khoih-prog/simplenn

Simple neural network builder and feeder without learning function.

SimpleNN

Simple arbitrary neural network builder and feeder without learning function for arduino. Implement trained neural network into your project.

For use you need:

• Predefined biases
• Predefined weights
• Neural network structure
• Activation function

Features

• User custom activation function

How to define neural network in SimpleNN library

The given example is in the 'examples' folder.

1. Define weights:

float weights[] = {
  W0,   W1,
  W2,   W3,
  W4,   W5,
  
  W6,   W7,   W8,   W9,
  W10,  W11,  W12,  W13
}

2. Define biases for each node (except nodes on first layer):

float biases[] = {
  B20, B21, B30, B31, B32, B33
}

3. Define neural network structure (number of nodes on each layer):

unsigned int NNStructure[] = {
  3, 2, 4
}

4. Define your activation function (for example Rely function):

float fRely(float x){ 
  return x > 0 ? x : 0;
}

5. Define neural network:

SimpleNN NeuralNetwork( NNStructure, 3, &fRely, weights, biases);

Feedforward:

Reading sensors (just for the example. You are free to use your values as you want), normalizing values from 0.0f to 1.0f by dividing them by 1023, because we know, that values from analog input range between 0 and 1023. Then reading the output. Output is index of maximum value in output layer.

float input[3];
int state;
input[0] = analogRead(L_EYE) / 1023.0f;
input[1] = analogRead(R_EYE) / 1023.0f;
input[2] = analogRead(LD) / 1023.0f;
state = NeuralNetwork.feedForward(input);
Serial.println(state);