Research map / Label-efficient and robust learning / Recognition and long-tail
Tabular Learnwares Can Be Repurposed for Seemingly Irrelevant New Tasks
Peng Tan, Feifan Yang, Zhi-Hao Tan, Zhi-Hua Zhou
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 |
|---|---|---|---|
| guide feature mapping with a heuristically searched label mapping (cross-class performance matrix) (ablation) | LwRepurpose · cross-task learnware reuse performance | reuse accuracy distribution | Improved |
| add system-recommended learnwares from other tasks via LwRepurpose | user self-trained model ensemble · TabZilla binary (limited labels) | accuracy: 80.41 -> 81.12 | Improved |
| add system-recommended learnwares from other tasks via LwRepurpose | user LGBtuned ensemble · TabZilla binary (full labels) | accuracy: 88.81 -> 89.17 | Improved |
| reuse learnwares from other tasks with previous distribution-matching adaptation (MMD, JOINT, GW, FA) | model hub baselines · TabZilla binary (limited labels) | accuracy gain after normalization: MMD -1.00, GW -0.35, JOINT -0.21, FA -0.03 | Hurt |
| use a shared backbone across diverse tasks (Xtab) instead of per-task models | Xtab · TabZilla binary (limited labels) | accuracy: 52.30 vs 80.28 (TabPFNv2-basic) | Hurt |
| optimize feature mapping on marginal distribution alone, then label mapping (no label guidance) (ablation) | cross-task reuse · TabZilla binary | reuse accuracy | Hurt |
| recommend learnwares by estimated reuse performance (top-k) | LwRepurpose · TabZilla binary learnware identification | average accuracy: 66.98% average -> close to oracle top-k | Improved |
| ensemble a LightGBM learnware from an irrelevant task with a self-trained TabPFN | label-scarce target tasks · TabZilla binary (limited labels) | accuracy: gains up to 4.67 (Figure 4) | Improved |
Closest papers by meaning
- TabSTAR: A Tabular Foundation Model for Tabular Data with Text Fields (NeurIPS 2025 · similarity 0.88)
- TabM: Advancing tabular deep learning with parameter-efficient ensembling (ICLR 2025 · similarity 0.87)
- Understanding the Surprising Generalization Properties of Tabular Foundation Models (NeurIPS 2026 · similarity 0.87)
- Large Scale Transfer Learning for Tabular Data via Language Modeling (NeurIPS 2024 · similarity 0.87)
- ConTextTab: A Semantics-Aware Tabular In-Context Learner (NeurIPS 2025 · similarity 0.87)
- Retrieval & Fine-Tuning for In-Context Tabular Models (NeurIPS 2024 · similarity 0.86)
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.
