Research map / Learning theory and algorithms / Prediction and decision losses
Generalizing Importance Weighting to A Universal Solver for Distribution Shift Problems
Tongtong Fang, Nan Lu, Gang Niu, Masashi Sugiyama
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 GIW instead of DIW under support shift | importance weighting · MNIST (iii) | accuracy: 80.53 -> 95.16 | Improved |
| use GIW instead of DIW under support shift | importance weighting · Color-MNIST (iii) | accuracy: 39.50 -> 93.87 | Improved |
| use GIW instead of DIW under support shift | importance weighting · CIFAR-20 (iii) | accuracy: 55.78 -> 59.73 | Improved |
| use CCSA domain adaptation instead of GIW | GIW · CIFAR-20 (iii) | accuracy: 59.73 -> 60.87 | Mixed |
| add label noise 0.4 and compare DANN | DANN · MNIST LN 0.4 | accuracy: 87.24 -> 65.56 | Hurt |
| class-prior shift rho=100 for CCSA | CCSA · MNIST CS 100 | accuracy: 94.37 -> 72.67 | Hurt |
| flip OOT validation data into IT (rate 0.7) (ablation) | GIW · MNIST LN 0.2 | accuracy: 94.60 -> 86.76 | Hurt |
| flip IT validation data into OOT (rate 0.7) (ablation) | GIW · MNIST LN 0.2 | accuracy: 94.60 -> 93.55 | Hurt |
| flip IT validation data into OOT (rate 0.7) (ablation) | GIW · MNIST LN 0.4 | accuracy: 93.74 -> 90.53 | Hurt |
| vary training size n_tr from 10000 to 20000 (ablation) | GIW · MNIST LN 0.2 | accuracy | No clear effect |
| increase validation data n_v from 10 to 20 (ablation) | GIW · MNIST LN 0.2 | accuracy | Improved |
| use loss-value representation (GIW-L) instead of hidden-layer output (ablation) | GIW · MNIST LN 0.2 | accuracy | Improved |
Closest papers by meaning
- Optimizing importance weighting in the presence of sub-population shifts (ICLR 2025 · similarity 0.83)
- Distributionally Robust Policy Evaluation and Learning for Continuous Treatment with Observational Data (AAAI 2025 · similarity 0.83)
- Positive-unlabeled AUC Maximization under Covariate Shift (ICML 2025 · similarity 0.83)
- Revealing Distribution Discrepancy by Sampling Transfer in Unlabeled Data (NeurIPS 2024 · similarity 0.82)
- ReTaSA: A Nonparametric Functional Estimation Approach for Addressing Continuous Target Shift (ICLR 2024 · similarity 0.82)
- Scaling laws for learning with real and surrogate data (NeurIPS 2024 · similarity 0.82)
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.
