#deep-learning #spiking #neuromorphic #snn #burn

neuburn

Spiking neural network framework on top of Burn — Rust-native alternative to snntorch

1 unstable release

0.1.0 Mar 8, 2026

#1505 in Machine learning

MIT/Apache

52KB
1K SLoC

neuBurn — Spiking neural network framework on top of Burn. Rust-native alternative to snntorch.


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

[dependencies]
neuburn = "0.1"
burn = { version = "0.14", features = ["train"] }
use neuburn::prelude::*;

License

MIT OR Apache-2.0

Dependencies

~61–100MB
~2M SLoC