<N; i++) {
    fprintf(f, "%c", (Input[i]) ? 'O' : ' ');
  }
  fprintf(f, " -> ");
}
void WriteOutput(NET* Net, BOOL* Output)
{
  if (Net->Winner != NO_WINNER)
    fprintf(f, "Class %in", Net->Winner);
  else
    fprintf(f, "new Input and all Classes exhaustedn");
}
void FinalizeApplication(NET* Net)
{
  fclose(f);
}
/******************************************************************************
                          I N I T I A L I Z A T I O N
 ******************************************************************************/
void GenerateNetwork(NET* Net)
{
  INT i;
  Net->F1 = (LAYER*) malloc(sizeof(LAYER));
  Net->F2 = (LAYER*) malloc(sizeof(LAYER));
      
  Net->F1->Units     = N;
  Net->F1->Output    = (BOOL*)  calloc(N, sizeof(BOOL));
  Net->F1->Weight    = (REAL**) calloc(N, sizeof(REAL*));
  Net->F2->Units     = M;
  Net->F2->Output    = (BOOL*)  calloc(M, sizeof(BOOL));
  Net->F2->Weight    = (REAL**) calloc(M, sizeof(REAL*));
  Net->F2->Inhibited = (BOOL*)  calloc(M, sizeof(BOOL));
  for (i=0; i<N; i++) {
    Net->F1->Weight[i] = (REAL*) calloc(M, sizeof(REAL));
  }
  for (i=0; i<M; i++) {
    Net->F2->Weight[i] = (REAL*) calloc(N, sizeof(REAL));
  }
}
void SetInput(NET* Net, BOOL* Input)
{
  INT  i;
  REAL Activation;
   
  for (i=0; i<Net->F1->Units; i++) {
    Activation = Input[i] / (1 + Net->A1 * (Input[i] + Net->B1) + Net->C1);
    Net->F1->Output[i] = (Activation > 0);
  }
}
void GetOutput(NET* Net, BOOL* Output)
{
  INT i;
   
  for (i=0; i<Net->F2->Units; i++) {
    Output[i] = Net->F2->Output[i];
  }
}
/******************************************************************************
                     P R O P A G A T I N G   S I G N A L S
 ******************************************************************************/
void PropagateToF2(NET* Net)
{
  INT  i,j;
  REAL Sum, MaxOut;
   
  MaxOut = MIN_REAL;
  Net->Winner = NO_WINNER;
  for (i=0; i<Net->F2->Units; i++) {
    if (NOT Net->F2->Inhibited[i]) {
      Sum = 0;
      for (j=0; j<Net->F1->Units; j++) {
        Sum += Net->F2->Weight[i][j] * Net->F1->Output[j];
      }
      if (Sum > MaxOut) {
        MaxOut = Sum;
        Net->Winner = i;
      }
    }
    Net->F2->Output[i] = FALSE;
  }
  if (Net->Winner != NO_WINNER)
    Net->F2->Output[Net->Winner] = TRUE;
}
void PropagateToF1(NET* Net, BOOL* Input)
{
  INT  i;
  REAL Sum, Activation;
   
  for (i=0; i<Net->F1->Units; i++) {
    Sum = Net->F1->Weight[i][Net->Winner] * Net->F2->Output[Net->Winner];
    Activation = (Input[i] + Net->D1 * Sum - Net->B1) /
                 (1 + Net->A1 * (Input[i] + Net->D1 * Sum) + Net->C1);
    Net->F1->Output[i] = (Activation > 0);
  }
}
/******************************************************************************
                        A D J U S T I N G   W E I G H T S
 ******************************************************************************/
REAL Magnitude(NET* Net, BOOL* Input)
{
  INT  i;
  REAL Magnitude;
  Magnitude = 0;
  for (i=0; i<Net->F1->Units; i++) {
    Magnitude += Input[i];
  }
  return Magnitude;
}
void AdjustWeights(NET* Net)
{
  INT  i;
  REAL MagnitudeInput_;
  for (i=0; i<Net->F1->Units; i++) {
    if (Net->F1->Output[i]) {
      MagnitudeInput_ = Magnitude(Net, Net->F1->Output);
      Net->F1->Weight[i][Net->Winner] = 1;
      Net->F2->Weight[Net->Winner][i] = Net->L / (Net->L - 1 + MagnitudeInput_);
    }
    else {
      Net->F1->Weight[i][Net->Winner] = 0;
      Net->F2->Weight[Net->Winner][i] = 0;
    }
  }
}
/******************************************************************************
                      S I M U L A T I N G   T H E   N E T
 ******************************************************************************/
void SimulateNet(NET* Net, BOOL* Input, BOOL* Output)
{
  INT  i;
  BOOL Resonance, Exhausted;
  REAL MagnitudeInput, MagnitudeInput_;
  WriteInput(Net, Input);      
  for (i=0; i<Net->F2->Units; i++) {
    Net->F2->Inhibited[i] = FALSE;
  }
  Resonance = FALSE;
  Exhausted = FALSE;
  do {
    SetInput(Net, Input);
    PropagateToF2(Net);
    GetOutput(Net, Output);
    if (Net->Winner != NO_WINNER) {
      PropagateToF1(Net, Input);
      MagnitudeInput = Magnitude(Net, Input);
      MagnitudeInput_ = Magnitude(Net, Net->F1->Output);
      if ((MagnitudeInput_ / MagnitudeInput) < Net->Rho)
        Net->F2->Inhibited[Net->Winner] = TRUE;
      else
        Resonance = TRUE;
    }
    else Exhausted = TRUE;
  } while (NOT (Resonance OR Exhausted));
  if (Resonance)
    AdjustWeights(Net);
  WriteOutput(Net, Output);      
}
/******************************************************************************
                                    M A I N
 ******************************************************************************/
void main()
{
  NET Net;
  INT n;
  GenerateNetwork(&Net);
  InitializeApplication(&Net);
  for (n=0; n<NUM_DATA; n++) {
    SimulateNet(&Net, Input[n], Output[n]);
  }
  FinalizeApplication(&Net);
}

						