Research map / Label-efficient and robust learning / Test-time adaptation
MoETTA: Test-Time Adaptation Under Mixed Distribution Shifts with MoE-LayerNorm
Xiao Fan, Jingyan Jiang, Zhaoru Chen, Fanding Huang, Xiao Chen, Qinting Jiang, Bowen Zhang, Xing Tang, Zhi Wang
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 load balancing loss (ablation) | MoETTA · Average over 3 settings | accuracy (%): 21.28 -> 66.20 | Improved |
| add sample-wise router (ablation) | MoETTA · Average over 3 settings | accuracy (%): 27.42 -> 66.20 | Improved |
| add MoE-LayerNorm module (ablation) | MoETTA · Average over 3 settings | accuracy (%): 22.42 -> 66.20 | Improved |
| add layer-wise router (ablation) | MoETTA · Average over 3 settings | accuracy (%): 19.89 -> 66.20 | Improved |
| add gradient flow from entropy loss to router (ablation) | MoETTA · Average over 3 settings | accuracy (%): 63.96 -> 66.20 | Improved |
| add entropy re-weighting (ablation) | MoETTA · Average over 3 settings | accuracy (%): 61.72 -> 66.20 | Improved |
| add sample selection (ablation) | MoETTA · Average over 3 settings | accuracy (%): 62.89 -> 66.20 | Improved |
| add sample selection (ablation) | MoETTA · Potpourri+ | accuracy (%): 57.61 -> 66.21 | Improved |
| add sample selection (ablation) | MoETTA · Classical mixed shifts | accuracy (%): 67.04 -> 67.25 | No clear effect |
| replace best baseline MGTTA with MoETTA | MoETTA vs MGTTA · Potpourri (ViT-B/16) | accuracy (%): 62.98 -> 65.12 | Improved |
| scale backbone to ViT-L/16, MoETTA vs EATA | MoETTA · Potpourri (ViT-L/16) | accuracy (%): 64.6 -> 64.4 | Hurt |
| use MoETTA instead of EATA on ViT-L/16 | MoETTA vs EATA · Classical mixed shifts (ViT-L/16) | accuracy (%): 66.6 -> 68.8 | Improved |
Closest papers by meaning
- A Layer Selection Approach to Test Time Adaptation (AAAI 2025 · similarity 0.91)
- ReservoirTTA: Prolonged Test-time Adaptation for Evolving and Recurring Domains (NeurIPS 2025 · similarity 0.91)
- AETTA: Label-Free Accuracy Estimation for Test-Time Adaptation (CVPR 2024 · similarity 0.89)
- Backpropagation-Free Test-Time Adaptation via Probabilistic Gaussian Alignment (NeurIPS 2025 · similarity 0.89)
- SNAP: Low-Latency Test-Time Adaptation with Sparse Updates (NeurIPS 2025 · similarity 0.89)
- Efficient Test-time Adaptation through Candidate Verification and Divergence Shifts (NeurIPS 2026 · 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.
