This project implements a distributed neural network where "neurons" act as WebSocket servers and "dendrites" act as WebSocket clients. Neurons process incoming data, compute activations, and handle backpropagation using tokio, tungstenite, and async Rust features.
- Asynchronous WebSocket servers (neurons) and clients (dendrites).
- Dynamic IPv6 address assignment.
- Real-time backpropagation and neuron replacement.
- Mutex locks for thread safety.
- Neuron: Acts as a WebSocket server, listens for incoming signals from dendrites, computes forward propagation using sigmoid activation, and sends feedback.
- Dendrite: Connects to neuron servers and sends signals. Monitors output and participates in backpropagation.
- Error Calculation & Replacement: Replaces the worst-performing dendrites periodically, simulating network pruning.
- Rust (latest version).
- Clone the repository: git clone https://github.com/stanta/neuromorph.git
- cargo build
- cargo run