Research map / Learning theory and algorithms / Prediction and decision losses
Preference Elicitation for Step-Wise Explanations in Logic Puzzles
Marco Foschini, Marianne Defresne, Emilio Gamba, Bart Bogaerts, Tias Guns
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 MACHOP (non-domination + local normalization + UCB weighting) instead of Choice Perceptron | Choice Perceptron · Sudoku (simulated users) | relative regret: 2.0 -> 0.4 | Improved |
| use MACHOP instead of Choice Perceptron | Choice Perceptron · Logic-Grid puzzles (simulated users) | relative regret: 3.8 -> 0.9 | Improved |
| add non-domination disjunctive constraint to query generation (ablation) | Choice Perceptron · simulated users | relative regret | Improved |
| use local or cumulative normalization instead of nadir-point normalization (ablation) | preference learning · simulated users | relative regret | Improved |
| add UCB weights to diversification (ablation) | diversification metric · Sudoku and Logic-Grid (simulated users) | relative regret: about 40% lower regret | Improved |
| weight diversification by learned weights (ablation) | diversification metric · Logic-Grid puzzles | relative regret | Mixed |
| precompute facts with SES (offline) instead of online selection (ablation) | MACHOP · Logic-Grid puzzles | query generation time (s): 49.2 -> 8.3 | Improved |
| precompute facts with random order (offline) (ablation) | MACHOP · Sudoku | query generation time (s): 35.6 -> 44.2 | Hurt |
| precompute facts offline (SES) (ablation) | MACHOP · Logic-Grid puzzles | relative regret: 0.9 -> 1.4 | Hurt |
| train with MACHOP vs Choice Perceptron, 50 queries | Choice Perceptron · Sudoku user study | % preferred over SES: 38.8 -> 72.6 | Improved |
Closest papers by meaning
- HardcoreLogic: Challenging Large Reasoning Models with Long-tail Logic Puzzle Games (ICLR 2026 · similarity 0.86)
- Advancing LLM Reasoning Generalists with Preference Trees (ICLR 2025 · similarity 0.85)
- Preference Optimization for Reasoning with Pseudo Feedback (ICLR 2025 · similarity 0.85)
- Learning How to Cube (NeurIPS 2026 · similarity 0.85)
- Pruning Long Chain-of-Thought of Large Reasoning Models via Small-Scale Preference Optimization (ICLR 2026 · similarity 0.84)
- Causal language modeling can elicit search and reasoning capabilities on logic puzzles (NeurIPS 2024 · 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.
