Research map / Optimization / Nonconvex and smooth optimization
Discretization Drift in Two-Player Games
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Cancel the DD interaction terms via explicit gradient-norm regularization (c1=αh/4, c2=λh/4) in simultaneous GD (ablation) | zero-sum GAN trained with simultaneous SGD · CIFAR10 GAN | Inception Score | Helped |
| Use alternating instead of simultaneous updates (ablation) | zero-sum GAN trained with SGD · CIFAR10 GAN | Inception Score | Helped |
| Strengthen self terms with coefficients proportional to drift (s1=αh/4, s2=λh/4) in simultaneous GD (ablation) | zero-sum GAN trained with simultaneous SGD · CIFAR10 GAN | Inception Score | Mixed |
| Cancel only the discriminator interaction term in alternating updates (ablation) | zero-sum GAN trained with alternating SGD · CIFAR10 GAN | training variance / sensitivity across hyperparameters | Helped |
| Cancel both interaction terms in alternating updates (ablation) | zero-sum GAN trained with alternating SGD · CIFAR10 GAN | Inception Score | Hurt |
| Cancel interaction terms for individual learning-rate choices (not across sweep) (ablation) | zero-sum GAN trained with simultaneous SGD · CIFAR10 GAN | Inception Score | No clear effect |
| Use learning rate ratio alpha/lambda > 0.5 with alternating updates (ablation) | zero-sum GAN trained with alternating SGD · CIFAR10 GAN | Inception Score | Helped |
| Use the non-saturating generator loss instead of the zero-sum loss; switch simultaneous to alternating updates | non-zero-sum GAN · CIFAR10 GAN | Inception Score | No clear effect |
| Use Runge-Kutta 4 instead of simultaneous Euler (removing drift) (ablation) | zero-sum GAN trained with SGD · CIFAR10 GAN | Inception Score | Helped |
Closest papers by meaning
- The Mechanics of n-Player Differentiable Games (ICML 2018 · similarity 0.89)
- Extra-gradient with player sampling for faster convergence in n-player games (ICML 2020 · similarity 0.89)
- A mean-field analysis of two-player zero-sum games (NeurIPS 2020 · similarity 0.88)
- Many Paths to Equilibrium: GANs Do Not Need to Decrease a Divergence At Every Step (ICLR 2018 · similarity 0.88)
- Finding Mixed Nash Equilibria of Generative Adversarial Networks (ICML 2019 · similarity 0.87)
- Solving Neural Min-Max Games: The Role of Architecture, Initialization & Dynamics (NeurIPS 2025 · 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.