Research map / Bandits and online learning / Online convex optimization and games
Learning to Identify Top Elo Ratings: A Dueling Bandits Approach
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 |
|---|---|---|---|
| Select maximum-uncertainty pairs from a UCB-maintained set of promising players instead of random or uniform matchups | MaxIn-Elo · Real-world transitive games | cumulative regret | Helped |
| Select maximum-uncertainty pairs from a UCB-maintained set of promising players | MaxIn-mElo · Six real-world intransitive games | cumulative regret | Helped |
| Add a cyclic interaction vector to the Elo prediction model | MaxIn-mElo · Intransitive games | cumulative regret | Helped |
| Replace repeated historical-data MLE with online SGD updates | MaxIn-Elo · Top-player identification | time complexity: O(nT^2+n^2T) -> O(n^2T) | Helped |
| Replace repeated historical-data MLE with online SGD updates | MaxIn-Elo · Top-player identification | memory: O(nT) -> O(n^2) | Helped |
| Set the exploration balance parameter to a smaller value (ablation) | MaxIn-Elo · Top-1 identification on transitive games | top-1 performance | Mixed |
| Increase the exploration balance parameter (ablation) | MaxIn-Elo · Top-k identification on transitive games | top-k performance | Mixed |
| Use γ=0.4 rather than the best-performing reported setting (ablation) | MaxIn-Elo · Top-1 identification on Elo game and its variants, and Triangular game | top-1 identification | Hurt |
| Increase the mElo cyclic-vector dimension up to 8 (ablation) | MaxIn-mElo · Intransitive games | top-1 identification | Helped |
| Increase the mElo cyclic-vector dimension to 16 (ablation) | MaxIn-mElo · Intransitive games | top-1 identification | Hurt |
| Use batch size τ=1 (ablation) | MaxIn-Elo and MaxIn-mElo · Elo game and Kuhn-poker | top-1 identification | Hurt |
| Use a batch size around 0.5n to 1.0n (ablation) | MaxIn-Elo · Elo game | top-1 identification: 0.5n~1.0n | Helped |
Closest papers by meaning
- An Analysis of Elo Rating Systems via Markov Chains (NeurIPS 2024 · similarity 0.89)
- Active Ranking and Matchmaking, with Perfect Matchings (ICML 2024 · similarity 0.87)
- Neural Dueling Bandits: Preference-Based Optimization with Human Feedback (ICLR 2025 · similarity 0.86)
- EnMatch: Matchmaking for Better Player Engagement via Neural Combinatorial Optimization (AAAI 2024 · similarity 0.85)
- Eliciting Kemeny Rankings (AAAI 2024 · similarity 0.85)
- From Average-Iterate to Last-Iterate Convergence in Games: A Reduction and Its Applications (NeurIPS 2025 · similarity 0.84)
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.