# Brain.js — GPU-Accelerated Neural Networks in JavaScript

> A JavaScript library for building and training neural networks in the browser and Node.js with optional GPU acceleration via headless-gl.

## Install

Save in your project root:

# Brain.js

## Quick Use
```bash
npm install brain.js
```
```js
import { NeuralNetwork } from 'brain.js';
const net = new NeuralNetwork();
net.train([
  { input: [0, 0], output: [0] },
  { input: [1, 0], output: [1] },
  { input: [0, 1], output: [1] },
  { input: [1, 1], output: [0] },
]);
console.log(net.run([1, 0])); // ~[1]
```

## Introduction
Brain.js is a lightweight JavaScript library for neural networks that runs in both the browser and Node.js. It focuses on simplicity, letting developers train feedforward, recurrent, and LSTM networks with just a few lines of code and no ML background required.

## What Brain.js Does
- Trains feedforward, recurrent, LSTM, and GRU neural networks in JavaScript
- Accelerates training with GPU.js for WebGL-based parallel computation
- Serializes trained models to JSON for storage and later reuse
- Supports time-series prediction, pattern recognition, and text classification
- Runs entirely client-side with no server dependency for inference

## Architecture Overview
Brain.js provides a set of network classes: NeuralNetwork for feedforward, RNNTimeStep and LSTMTimeStep for sequential data, and recurrent variants for text. Training runs a configurable backpropagation loop with learning rate, momentum, and error threshold parameters. GPU acceleration is opt-in via the NeuralNetworkGPU class which offloads matrix operations to WebGL shaders through GPU.js.

## Self-Hosting & Configuration
- Install via npm for Node.js or include via CDN for browser usage
- Choose the network type matching your data: feedforward for classification, LSTM for sequences
- Configure training options like iterations, error threshold, learning rate, and log interval
- Export trained models with toJSON and reload them with fromJSON
- For GPU acceleration in Node.js install gpu.js as a peer dependency

## Key Features
- Simple API that requires no knowledge of linear algebra or ML theory
- GPU acceleration via WebGL for faster training on supported environments
- JSON model serialization for saving, sharing, and deploying trained networks
- Multiple network architectures including LSTM and GRU for sequence tasks
- Runs in browsers and Node.js with the same API

## Comparison with Similar Tools
- **TensorFlow.js** — full ML framework with broader model support; Brain.js is simpler for basic networks
- **Synaptic** — similar concept but unmaintained; Brain.js is actively developed
- **ConvNetJS** — focused on CNNs and archived; Brain.js covers RNNs and LSTMs
- **ml5.js** — higher-level wrapper over TensorFlow.js; Brain.js is a standalone lightweight library
- **ONNX Runtime Web** — inference-only; Brain.js supports both training and inference

## FAQ
**Q: What tasks is Brain.js good for?**
A: Pattern recognition, simple classification, time-series prediction, and text categorization with small to medium datasets.

**Q: Can I use it for image recognition?**
A: It lacks convolutional layers, so for images consider TensorFlow.js or ONNX Runtime instead.

**Q: Does GPU acceleration work in all browsers?**
A: It requires WebGL support. Most modern desktop and mobile browsers support it.

**Q: How large can my training data be?**
A: Brain.js works best with small to medium datasets. For large-scale training, consider server-side frameworks.

## Sources
- https://github.com/BrainJS/brain.js
- https://brain.js.org

---
Source: https://tokrepo.com/en/workflows/asset-83222af5
Author: AI Open Source