Building a Tiny Neural Network

You can use Openmadness to build a basic neural network prototype — just enough to demonstrate a forward pass using matrix operations.

This section walks you through creating a simple, 3-layer neural net: one input layer, one hidden layer, and one output layer. There's no training involved — this is a static prototype for understanding how data flows through layers.

Step 1: Import and Define Inputs

import { omArray } from 'openmadness';

const inputs = omArray([
  [1, 0],
  [0, 1]
]);

Each row represents one input sample with two features.

Step 2: Define Random Weights

const weights1 = omArray([
  [Math.random(), Math.random()],
  [Math.random(), Math.random()]
]);

const weights2 = omArray([
  [Math.random()],
  [Math.random()]
]);
  • weights1 connects the input layer to the hidden layer (2x2)
  • weights2 connects the hidden layer to the output layer (2x1)

Step 3: Add Sigmoid Function

const sigmoid = (x) => 1 / (1 + Math.exp(-x));

Step 4: Forward Pass

const hidden = inputs
  .dot(weights1)
  .map(row => row.map(sigmoid));

const output = omArray(hidden)
  .dot(weights2)
  .map(row => row.map(sigmoid));

console.log('Output:', output.values);

Run the Code

To run the script:

node tiny-net.js

Example output:

Output: [ [ 0.74 ], [ 0.68 ] ]

Each output value corresponds to a prediction for the input sample.

Next Steps

This forward-pass prototype can be expanded by:

  • Adding more hidden layers
  • Replacing sigmoid with ReLU or another activation
  • Implementing error calculation and training via backpropagation.