Research map / Bandits and online learning / Online convex optimization and games
Solving Imperfect-Information Games via Discounted Regret Minimization
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 DCFR(3/2, 0, 2) instead of CFR+ | CFR · HUNL subgames 1–4 | exploitability | Helped |
| Use DCFR(3/2, 0, 2) instead of CFR+ | CFR · 5-card Goofspiel | exploitability | No clear effect |
| Discount earlier iterations’ regrets linearly | CFR · HUNL subgames 1 and 3 | exploitability | Helped |
| Discount earlier iterations’ regrets linearly | CFR · 5-card Goofspiel | exploitability | Hurt |
| Set β = 1/2 rather than β = 0, allowing negative regrets to keep decreasing (ablation) | DCFR(3/2, 0, 2) · HUNL subgames | exploitability | No clear effect |
| Set β = 1/2 rather than β = 0, allowing negative regrets to keep decreasing (ablation) | DCFR(3/2, 0, 2) · 5-card Goofspiel | exploitability | Hurt |
| Set negative regrets to zero rather than discounting them by 1/2 (ablation) | DCFR(3/2, 0, 2) · HUNL subgames | exploitability | Mixed |
| Combine linear regret weighting with CFR+’s floor at zero (ablation) | LCFR · Games tested | exploitability | Hurt |
| Replace regret matching with NormalHedge (ablation) | DCFR(3/2, 0, 2) · HUNL subgames | exploitability | Mixed |
| Replace regret matching with NormalHedge (ablation) | LCFR · HUNL subgames | exploitability | Hurt |
| Use NormalHedge instead of regret matching (ablation) | CFR variants · HUNL subgames | iteration time: five times longer | Hurt |
| Count the latest iteration twice when computing the next strategy (ablation) | DCFR(3/2, 0, 2) · HUNL subgames | exploitability | Hurt |
Closest papers by meaning
- Lazy-CFR: fast and near-optimal regret minimization for extensive games with imperfect information (ICLR 2020 · similarity 0.95)
- Dynamic Thresholding and Pruning for Regret Minimization (AAAI 2017 · similarity 0.92)
- Double Neural Counterfactual Regret Minimization (ICLR 2020 · similarity 0.92)
- Reduced Space and Faster Convergence in Imperfect-Information Games via Pruning (ICML 2017 · similarity 0.92)
- Stable-Predictive Optimistic Counterfactual Regret Minimization (ICML 2019 · similarity 0.92)
- Accelerating Nash Equilibrium Convergence in Monte Carlo Settings Through Counterfactual Value Based Fictitious Play (NeurIPS 2024 · similarity 0.90)
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.