Research map / Learning theory and algorithms / Prediction and decision losses
Simultaneous Preference and Metric Learning from Paired Comparisons
Austin Xu, Mark Davenport
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 |
|---|---|---|---|
| learn a Mahalanobis metric instead of assuming Euclidean distance | Ideal-point estimation from paired comparisons · Synthetic data with M ≠ I | ideal-point reconstruction, ranking, and top-K identification | Improved |
| learn a flexible Mahalanobis metric when the true metric is I | Single-step ideal-point estimation · Synthetic data with M = I | ideal-point estimation performance | No clear effect |
| alternate between refining the metric and ideal point (ablation) | Single-step joint estimator · Synthetic data, 40–200 comparisons | user reconstruction error: nearly 60% | Improved |
| alternate between refining the metric and ideal point (ablation) | Single-step joint estimator · Synthetic data, fewer than 40 or more than 200 comparisons | user reconstruction error | No clear effect |
| use a Mahalanobis metric rather than an assumed Euclidean metric | Joint ideal-point and metric learner · Synthetic data as comparison count increases | fraction of top 10 items identified | Improved |
| use fewer than 20% of the available paired comparisons | Joint Mahalanobis metric and ideal-point learning · Ranked Candidates, top 22 and top 17 | fraction correctly identified: over 90% | Improved |
| use fewer than 20% of the available paired comparisons | Joint Mahalanobis metric and ideal-point learning · Ranked Candidates, top 11 | fraction correctly identified: over 80% | Improved |
Closest papers by meaning
- One for All: Simultaneous Metric and Preference Learning over Multiple Users (NeurIPS 2022 · similarity 0.84)
- Preference Modeling with Context-Dependent Salient Features (ICML 2020 · similarity 0.82)
- Contextual-Bandit Based Personalized Recommendation with Time-Varying User Interests (AAAI 2020 · similarity 0.82)
- Everyone's Preference Changes Differently: A Weighted Multi-Interest Model For Retrieval (ICML 2023 · similarity 0.82)
- Preference Diffusion for Recommendation (ICLR 2025 · similarity 0.81)
- Mult-DPO: Multinomial Direct Preference Optimization for Recommender Systems (NeurIPS 2026 · similarity 0.80)
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.
