Research map / Label-efficient and robust learning / Test-time adaptation
Learning Loss for Test-Time Augmentation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| select transform with learned loss predictor (k=2) instead of random selection (k=2) (ablation) | test-time augmentation · ImageNet-C corrupted set | mCE: 79.91 -> 75.00 | Helped |
| loss-predictor selection (k=2) vs greedy policy search GPS (k=2) | test-time augmentation · ImageNet-C corrupted set | mCE: 76.45 -> 75.00 | Helped |
| search policies on corrupted data (GPS-Cross, k=2) vs clean data | GPS · ImageNet clean test set | error rate (%): 23.78 -> 27.04 | Hurt |
| add horizontal flip to Ours (k=2) (ablation) | Ours · ImageNet-C corrupted set | mCE: 75.00 -> 74.00 | Helped |
| increase k from 1 to 2 selected transforms (ablation) | Ours · ImageNet-C corrupted set | mCE: 75.52 -> 75.00 | Helped |
| increase k for random selection from 1 to 4 (ablation) | Random baseline · ImageNet clean test set | error rate (%): 26.89 -> 24.29 | Helped |
Closest papers by meaning
- MEMO: Test Time Robustness via Adaptation and Augmentation (NeurIPS 2022 · similarity 0.90)
- TeSLA: Test-Time Self-Learning With Automatic Adversarial Augmentation (CVPR 2023 · similarity 0.87)
- Test-Time Training with Self-Supervision for Generalization under Distribution Shifts (ICML 2020 · similarity 0.85)
- Fixing the train-test resolution discrepancy (NeurIPS 2019 · similarity 0.85)
- Self-Bootstrapping for Versatile Test-Time Adaptation (ICML 2025 · similarity 0.85)
- Improved Self-Training for Test-Time Adaptation (CVPR 2024 · 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.