AI Paper Atlas

Research map / Learning theory and algorithms / Submodular and game algorithms

EigenGame: PCA as a Nash Equilibrium

Ian Gemp, Brian McWilliams, Claire Vernade, Thore Graepel

ICLR 2021 · Learning theory and algorithms › Submodular and game algorithms · approach: Zero-sum game learning dynamics · 6 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
Use EigenGame instead of Frequent DirectionsPCA on small ImageNet (ResNet-50 activations) · Small ImageNet (ResNet-50 activations)runtime: 9 hours -> 2 hoursImproved
Scale EigenGame to the full ResNet-200 activation representationEigenGame · ImageNet (ResNet-200 activations)runtimeMixed
Remove the hierarchy (ablation)EigenGame · Synthetic dataprincipal components learnedHurt
Remove the hierarchy and normalization (ablation)EigenGame · Synthetic dataprincipal components learnedHurt
Use EigenGame rather than Oja's algorithm with QR orthogonalizationPCA on synthetic data · Synthetic dataprincipal component accuracyMixed
Use EigenGame with a smaller mini-batchPCA on MNIST · MNIST training setlongest streakHurt
Use EigenGame in a high batch-size regimePCA on MNIST · MNIST training setlongest streakNo clear effect
Omit projection of the update direction onto the sphere's tangent space (ablation)EigenGame update · PCAstability near equilibriumImproved

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