Research map / Learning theory and algorithms / Prediction and decision losses
Unifying and Optimizing Data Values for Selection via Sequential Decision-Making
Hongliang Chi, Qiong Wu, Zhengyi Zhou, Jonathan Light, Emily Dodwell, Yao Ma
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 |
|---|---|---|---|
| rank by existing game-theoretic data values instead of optimal sequential selection (DP) | Best existing data value per dataset · 8 OpenML datasets | relative performance drop vs optimal: 8.39% to 23.76% below optimal | Hurt |
| increase within-class feature averaging to raise substitutability (curvature proxy) (ablation) | Shapley, BetaShapley, Banzhaf · Synthetic message-passing data | mean accuracy: Shapley 0.738 -> 0.595; BetaShapley 0.706 -> 0.597; Banzhaf 0.720 -> 0.590 | Hurt |
| replace linear surrogate with bipartite coverage surrogate for utility approximation | Data selection utility model · Utility approximation (test) | MSE: 0.089 -> 0.019 | Improved |
| use bipartite graph greedy selection instead of game-theoretic values | OpenML data selection · Budget 500, average over 8 datasets | average accuracy: 0.821 -> 0.828 vs BetaShap | Improved |
| larger selection budget (500) narrows gap between methods | Bipartite vs game-theoretic values · Budget 500 | accuracy | No clear effect |
| use coverage-prioritized BipCov selection instead of similarity-only RDS+ | Llama-3.1-8B LoRA fine-tuning · DATE-LM (MMLU/GSM8K/BBH average) | average accuracy (%): 62.88 -> 63.89 | Improved |
| BipCov versus random selection on DATE-LM | Llama-3.1-8B LoRA fine-tuning · DATE-LM average | average accuracy (%): 62.82 -> 63.89 | Improved |
| ranking by semi-values on highly redundant datasets | Shapley, Banzhaf, Beta Shapley · Highly redundant data | selection optimality guarantee: U(S_k) >= (1-c)^2 U(OPT_k) | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- DataOob — Data-OOB: Out-of-bag Estimate as a Simple and Efficient Data Value (ICML 2023) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Compute-Constrained Data Selection (ICLR 2025 · similarity 0.87)
- Rethinking Data Shapley for Data Selection Tasks: Misleads and Merits (ICML 2024 · similarity 0.87)
- What Data Enables Optimal Decisions? An Exact Characterization for Linear Optimization (NeurIPS 2025 · similarity 0.87)
- Real-Time Selection Under General Constraints via Predictive Inference (NeurIPS 2024 · similarity 0.86)
- Adaptive Data Optimization: Dynamic Sample Selection with Scaling Laws (ICLR 2025 · similarity 0.85)
- Computational Budget Should Be Considered in Data Selection (NeurIPS 2025 · 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.
