Research map / Optimization / Optimizers
Recursive Reasoning in Minimax Games: A Level $k$ Gradient Play Method
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 |
|---|---|---|---|
| use Lv.6 Adam instead of standard Adam | SN-GAN · CIFAR-10 | FID: 21.70 ± 0.21 -> 10.17± 0.16 | Helped |
| use Lv.6 Adam instead of Adam | SN-GAN · STL-10 | FID: 30.25± 0.26 -> 25.43±0.18 | Helped |
| increase reasoning steps from 2 to 6 (ablation) | Lv.k Adam with non-saturated loss · CIFAR-10 | FID: 11.33± 0.18 -> 10.93± 0.24 | Helped |
| increase reasoning steps from 2 to 6 (ablation) | Lv.k Adam with hinge loss · CIFAR-10 | FID: 10.68 ± 0.20 -> 10.17 ± 0.16 | Helped |
| replace non-saturated loss with hinge loss (ablation) | Lv.2 Adam · CIFAR-10 | FID: 11.33± 0.18 -> 10.68 ± 0.20 | Helped |
| replace non-saturated loss with hinge loss (ablation) | Lv.4 Adam · CIFAR-10 | FID: 11.62± 0.25 -> 10.33 ± 0.22 | Helped |
| replace non-saturated loss with hinge loss (ablation) | Lv.6 Adam · CIFAR-10 | FID: 10.93± 0.24 -> 10.17 ± 0.16 | Helped |
| increase reasoning steps from 4 to 6 (ablation) | Lv.k Adam with non-saturated loss · CIFAR-10 | FID: 11.62± 0.25 -> 10.93± 0.24 | Helped |
| increase reasoning steps from 4 to 6 (ablation) | Lv.k Adam with hinge loss · CIFAR-10 | FID: 10.33 ± 0.22 -> 10.17 ± 0.16 | Helped |
| increase reasoning steps (ablation) | Lv.k Adam · 8-Gaussians | average difference between consecutive reasoning states: 1.04\times 10^{-1} -> 1.68\times 10^{-4} | Helped |
Closest papers by meaning
- Finding Mixed Nash Equilibria of Generative Adversarial Networks (ICML 2019 · similarity 0.90)
- The Mechanics of n-Player Differentiable Games (ICML 2018 · similarity 0.88)
- Fast and Knowledge-Free Deep Learning for General Game Playing (Student Abstract) (AAAI 2024 · similarity 0.87)
- Vision-Zero: Scalable VLM Self-Improvement via Strategic Gamified Self-Play (ICLR 2026 · 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.