AI Paper Atlas

Research map / Architectures and efficiency / Spiking networks

LIF Recurrent Memory Enables Long-Horizon Spiking Computation

Fenghao Liu, Yipeng Shen, Peng Chen, Qian Zheng, Peng Lin, Gang Pan

ICML 2026 · Architectures and efficiency › Spiking networks · approach: Bio-inspired spiking neuron models

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 recurrent feedback from context neuron NC to memory neuron NM (ablation)LRMM · PS-MNISTaccuracy: 86.11 -> 96.52Improved
Add recurrent feedback from context neuron NC to memory neuron NM (ablation)LRMM · SHDaccuracy: 86.28 -> 94.70Improved
Add global output recurrence from NO to the current layer (ablation)LRMM · PS-MNISTaccuracy: 93.74 -> 96.52Improved
Use three separate modulatory gates instead of one shared static gate (ablation)LRMM · PS-MNISTaccuracy: 92.35 -> 96.52Improved
Use three separate modulatory gates instead of one shared static gate (ablation)LRMM · SHDaccuracy: 90.81 -> 94.70Improved
Use all three modulatory currents (II, IF, IO) (ablation)LRMM · supplementary ablations (Appendix Tables 7-9)accuracyImproved
Vary the leak factor over a broad range (ablation)LRMM · supplementary ablationsaccuracyNo clear effect
Train with modulation-function annealing instead of removing modulation (ablation)LRMM · supplementary ablationsaccuracy and stabilityImproved
Replace NI and NO with ALIF neuronsLRMM · PS-MNISTaccuracy: 96.52 -> 97.39Improved
Use LRMM memory loop instead of a much larger ALIF baselinecopy task, delay 20 · copy task L=8, K=8accuracy: 0.58 -> 0.89Improved
Use LRMM instead of LSTM for energySHD inference · SHD, 2 layers x 512 neuronsenergy (nJ): 21145 -> 76.11Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 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