Direct binary updates
Flip a binary weight when the product of gradient momentum and the weight crosses a threshold, removing the need to maintain latent weights.
EFFICIENT AI / RESEARCH PROJECT
ICML 2025
BSO
BSO: Binary Spiking Online Optimization Algorithm
Online optimization for binary weights with less training memory.
Overview
BSO reduces the memory burden of training binary spiking networks by updating binary weights directly. Its temporal-aware variant, T-BSO, uses gradient information across time steps to adjust the update threshold.
How it works
Temporal-aware thresholds
T-BSO incorporates temporal gradient information into adaptive threshold selection.
Key results
direct weight updates within an online training framework
formal regret bounds for BSO and T-BSO
The paper reports improved optimization performance over the evaluated binary spiking training methods. Consult its experiments for accuracy and memory comparisons.
Results and experimental details ↗Citation
@inproceedings{pmlr-v267-liang25r,
title = {{BSO}: Binary Spiking Online Optimization Algorithm},
author = {Liang, Yu and Yang, Yu and Wei, Wenjie and Belatreche, Ammar and Wang, Shuai and Zhang, Malu and Yang, Yang},
booktitle = {Proceedings of the 42nd International Conference on Machine Learning},
series = {Proceedings of Machine Learning Research},
volume = {267},
pages = {37442--37455},
year = {2025},
url = {https://proceedings.mlr.press/v267/liang25r.html}
}