ICLR

Leveraging Flatness to Improve Information-Theoretic Generalization Bounds for SGD

利用平坦性改进SGD的信息论通用性界限

Ze Peng, Jian Zhang, Yisen Wang, Lei Qi, Yinghuan Shi, Yang Gao

AI总结 本文提出了一种更注重SGD平坦性偏置的信息论通用性界限，通过'omniscient trajectory'技术改进了现有界限的紧致性，并揭示了记忆与通用性的权衡问题。

Comments Published as a conference paper at ICLR 2025

Journal ref The Thirteenth International Conference on Learning Representations, ICLR 2025