Sakana AI 用 Error Diffusion 训练双流网络，MNIST 96.7%/CIFAR-10 61.7% 无反向传播

精选理由

Sakana AI 搞出了不用反向传播也能训练的网络，MNIST 和 CIFAR-10 成绩还不错，研究生物合理学习的人可以看看。

AI 摘要

Sakana AI 提出 Error Diffusion 方法，可训练符合 Dale 原则的双流兴奋/抑制网络，无需依赖反向传播。在 MNIST 基准上达到 96.7% 准确率，在 CIFAR-10 上达到 61.7% 准确率。该方法通过模数误差路由将训练规则从 MNIST 扩展到 CIFAR-10 及强化学习场景。消融实验显示，不同任务所依赖的机制具有特定性。

Sakana AI 用 Error Diffusion 训练双流网络，MNIST 96.7%/CIFAR-10 61.7% 无反向传播 — 图片来源 · marktechpost

原文 · marktechpost

Sakana AI’s Error Diffusion Trains Dale-Compliant Dual-Stream Networks, Reaching 96.7% MNIST and 61.7% CIFAR-10 Without Backpropagation

Backpropagation relies on weight transport, which biological circuits likely cannot implement. Sakana AI's Error Diffusion sidesteps that constraint, training dual-stream excitatory/inhibitory networks that obey Dale's principle. This piece breaks down how modulo error routing scales the rule from MNIST to CIFAR-10 and reinforcement learning, and what its task-dependent ablations reveal. The post Sakana AI’s Error Diffusion Trains Dale-Compliant Dual-Stream Networks, Reaching 96.7% MNIST and 61.7% CIFAR-10 Without Backpropagation appeared first on MarkTechPost .

arXiv cs.LG07-16 17:18原文

阅读原文