neuBurn
Spiking neural network (SNN) framework built on Burn — a Rust-native alternative to snntorch.
Features
- Neuron models: Leaky (LIF), Synaptic, Alpha, RLeaky, RSynaptic, Lapicque
- Reset modes: Subtract, Zero, None (snntorch-style)
- Learnable membrane decay (β) and optional learnable threshold
- Surrogate gradients: FastSigmoid, Sigmoid, StraightThrough, ATan
- Encoding: rate (Poisson), latency, delta
- Losses: MSE count, CE rate, CE temporal
- Layers: SpikingConv2d, LeakyParallel, BNTT (BatchNorm through time), SLSTM, KAN-based event-driven blocks
Usage
[]
= "0.1"
= { = "0.14", = ["train"] }
use *;
License
MIT OR Apache-2.0