5-Layer Neural Network

Deep Learning from Scratch, Visualized in Haskell

This is a machine learning XOR logic using pure functional programming.

---

The Engine (The Math)

The network is a chain of matrix transformations. Every neuron is a tiny calculator performing:

$$Output = \sigma(\sum(Input \cdot Weight) + Bias)$$

• Weights ($W$): The synapses. Visualized as cyan (strong positive) and orange (strong negative) connections.
• Backpropagation: The math behind the magic. The network calculates the Gradient ($\nabla C$) to determine how to nudge weights to reduce error.
• Cost Function: We use Mean Squared Error to quantify the "confusion": $$MSE = \frac{1}{n} \sum (actual - predicted)^2$$

---

Interactive Controls

Key	Action	Visual Feedback
R	Reset Brain	Scrambles weights; watch the loss graph spike.
Watch	Signal Flow	White sparks trace paths through high-weight connections.
Analyze	Loss Graph	The green line tracks "confusion." Flatline = Convergence.

---

Project Structure

.
├── .github/
│   └── PULL_REQUEST_TEMPLATE.md
├── Assets/           
│   └── NeuralNetwork.gif
├── .gitignore        
├── CODE_OF_CONDUCT.md
├── CONTRIBUTING.md      
├── LICENSE           
├── SECURITY.md          
├── Main.hs           
├── NeuralNet.hs      
└── README.md

---

Run it

Get the Dependencies:

cabal update
cabal install gloss

Build & Fire it up:

ghc -O2 Main.hs -o main
./main

[!TIP] Running on a potato? If your FPS stutters, open Main.hs, find the step function, and change the training iterations from !! 10 to !! 5. Your CPU will thank you.

---

Contributing

Love this project? Whether it's fixing a bug or adding a new "brain" feature, your help is welcome!

1. Check out the Contributing Guidelines.
2. Read the Code of Conduct.
3. Fork it and submit a PR!

---

Security

If you discover any security-related issues, please review our Security Policy.

---

License & Disclaimer

Built for educational purposes. This is a "vanilla" implementation—no heavy ML frameworks, no TensorFlow, no PyTorch. Just raw Haskell and pure math.

This project is licensed under the MIT License - see the LICENSE file for details.

---

Built by mazerissa