AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Reliability of CKA as a Similarity Measure in Deep Learning

MohammadReza Davari, Stefan Horoi, Amine Natik, Guillaume Lajoie, Guy Wolf, Eugene Belilovsky

ICLR 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Contrastive representation theory · 3 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Translate a subset of representations from one cube in a random directionArtificial representations · Artificial subset translationlinear CKA: drops below 0.2Hurt
Translate a subset of representations from one cube in a random directionArtificial representations · Artificial subset translationRBF CKA: drops below 0.2Hurt
Translate one class of representations in a direction that preserves the separating hyperplanes9-layer CNN representations · CIFAR10 training representationslinear CKA: quickly drop to 0Hurt
Translate a single representation as an outlier9-layer CNN representations · CIFAR10 training representationslinear CKAHurt
Translate a subset while preserving the separating hyperplanes9-layer CNN representations · CIFAR10 training representationsclassification accuracy: >90% accuracy on both setsNo clear effect
Optimize the CKA map while preserving outputs with distillationTrained CNN · CIFAR10model performanceNo clear effect
Optimize the CKA map to increase similarity between the first and last layersTrained CNN · CIFAR10model performanceNo clear effect
Optimize the CKA map to increase similarity among all layersTrained CNN · CIFAR10model performanceNo clear effect
Optimize the CKA map to decrease similarity among all layersTrained CNN · CIFAR10model performanceNo clear effect
Optimize wide-network CKA maps to match a thin network's target mapResNet-34 · CIFAR10model performanceNo clear effect
Replace distillation loss with cross-entropy loss (ablation)CKA map optimization · CIFAR10CKA map manipulationNo clear effect
Use CKA-map optimization on a Vision TransformerVision Transformer · CIFAR10model performanceNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy