已入深夜,您辛苦了!由于当前在线用户较少,发布求助请尽量完整地填写文献信息,科研通机器人24小时在线,伴您度过漫漫科研夜!祝你早点完成任务,早点休息,好梦!

Shift-Curvature, SGD, and Generalization

作者
Arwen V. Bradley,Carlos Alberto Gomez-Uribe,Manish Reddy Vuyyuru
出处
期刊:Cornell University - arXiv [Cornell University]
标识
DOI:10.48550/arxiv.2108.09507
摘要

A longstanding debate surrounds the related hypotheses that low-curvature minima generalize better, and that SGD discourages curvature. We offer a more complete and nuanced view in support of both. First, we show that curvature harms test performance through two new mechanisms, the shift-curvature and bias-curvature, in addition to a known parameter-covariance mechanism. The three curvature-mediated contributions to test performance are reparametrization-invariant although curvature is not. The shift in the shift-curvature is the line connecting train and test local minima, which differ due to dataset sampling or distribution shift. Although the shift is unknown at training time, the shift-curvature can still be mitigated by minimizing overall curvature. Second, we derive a new, explicit SGD steady-state distribution showing that SGD optimizes an effective potential related to but different from train loss, and that SGD noise mediates a trade-off between deep versus low-curvature regions of this effective potential. Third, combining our test performance analysis with the SGD steady state shows that for small SGD noise, the shift-curvature may be the most significant of the three mechanisms. Our experiments confirm the impact of shift-curvature on test loss, and further explore the relationship between SGD noise and curvature.

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
刚刚
LA完成签到,获得积分10
1秒前
1秒前
LMX发布了新的文献求助10
1秒前
飞飞飞发布了新的文献求助10
2秒前
科研通AI6.4应助limeng采纳,获得10
3秒前
坚强煜城发布了新的文献求助10
3秒前
sss发布了新的文献求助10
4秒前
王特发布了新的文献求助10
6秒前
斯文败类应助青屿采纳,获得10
6秒前
7秒前
思源应助坚强煜城采纳,获得10
8秒前
10秒前
Gtingting发布了新的文献求助10
12秒前
鸽子完成签到,获得积分10
13秒前
14秒前
liuwei发布了新的文献求助10
15秒前
Cynosureo921完成签到,获得积分20
16秒前
16秒前
cincrady完成签到,获得积分10
18秒前
哈哈哈发布了新的文献求助10
19秒前
nanshaokuingh发布了新的文献求助10
19秒前
Jane发布了新的文献求助10
19秒前
埃维发布了新的文献求助10
20秒前
kexing发布了新的文献求助10
22秒前
25秒前
隐形曼青应助明年采纳,获得10
27秒前
27秒前
28秒前
30秒前
30秒前
31秒前
31秒前
英俊的铭应助科研通管家采纳,获得10
31秒前
李健应助科研通管家采纳,获得10
32秒前
在水一方应助科研通管家采纳,获得10
32秒前
Su发布了新的文献求助10
32秒前
废废废完成签到,获得积分10
32秒前
英姑应助科研通管家采纳,获得10
32秒前
32秒前
高分求助中
Principles of Economics, 11th Edition 10000
University Physics with Modern Physics, 16th edition 10000
(应助此贴封号)【重要!!请各用户(尤其是新用户)详细阅读】【科研通的精品贴汇总】 10000
Molecular Mechanisms of Photosynthesis, 4th Edition 1000
Organic Reactions, Volume 116 1000
Matrix Methods in Data Mining and Pattern Recognition 510
Reading and Understanding Health Research 500
热门求助领域 (近24小时)
化学 材料科学 医学 生物 纳米技术 工程类 有机化学 化学工程 生物化学 计算机科学 内科学 物理 复合材料 催化作用 细胞生物学 无机化学 光电子学 物理化学 电极 基因
热门帖子
关注 科研通微信公众号,转发送积分 7252185
求助须知:如何正确求助?哪些是违规求助? 8874579
关于积分的说明 18732879
捐赠科研通 6932240
什么是DOI,文献DOI怎么找? 3199651
关于科研通互助平台的介绍 2374362
邀请新用户注册赠送积分活动 2174251