Research map / Optimization / Nonconvex and smooth optimization
Differentiable Game Mechanics
Alistair Letcher, David Balduzzi, Sébastien Racanière, James Martens, Jakob Foerster, Karl Tuyls, 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 simultaneous gradient descent with SGA on a weak attractor plus strong rotation game | SGA · Two-player game l1=x^2/2+10xy | convergence vs learning rate: SGA converges faster with learning rates eta=0.01 and eta=0.032, and only starts overshooting for eta=0.1 | Improved |
| increase learning rate for plain gradient descent (ablation) | Simultaneous gradient descent · Two-player game l1=x^2/2+10xy | convergence: eta 0.01, 0.032, 0.1: converging extremely slowly, diverging slowly, diverging rapidly | Hurt |
| use SGA instead of optimistic mirror descent | SGA · Zero-sum bimatrix and four-player games | range of converging learning rates | Improved |
| add symplectic adjustment to simultaneous gradient descent | GAN on 16-Gaussian mixture · 2D mixture of 16 Gaussians | mode coverage | Improved |
| choose the sign of the adjustment correctly (aligned) rather than fixed (ablation) | SGA · 2D mixture of 16 Gaussians | convergence | Improved |
| add SGA to vanilla GAN with RMSProp | GAN · 75-dimensional spherical Gaussian | learned covariance eigenvalues | Improved |
| drop the first term xi from aligned consensus optimization (ablation) | Aligned consensus optimization · experiments (not shown) | performance | Hurt |
Closest papers by meaning
- The Mechanics of n-Player Differentiable Games (ICML 2018 · similarity 0.97)
- Learning to Optimize Differentiable Games (ICML 2023 · similarity 0.88)
- Discretization Drift in Two-Player Games (ICML 2021 · similarity 0.88)
- Exploiting hidden structures in non-convex games for convergence to Nash equilibrium (NeurIPS 2023 · similarity 0.87)
- Solving Neural Min-Max Games: The Role of Architecture, Initialization & Dynamics (NeurIPS 2025 · similarity 0.87)
- Generative Adversarial Equilibrium Solvers (ICLR 2024 · similarity 0.86)
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.
