Research map / Label-efficient and robust learning / Test-time adaptation
Multi-Hypothesis Test-Time Adaptation to Mitigate Underspecification
Afshar Shamsi, Xiao-Yu Guo, Hamid Alinejad-Rokny, Arash Mohammadi, Damien Teney, Ehsan Abbasnejad
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 |
|---|---|---|---|
| add gradient diversification | naive ensemble of 3 particles · ImageNet-C batch size 1 (ViTBase-LN) | accuracy: 62.17 -> 63.14 | Improved |
| use three particles instead of DeYO's single model (ablation) | DeYO · ImageNet-C batch size 1 (ViTBase-LN) | accuracy: 60.57 -> 62.17 | Improved |
| add gradient diversification to DeYO-style baseline | DeYO · ImageNet-C batch size 1 (ResNet-50-GN) | accuracy: 44.40 -> 46.90 | Improved |
| add gradient diversification | DeYO · ImageNet-C label shifts (ViTBase-LN) | accuracy: 62.37 -> 63.36 | Improved |
| add gradient diversification | DeYO · ImageNet-C mixed shifts (ViTBase-LN) | top-1 accuracy: 57.05 -> 61.36 | Improved |
| add gradient diversification | DeYO · ImageNet-C mixed shifts (ResNet-50-GN) | top-1 accuracy: 31.36 -> 34.36 | Mixed |
| add flip augmentation (input diversification) to gradient diversification (ablation) | Grad diversification · ImageNet-C batch size 1 (ViTBase-LN) | accuracy: 63.08 -> 64.72 | Improved |
| add flip augmentation to naive ensemble (ablation) | Naive ensemble · ImageNet-C batch size 1 (ViTBase-LN) | accuracy: 61.88 -> 62.26 | Improved |
| use different optimizers per particle plus Grad regularizer (ablation) | Grad diversification · ImageNet-C batch size 1 (ViTBase-LN) | accuracy: 63.14 -> 63.08 | No clear effect |
| add gradient diversification to Tent | Tent · ImageNet-C batch size 1 (7 corruptions) | accuracy: 51.52 -> 53.63 | Improved |
| increase particles from N=1 to N=3 (ablation) | Grad diversification · ImageNet-C Gaussian noise | accuracy | Improved |
| dot-product gradient diversity instead of cosine or L2 difference (ablation) | Grad diversification · ImageNet-C zoom blur | accuracy | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Tent — Tent: Fully Test-Time Adaptation by Entropy Minimization (ICLR 2021) · better on 2 of 2 shared results
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
- Rethinking Entropy in Test-Time Adaptation: The Missing Piece from Energy Duality (NeurIPS 2025 · similarity 0.91)
- Backpropagation-Free Test-Time Adaptation via Probabilistic Gaussian Alignment (NeurIPS 2025 · similarity 0.90)
- A Layer Selection Approach to Test Time Adaptation (AAAI 2025 · similarity 0.90)
- Unified Entropy Optimization for Open-Set Test-Time Adaptation (CVPR 2024 · similarity 0.90)
- Evaluation of Test-Time Adaptation Under Computational Time Constraints (ICML 2024 · similarity 0.88)
- Bayesian Weight Enhancement with Steady-State Adaptation for Test-time Adaptation in Dynamic Environments (ICML 2025 · similarity 0.88)
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.
