AI Paper Atlas

Research map / Architectures and efficiency / Spiking networks

Towards Lossless Memory-efficient Training of Spiking Neural Networks via Gradient Checkpointing and Spike Compression

Yifan Huang, Wei Fang, Zecheng Hao, Zhengyu Ma, Yonghong Tian

ICLR 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
replace SpikingJelly CuPy LIF with Triton MELIF kernel (ablation)SNN BPTT training · CIFAR10-DVSpeak allocated memory (MB): 6131.07 -> 4865.06Improved
replace SpikingJelly CuPy LIF with Triton MELIF kernel (ablation)SNN BPTT training · CIFAR10-DVSthroughput (samples/s): 290.26 -> 331.30Improved
add layer-wise gradient checkpointing and spike compression (O1) (ablation)MELIF Spiking VGG · CIFAR10-DVSpeak allocated memory (MB): 4865.06 -> 2887.75Improved
add layer-wise gradient checkpointing and spike compression (O1) (ablation)MELIF Spiking VGG · CIFAR10-DVSthroughput (samples/s): 331.30 -> 246.81Hurt
add spatio-temporal segment partitioning (O3) (ablation)MELIF + O1 · CIFAR10-DVSpeak allocated memory (MB): 2887.75 -> 2349.39Improved
add greedy segment restoration (O4) (ablation)MELIF + O3 · CIFAR10-DVSthroughput (samples/s): 247.83 -> 270.79Improved
add spike compression at O1 (ablation)MELIF + GC Spiking VGG · CIFAR10-DVSpeak allocated memory (MB): 2892.63 -> 2887.75Improved
add spike compression at O3 (ablation)MELIF + GC + partitioning · CIFAR10-DVSpeak allocated memory (MB): 2530.66 -> 2349.39Improved
add AMPMELIF + O4 QKFormer · ImageNet 1/32 subsetpeak allocated memory (MB): 5219.93 -> 3158.02Improved
add LOMOMELIF + O4 QKFormer · ImageNet 1/32 subsetpeak allocated memory (MB): 5219.93 -> 5190.60No clear effect
add LOMO on top of AMPMELIF + O4 QKFormer · ImageNet 1/32 subsetpeak allocated memory (MB): 3158.02 -> 3142.86No clear effect
apply O4 pipeline to parallel spiking neuronsSEW ResNet-34 with PSN · ImageNet 1/32 subsetpeak allocated memory (MB): 7602.64 -> 2544.28Improved

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