Introduction: Arduino-Nano and Artificial Neural Networks

By plus.google.com/+SaiSurya300576
About: Software Technical Architect by profession; Author of 'Devil's Vengeance' (on Kindle); Blogger; Electronics Hobbyist;

Hi Friends,

Here is an implementation of Artificial Neural Networks (ANN) on Arduino Nano board, that I have done recently.


The steps below will explain how a sample ANN program can be trained to learn the XOR truth table outputs very efficiently.

This can further be extended to any truth table .

Thanks to these resources I have referenced and they are so amazingly easy to learn.

Check these for your reading!

Step 1: Components Required

  • Arduino Nano
  • OLED Display Module
  • Joystick Module (Analog 2 Axis) Breakout Board

The reason I chose Arduino Nano is that it is best suited for compact design

Few details below, from the Arduino site >> https://store.arduino.cc/usa/arduino-nano

The Arduino Nano is a small, complete, and breadboard-friendly board based on the ATmega328 (Arduino Nano 3.x). It has more or less the same functionality of the Arduino Duemilanove, but in a different package. It lacks only a DC power jack, and works with a Mini-B USB cable instead of a standard one.

Step 2: The Connections for Display and Inputs (joystick)

Oled Module

This is very easy to connect and a wide range of libraries are available from simple to advanced display requirements.

JoyStick Module

This type of input control is chosen because it works well with menu item navigation and selection.

Step 3: Menu Creation for Sample Input for the Neural Network


Training Data & User Inputs

As mentioned earlier we use the above XOR Truth table as the training data and test the program using these inputs to verify the actual ouputs with these outputs.

  • The user needs to be able to supply (0,0), (1,0), (0,1) & (1,1) from the device we are building.



Oled Menu Creation

The menu is created using the standard program from the examples in the Arduino Editor, available once the u8glib library is installed. Here is the menu in use in the picture.

  • Set Sample Input Menu Item is the one the user selects to send inputs (listed above) to the neural network program

Step 4: The Artificial Neural Network Program on Arduino

NOTE : NOT THE ACTUAL PROGRAM!!
BELOW IS SHOWN FROM hobbizine, FOR REFERENCE Only



REF : <a href="http://robotics.hobbizine.com/arduinoann.html">hobbizine</a> -> A great article to learn ANN programming on Arduino!

#include <math.h>

/******************************************************************
 * Network Configuration - customized per network 
 ******************************************************************/

const int PatternCount = 10;
const int InputNodes = 7;
const int HiddenNodes = 8;
const int OutputNodes = 4;
const float LearningRate = 0.3;
const float Momentum = 0.9;
const float InitialWeightMax = 0.5;
const float Success = 0.0004;

const byte Input[PatternCount][InputNodes] = {
  { 1, 1, 1, 1, 1, 1, 0 },  // 0
  { 0, 1, 1, 0, 0, 0, 0 },  // 1
  { 1, 1, 0, 1, 1, 0, 1 },  // 2
  { 1, 1, 1, 1, 0, 0, 1 },  // 3
  { 0, 1, 1, 0, 0, 1, 1 },  // 4
  { 1, 0, 1, 1, 0, 1, 1 },  // 5
  { 0, 0, 1, 1, 1, 1, 1 },  // 6
  { 1, 1, 1, 0, 0, 0, 0 },  // 7 
  { 1, 1, 1, 1, 1, 1, 1 },  // 8
  { 1, 1, 1, 0, 0, 1, 1 }   // 9
}; 

const byte Target[PatternCount][OutputNodes] = {
  { 0, 0, 0, 0 },  
  { 0, 0, 0, 1 }, 
  { 0, 0, 1, 0 }, 
  { 0, 0, 1, 1 }, 
  { 0, 1, 0, 0 }, 
  { 0, 1, 0, 1 }, 
  { 0, 1, 1, 0 }, 
  { 0, 1, 1, 1 }, 
  { 1, 0, 0, 0 }, 
  { 1, 0, 0, 1 } 
};

/******************************************************************
 * End Network Configuration
 ******************************************************************/


int i, j, p, q, r;
int ReportEvery1000;
int RandomizedIndex[PatternCount];
long  TrainingCycle;
float Rando;
float Error;
float Accum;


float Hidden[HiddenNodes];
float Output[OutputNodes];
float HiddenWeights[InputNodes+1][HiddenNodes];
float OutputWeights[HiddenNodes+1][OutputNodes];
float HiddenDelta[HiddenNodes];
float OutputDelta[OutputNodes];
float ChangeHiddenWeights[InputNodes+1][HiddenNodes];
float ChangeOutputWeights[HiddenNodes+1][OutputNodes];

void setup(){
  Serial.begin(9600);
  randomSeed(analogRead(3));
  ReportEvery1000 = 1;
  for( p = 0 ; p < PatternCount ; p++ ) {    
    RandomizedIndex[p] = p ;
  }
}  

void loop (){


/******************************************************************
* Initialize HiddenWeights and ChangeHiddenWeights 
******************************************************************/

  for( i = 0 ; i < HiddenNodes ; i++ ) {    
    for( j = 0 ; j <= InputNodes ; j++ ) { 
      ChangeHiddenWeights[j][i] = 0.0 ;
      Rando = float(random(100))/100;
      HiddenWeights[j][i] = 2.0 * ( Rando - 0.5 ) * InitialWeightMax ;
    }
  }
/******************************************************************
* Initialize OutputWeights and ChangeOutputWeights
******************************************************************/

  for( i = 0 ; i < OutputNodes ; i ++ ) {    
    for( j = 0 ; j <= HiddenNodes ; j++ ) {
      ChangeOutputWeights[j][i] = 0.0 ;  
      Rando = float(random(100))/100;        
      OutputWeights[j][i] = 2.0 * ( Rando - 0.5 ) * InitialWeightMax ;
    }
  }
  Serial.println("Initial/Untrained Outputs: ");
  toTerminal();
/******************************************************************
* Begin training 
******************************************************************/

  for( TrainingCycle = 1 ; TrainingCycle < 2147483647 ; TrainingCycle++) {    

/******************************************************************
* Randomize order of training patterns
******************************************************************/

    for( p = 0 ; p < PatternCount ; p++) {
      q = random(PatternCount);
      r = RandomizedIndex[p] ; 
      RandomizedIndex[p] = RandomizedIndex[q] ; 
      RandomizedIndex[q] = r ;
    }
    Error = 0.0 ;
/******************************************************************
* Cycle through each training pattern in the randomized order
******************************************************************/
    for( q = 0 ; q < PatternCount ; q++ ) {    
      p = RandomizedIndex[q];

/******************************************************************
* Compute hidden layer activations
******************************************************************/

      for( i = 0 ; i < HiddenNodes ; i++ ) {    
        Accum = HiddenWeights[InputNodes][i] ;
        for( j = 0 ; j < InputNodes ; j++ ) {
          Accum += Input[p][j] * HiddenWeights[j][i] ;
        }
        Hidden[i] = 1.0/(1.0 + exp(-Accum)) ;
      }

/******************************************************************
* Compute output layer activations and calculate errors
******************************************************************/

      for( i = 0 ; i < OutputNodes ; i++ ) {    
        Accum = OutputWeights[HiddenNodes][i] ;
        for( j = 0 ; j < HiddenNodes ; j++ ) {
          Accum += Hidden[j] * OutputWeights[j][i] ;
        }
        Output[i] = 1.0/(1.0 + exp(-Accum)) ;   
        OutputDelta[i] = (Target[p][i] - Output[i]) * Output[i] * (1.0 - Output[i]) ;   
        Error += 0.5 * (Target[p][i] - Output[i]) * (Target[p][i] - Output[i]) ;
      }

/******************************************************************
* Backpropagate errors to hidden layer
******************************************************************/

      for( i = 0 ; i < HiddenNodes ; i++ ) {    
        Accum = 0.0 ;
        for( j = 0 ; j < OutputNodes ; j++ ) {
          Accum += OutputWeights[i][j] * OutputDelta[j] ;
        }
        HiddenDelta[i] = Accum * Hidden[i] * (1.0 - Hidden[i]) ;
      }


