AI Paper Atlas

Research map / Architectures and efficiency / Spiking networks

Spik4lite: Refactoring Neuromorphic Sparsity for Efficient Spiking Neural Networks on Commodity Edge Devices

Yongzhi She, Qihua Zhou, Yuhao Wang, Yaodong Huang, Laizhong Cui, Jingcai Guo

ICML 2026 · Architectures and efficiency › Spiking networks · approach: Spiking transformers

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add Spik4lite compressionSpikingformer-4-384 · CIFAR10parameters (M): 9.32 -> 2.86Improved
add Spik4lite compressionSpikingformer-4-384 · CIFAR10Top-1 accuracy (%): 95.07 -> 94.43Hurt
add Spik4lite compressionSpikingformer-4-384 · CIFAR10energy per batch (J): 5.42 -> 3.42Improved
add Spik4lite compressionSpikformer-4-384 · CIFAR10peak memory (MiB): 2614 -> 1613Improved
add Spik4lite compressionSD-Transformer-2-256 · DVS-128 Gestureenergy per batch (J): 9.24 -> 2.85Improved
add Spik4lite compressionSD-Transformer-2-256 · DVS-128 Gesturelatency (s): 1.65 -> 0.46Improved
add Spik4lite compressionSpikformer-4-384 · CIFAR100Top-1 accuracy (%): 78.0 -> 74.19Hurt
add Spik4lite compressionSpikformer-2-256 · DVS-CIFAR10Top-1 accuracy (%): 79.9 -> 76.4Hurt
add Spik4lite compressionSpikingformer · CIFAR100 (Jetson Orin Nano)inference duration: approximately 130 s -> roughly 57 sImproved
increase penalty to prune 32% of channels (ablation)Spikingformer + Spik4lite · CIFAR10Top-1 accuracy (%): 95.07 -> 95.07No clear effect
increase penalty to prune 55% of channels (ablation)Spikingformer + Spik4lite · CIFAR10Top-1 accuracy (%): 95.07 -> 94.4Hurt
increase penalty to prune 90% of channels (ablation)Spikingformer + Spik4lite · CIFAR10Top-1 accuracy (%): 95.07 -> 83.98Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy