Research map / Label-efficient and robust learning / Recognition and long-tail
Detecting Individual Decision-Making Style: Exploring Behavioral Stylometry in Chess
Reid McIlroy-Young, Russell Wang, Siddhartha Sen, Jon Kleinberg, Ashton Anderson
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 the transformer-based behavioral stylometry model instead of the opening baseline | Lichess players with 10K+ games, unseen targets, k=15 · 10K+ Lichess, candidate pool 2844 | P@1: 0.244 -> 0.860 | Improved |
| Include opening moves at inference (k=0) rather than excluding them (k=15) | Behavioral stylometry model · 10K+ Lichess, unseen targets, candidate pool 2844 | P@1: 0.860 -> 0.979 | Improved |
| Replace the personalized move-matching approach with the shared behavioral stylometry model | McIlroy-Young et al. task · 400-player candidate pool, k=15 | P@1: 0.552 -> 0.953 | Improved |
| Apply the amateur-trained model to high-ranked players | Behavioral stylometry model · High-ranked unseen players, candidate pool 2157, k=15 | P@1: 0.039 -> 0.308 | Improved |
| Add Lichess amateurs to the candidate pool of high-ranked players | Behavioral stylometry model · High-ranked unseen players, k=15 | P@1: 0.308 -> 0.301 | No clear effect |
| Use the transformer instead of an LSTM to aggregate move features (ablation) | Behavioral stylometry model · 10K--40K+ Lichess unseen, k=15, candidate pool 2844 | P@1: 0.7803 -> 0.8599 | Improved |
| Use all moves instead of a five-move opening window (ablation) | Opening baseline · 10K+ Lichess unseen, k=15 | P@1: 0.234 -> 0.277 | Improved |
| Train with 32-move windows instead of 10-move windows (ablation) | Behavioral stylometry model · Training validation loss | validation loss | Improved |
| Train only on windows starting at move 15 or later (ablation) | Behavioral stylometry model · Candidate pool and evaluation pool of 232 players | accuracy: 0.9927 -> 0.390 | Hurt |
| Replace the 512-dimensional embedding with a 128-dimensional embedding (ablation) | Behavioral stylometry model · Training validation loss | validation loss | Hurt |
| Replace GE2E loss with ArcFace (ablation) | Behavioral stylometry model · Evaluated settings | convergence | Hurt |
| Train on only 400 players rather than the full training population (ablation) | Behavioral stylometry model · McIlroy-Young et al. task, k=15 | P@1: 0.953 -> 0.825 | Hurt |
Closest papers by meaning
- Maia-2: A Unified Model for Human-AI Alignment in Chess (NeurIPS 2024 · similarity 0.90)
- Human-Aligned Chess With a Bit of Search (ICLR 2025 · similarity 0.89)
- Amortized Planning with Large-Scale Transformers: A Case Study on Chess (NeurIPS 2024 · similarity 0.86)
- Designing Skill-Compatible AI: Methodologies and Frameworks in Chess (ICLR 2024 · similarity 0.86)
- The Algorithm Is Not the Behavior: Learned Priors Override Look-Ahead in a Chess-Playing Neural Network (NeurIPS 2026 · similarity 0.85)
- Who’s Gaming the System? A Causally-Motivated Approach for Detecting Strategic Adaptation (NeurIPS 2024 · similarity 0.85)
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.
