Understand sign flipping
Analyze how repeated changes in binary weight signs interfere with learning.
EFFICIENT AI / RESEARCH PROJECT
AAAI 2025Oral
AGMM
Towards Accurate Binary Spiking Neural Networks: Learning with Adaptive Gradient Modulation Mechanism
More stable learning for accurate binary spiking networks.
Overview
Binary spiking networks combine event-driven computation with compact synaptic weights. AGMM investigates frequent weight sign flipping as a training bottleneck and adapts gradient magnitudes to improve learning stability.
How it works
Adapt gradient magnitudes
Modulate gradients during training to reduce unnecessary weight sign changes and improve convergence.
Key results
reported relative to the binary spiking baselines evaluated in the paper
evaluation on both conventional image and event-based datasets
The experiments report improved binary-network accuracy and a narrower gap to full-precision counterparts. Detailed settings and comparisons are available in the paper.
Results and experimental details ↗Citation
@inproceedings{liang2025towards,
title = {Towards Accurate Binary Spiking Neural Networks: Learning with Adaptive Gradient Modulation Mechanism},
author = {Liang, Yu and Wei, Wenjie and Belatreche, Ammar and Cao, Honglin and Zhou, Zijian and Wang, Shuai and Zhang, Malu and Yang, Yang},
booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence},
volume = {39},
number = {2},
pages = {1402--1410},
year = {2025},
doi = {10.1609/aaai.v39i2.32130}
}