Architecture: 40 → 512 (recurrent adLIF, spike-to-spike) → 12
Neuron model: Adaptive Leaky Integrate-and-Fire (adLIF) with spike-to-spike delta encoding
Training: Surrogate gradient BPTT, fast-sigmoid surrogate (scale=25)
Hardware target: Catalyst N1/N2/N3 neuromorphic processors
Quantization: Float32 weights -> int16, membrane decay -> 12-bit fixed-point

Results

Metric	Value
Float accuracy	88.0%
Quantized accuracy (int16)	87.5%
Parameters	290,828
Quantization loss	0.5%

Reproduce

bash

git clone https://github.com/catalyst-neuromorphic/catalyst-benchmarks.git
cd catalyst-benchmarks
pip install -e .
python gsc/train.py --device cuda:0

Deploy to Catalyst Hardware

python

import catalyst_cloud

client = catalyst_cloud.Client()
result = client.simulate(
    model="catalyst-neuromorphic/gsc-snn-benchmark",
    input_data=your_spikes,
    processor="n2"
)

Citation

bibtex

@misc{catalyst-benchmarks-2026,
  author = {Shulayev Barnes, Henry},
  title = {Catalyst Neuromorphic Benchmarks},
  year = {2026},
  url = {https://github.com/catalyst-neuromorphic/catalyst-benchmarks}
}

Catalyst-Neuromorphic/gsc-snn-benchmark

Catalyst GSC SNN Benchmark

Model Description

Results

Reproduce

Deploy to Catalyst Hardware

Links

Citation

Catalyst-Neuromorphic/
gsc-snn-benchmark