SVCCA: Singular Vector Canonical Correlation Analysis for Deep Learning Dynamics and Interpretability
SVCCA: Singular Vector Canonical Correlation Analysis for Deep Learning Dynamics and Interpretability
M. Raghu,J. Gilmer,J. Yosinski,Jascha Narain Sohl-Dickstein
2017 · DBLP: conf/nips/RaghuGYS17
Neural Information Processing Systems · 893 Citations
TLDR
This work proposes a new technique, Singular Vector Canonical Correlation Analysis (SVCCA), a tool for quickly comparing two representations in a way that is both invariant to affine transform and fast to compute.