Neuron· Jonathan Cornford, Roman Pogodin, Arna Ghosh, Kaiwen Sheng, Brendan A. Bicknell, Olivier Codol, Pingsheng Li, Beverley A. Clark, Guillaume Lajoie, Blake A. Richards·· 2026-09-28AI 评分62
Cornford 等：指数梯度学习产生类脑突触权重分布
Exponentiated gradient learning yields brain-like synaptic distributions
AI 导读
Cornford 等报告，指数梯度（EG）学习以乘性方式更新突触，在认知任务上表现与梯度下降相当，同时满足 Dale 定律并产生类脑的对数正态权重分布。依据摘要，EG 训练的网络在突触修剪下更稳健，且当大量输入与任务无关时学习更有效。该结果提示乘性突触更新规则可能比加性规则更贴近生物突触的统计特征，但材料仅为摘要，任务细节、网络规模与验证范围未报告。
摘要
Cornford et al. show that exponentiated gradient (EG) learning, which updates synapses multiplicatively, matches gradient descent on cognitive tasks while respecting Dale’s law and producing brain-like log-normal weight distributions. EG-trained networks better withstand synaptic pruning and learn more effectively when many inputs are task-irrelevant.
中文翻译还没有生成，先显示原文。
来源：Neuron · cell.com