javascript implementation of neural networks

Javascript Implementation of Neural Networks

Overview

Network Implementation


function evaluate(layers, input){
    let result = input;
    for(let i=0;i<layers.length;i++){
        let layer = layers[i];
        result = layer.evaluate(result);
    }
    return result;
}

Backpropagation API

A neural network can be created using the backpropagation library. A network consists of an array of layers. Each layer follows the object description above. The library provides default implementations for a number of different layers.

The following code creates a neural network with one layer of weights with a bias, following by the relu activation function.

The evaluate method in the library will calculate the output of the network given an input.


let layers = [];
layers.push(ba.affine([[1,1]],  [[1]]));
layers.push(ba.relu());

let input = [[1],[4]];

let test = evaluate(layers, input);