Research map / Optimization / Optimizers
EigenGame Unloaded: When playing games is better than optimizing
Ian Gemp, Brian McWilliams, Claire Vernade, Thore Graepel
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Replace α-EigenGame's update with the unbiased μ-EigenGame update | EigenGame · MNIST | longest correct eigenvector streak | Improved |
| Replace α-EigenGame's update with the unbiased μ-EigenGame update | EigenGame · MNIST | normalized subspace distance | Improved |
| Increase data parallelism and effective batch size | μ-EigenGame · Meena embeddings | eigenvector recovery performance: 16,384 samples -> 65,536 samples -> 131,072 samples | Improved |
| Increase data parallelism and effective batch size | α-EigenGame · Meena embeddings | eigenvector recovery performance: 16,384 samples -> 65,536 samples -> 131,072 samples | No clear effect |
| Use μ-EigenGame instead of α-EigenGame | EigenGame · Synthetic data with exponentially decaying spectrum | performance | Improved |
| Use μ-EigenGame instead of α-EigenGame | EigenGame · Synthetic data with linearly decaying spectrum | performance | No clear effect |
| Remove the stop-gradient operation and use the true gradient of the pseudo-utility (ablation) | μ-EigenGame · Eigenvector learning experiments | performance | Hurt |
Closest papers by meaning
- On the Robustness of Spectral Algorithms for Semirandom Stochastic Block Models (NeurIPS 2024 · similarity 0.86)
- Accelerated Regularized Learning in Finite N-Person Games (NeurIPS 2024 · similarity 0.84)
- Prediction Accuracy of Learning in Games : Follow-the-Regularized-Leader meets Heisenberg (ICML 2024 · similarity 0.84)
- Fast Last-Iterate Convergence of Learning in Games Requires Forgetful Algorithms (NeurIPS 2024 · similarity 0.84)
- From Average-Iterate to Last-Iterate Convergence in Games: A Reduction and Its Applications (NeurIPS 2025 · similarity 0.84)
- Computing Game Symmetries and Equilibria That Respect Them (AAAI 2025 · similarity 0.84)
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.
