Research map / Bandits and online learning / Online convex optimization and games
Double Neural Counterfactual 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 |
|---|---|---|---|
| Represent cumulative regret and average strategy numerator with separate neural networks | CFR for Leduc Hold'em · Leduc Hold'em | iterations to 0.06 exploitability: 10^{6} -> 200 (versus NFSP) | Helped |
| Use double neural CFR | CFR for Leduc Hold'em · Leduc Hold'em | exploitability: 0.02 after 1000 iterations | Helped |
| Use double neural CFR with mini-batch sizes 50, 100, or 500 | Double neural CFR · Leduc Hold'em | exploitability: less than 0.1 within only 1000 iterations | Helped |
| Increase mini-batch size | Robust sampling · Leduc Hold'em | exploitability: b=1, 1000, 5000, 10000 | Helped |
| Increase the number of sampled actions k | Robust sampling · Leduc Hold'em | convergence rate | Helped |
| Use robust sampling with k=1 instead of outcome sampling | MCCFR+ · One-Card Poker | convergence rate | Helped |
| Warm-start from an existing tabular CFR or RS-MCCFR+ strategy | Double neural CFR · Leduc Hold'em | exploitability: first 10 iterations | Helped |
| Use LSTM with attention rather than the other tested architectures | Double neural CFR · Leduc Hold'em | convergence rate | Helped |
| Use RSN only and remove ASN (ablation) | Double neural CFR · Leduc Hold'em | exploitability | Helped |
| Use ASN only and remove RSN (ablation) | Double neural CFR · Leduc Hold'em | exploitability | Helped |
| Use mini-batch MCCFR+ instead of mini-batch MCCFR | MCCFR · Leduc Hold'em | convergence rate | Helped |
Closest papers by meaning
- Deep (Predictive) Discounted Counterfactual Regret Minimization (AAAI 2026 · similarity 0.95)
- Solving Imperfect-Information Games via Discounted Regret Minimization (AAAI 2019 · similarity 0.92)
- An Efficient Deep Reinforcement Learning Algorithm for Solving Imperfect Information Extensive-Form Games (AAAI 2023 · similarity 0.92)
- Accelerating Nash Equilibrium Convergence in Monte Carlo Settings Through Counterfactual Value Based Fictitious Play (NeurIPS 2024 · similarity 0.89)
- Learning Not to Regret (AAAI 2024 · similarity 0.89)
- Preference-CFR: Beyond Nash Equilibrium for Better Game Strategies (ICML 2025 · similarity 0.88)
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.