/******************************************************************
* Update Inner-->Hidden Weights
******************************************************************/


      for( i = 0 ; i < HiddenNodes ; i++ ) {     
        ChangeHiddenWeights[InputNodes][i] = LearningRate * HiddenDelta[i] + Momentum * ChangeHiddenWeights[InputNodes][i] ;
        HiddenWeights[InputNodes][i] += ChangeHiddenWeights[InputNodes][i] ;
        for( j = 0 ; j < InputNodes ; j++ ) { 
          ChangeHiddenWeights[j][i] = LearningRate * Input[p][j] * HiddenDelta[i] + Momentum * ChangeHiddenWeights[j][i];
          HiddenWeights[j][i] += ChangeHiddenWeights[j][i] ;
        }
      }

/******************************************************************
* Update Hidden-->Output Weights
******************************************************************/

      for( i = 0 ; i < OutputNodes ; i ++ ) {    
        ChangeOutputWeights[HiddenNodes][i] = LearningRate * OutputDelta[i] + Momentum * ChangeOutputWeights[HiddenNodes][i] ;
        OutputWeights[HiddenNodes][i] += ChangeOutputWeights[HiddenNodes][i] ;
        for( j = 0 ; j < HiddenNodes ; j++ ) {
          ChangeOutputWeights[j][i] = LearningRate * Hidden[j] * OutputDelta[i] + Momentum * ChangeOutputWeights[j][i] ;
          OutputWeights[j][i] += ChangeOutputWeights[j][i] ;
        }
      }
    }

/******************************************************************
* Every 1000 cycles send data to terminal for display
******************************************************************/
    ReportEvery1000 = ReportEvery1000 - 1;
    if (ReportEvery1000 == 0)
    {
      Serial.println(); 
      Serial.println(); 
      Serial.print ("TrainingCycle: ");
      Serial.print (TrainingCycle);
      Serial.print ("  Error = ");
      Serial.println (Error, 5);

      toTerminal();

      if (TrainingCycle==1)
      {
        ReportEvery1000 = 999;
      }
      else
      {
        ReportEvery1000 = 1000;
      }
    }    


/******************************************************************
* If error rate is less than pre-determined threshold then end
******************************************************************/

    if( Error < Success ) break ;  
  }
  Serial.println ();
  Serial.println(); 
  Serial.print ("TrainingCycle: ");
  Serial.print (TrainingCycle);
  Serial.print ("  Error = ");
  Serial.println (Error, 5);

  toTerminal();

  Serial.println ();  
  Serial.println ();
  Serial.println ("Training Set Solved! ");
  Serial.println ("--------"); 
  Serial.println ();
  Serial.println ();  
  ReportEvery1000 = 1;
}

void toTerminal()
{

  for( p = 0 ; p < PatternCount ; p++ ) { 
    Serial.println(); 
    Serial.print ("  Training Pattern: ");
    Serial.println (p);      
    Serial.print ("  Input ");
    for( i = 0 ; i < InputNodes ; i++ ) {
      Serial.print (Input[p][i], DEC);
      Serial.print (" ");
    }
    Serial.print ("  Target ");
    for( i = 0 ; i < OutputNodes ; i++ ) {
      Serial.print (Target[p][i], DEC);
      Serial.print (" ");
    }
/******************************************************************
* Compute hidden layer activations
******************************************************************/

    for( i = 0 ; i < HiddenNodes ; i++ ) {    
      Accum = HiddenWeights[InputNodes][i] ;
      for( j = 0 ; j < InputNodes ; j++ ) {
        Accum += Input[p][j] * HiddenWeights[j][i] ;
      }
      Hidden[i] = 1.0/(1.0 + exp(-Accum)) ;
    }

/******************************************************************
* Compute output layer activations and calculate errors
******************************************************************/

    for( i = 0 ; i < OutputNodes ; i++ ) {    
      Accum = OutputWeights[HiddenNodes][i] ;
      for( j = 0 ; j < HiddenNodes ; j++ ) {
        Accum += Hidden[j] * OutputWeights[j][i] ;
      }
      Output[i] = 1.0/(1.0 + exp(-Accum)) ; 
    }
    Serial.print ("  Output ");
    for( i = 0 ; i < OutputNodes ; i++ ) {       
      Serial.print (Output[i], 5);
      Serial.print (" ");
    }
  }


}

Step 5: ANN in Action on Arduino Nano!!

Here is the demo of ANN for XOR training data in action!

Happy Programming!