BrainAI Spiking Core (LIF Family) + Surrogate Gradient Training

What problem does it solve?

Explicitly manage spiking neuron state to prevent hidden leakage and to enable stable, testable training with surrogate gradients, aligning behavior across CPU/GPU and research experiments.

Core Features & Use Cases

• Explicit state contract for LIF-family neurons (LIF, AdaptiveLIF, RecurrentLIF, AdvancedLIF) to support deterministic resets, carry, and clean sequence processing.
• Time-unrolled SNN support with a unified snn_unroll utility for truncated BPTT, plus surrogate gradient options (ATan, FastSigmoid, STE) and diagnostics.
• Extensibility and debugging aids: references, assets templates, and scripts for testing and validation.

Quick Start

Create a tiny network using LIFNeuron with a short input sequence and run it through snn_unroll to observe spikes and state transitions.