WeWorkBuddy

Manifold Approximation leads to Robust Kernel Alignment

arXiv:2510.22953v3 Announce Type: replace-cross Abstract: Centered kernel alignment (CKA) is a popular metric for comparing representations, determining equivalence of networks, and conducting neuroscience research. However, CKA does not account for the underlying manifold and relies on numerous heuristics that cause it to behave differently at different scales of data. In this work, we propose Manifold-approximated Kernel Alignment (MKA), which incorporates manifold geometry (i.e., local neighb

阅读原始来源
来源参考:Manifold Approximation leads to Robust Kernel Alignment。暂未取得截图。
来源记录 · 预览待采集 · arxiv.org打开来源查看完整网页。

来源记录

项目分析

暂无五维分析,可先查看上方来源资料。