Research map / Label-efficient and robust learning / Test-time adaptation
Tempora: Characterising the Time-Contingent Utility of Online Test-Time Adaptation
Sudarshan Sreeram, Young D. Kwon, Cecilia Mascolo
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 test-time adaptation with AdaBN to standard inference under hard deadlines | Standard ResNet-50 · ImageNet-C rho=100% | discrete utility (%): 18.2 -> 30.8 | Improved |
| use gradient-based ETA instead of gradient-free AdaBN | ETA · ImageNet-C rho=100% | discrete utility (%): 30.8 -> 18.7 | Hurt |
| use SAR instead of standard inference under hard deadlines | SAR · ImageNet-C rho=100% | discrete utility (%): 18.2 -> 7.8 | Hurt |
| use ETA instead of AdaBN under continuous latency penalty | ETA · ImageNet-C T=50ms | continuous utility (%): 28.42 -> 7.22 | Hurt |
| use gradient-based Tent under amortised budget B=1s instead of standard | Tent · ImageNet-C B=1s | amortised utility (%): 18.16 -> 0.84 | Hurt |
| use SHOT-IM instead of other gradient-based methods under amortised budget | SHOT-IM · ImageNet-C B=1s | amortised utility (%): 0.87 -> 32.22 | Improved |
| relax temporal pressure from rho=100% to rho=25% for rank agreement with offline | offline ranking · ImageNet-C discrete | mean Spearman correlation: -0.19 -> 0.97 | Improved |
| choose CMF (best offline) under temporal evaluation | CMF · ImageNet-C 240 evaluations | times ranked first: 10/15 offline wins -> 29 of 240 (12.1%) temporal wins | Hurt |
| choose ETA (no offline wins) under temporal evaluation | ETA · ImageNet-C 240 evaluations | times ranked first: 0/15 offline wins -> 103 of 240 (42.9%) | Improved |
| use LAME (designed for non-i.i.d. streams) in i.i.d. episodic evaluation | LAME · ImageNet-C 240 evaluations | accuracy vs standard | Hurt |
Closest papers by meaning
- A Layer Selection Approach to Test Time Adaptation (AAAI 2025 · similarity 0.90)
- Monitoring Risks in Test-Time Adaptation (NeurIPS 2025 · similarity 0.90)
- Evaluation of Test-Time Adaptation Under Computational Time Constraints (ICML 2024 · similarity 0.89)
- Backpropagation-Free Test-Time Adaptation via Probabilistic Gaussian Alignment (NeurIPS 2025 · similarity 0.89)
- Persistent Test-time Adaptation in Recurring Testing Scenarios (NeurIPS 2024 · similarity 0.88)
- TEA: Test-time Energy Adaptation (CVPR 2024 · 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.
