Research map / Label-efficient and robust learning / Test-time adaptation
A Probabilistic Framework for Lifelong Test-Time Adaptation
Dhanajit Brahma, Piyush Rai
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 the approximate source-model posterior regularizer and PETAL adaptation framework rather than CoTTA | Lifelong test-time adaptation · CIFAR10-to-CIFAR10C, severity 5 | classification error (%): 16.29 (0.02) -> 15.95 (0.04) | Improved |
| Add the approximate source-model posterior regularizer and PETAL adaptation framework rather than CoTTA | Lifelong test-time adaptation · CIFAR100-to-CIFAR100C, severity 5 | classification error (%): 32.48 (0.02) -> 31.46 (0.04) | Improved |
| Use FIM-driven parameter restoration instead of stochastic restoration (ablation) | PETAL · CIFAR10-to-CIFAR10C, severity 5 | classification error (%): 16.16 (0.02) -> 15.95 (0.04) | Improved |
| Use FIM-driven parameter restoration instead of stochastic restoration (ablation) | PETAL · CIFAR100-to-CIFAR100C, severity 5 | classification error (%): 31.62 (0.04) -> 31.46 (0.04) | Improved |
| Remove parameter restoration (ablation) | PETAL · CIFAR10-to-CIFAR10C | classification error (%): 15.95±0.04 -> 16.37±0.03 | Hurt |
| Use stochastic restoration at rate 0.03 instead of FIM-based restoration (ablation) | PETAL · CIFAR10-to-CIFAR10C | classification error (%): 15.95±0.04 -> 16.51±0.02 | Hurt |
| Add PETAL's posterior regularization and adaptation rather than CoTTA | Lifelong test-time adaptation · CIFAR100-to-CIFAR100C, severity 5 | NLL: 1.2767 (0.0006) -> 1.2206 (0.0011) | Improved |
| Add PETAL's posterior regularization and adaptation rather than CoTTA | Lifelong test-time adaptation · CIFAR100-to-CIFAR100C, severity 5 | Brier: 0.4430 (0.0003) -> 0.4317 (0.0003) | Improved |
| Add PETAL adaptation rather than CoTTA | Lifelong test-time adaptation · ImageNet-to-ImageNetC, severity 5 | Error (%): 63.18 -> 62.71 | Improved |
| Add PETAL adaptation rather than CoTTA | Lifelong test-time adaptation · ImageNet-to-ImageNet3DCC, severity 5 | Error (%): 59.91 -> 59.61 | Improved |
| Add PETAL adaptation rather than CoTTA | Lifelong test-time adaptation · CIFAR10-to-CIFAR10C, gradually changing severity | Error (%): 10.40 ± 0.22 -> 10.11 ± 0.23 | Improved |
Closest papers by meaning
- Persistent Test-time Adaptation in Recurring Testing Scenarios (NeurIPS 2024 · similarity 0.92)
- A Layer Selection Approach to Test Time Adaptation (AAAI 2025 · similarity 0.91)
- Evaluation of Test-Time Adaptation Under Computational Time Constraints (ICML 2024 · similarity 0.90)
- Backpropagation-Free Test-Time Adaptation via Probabilistic Gaussian Alignment (NeurIPS 2025 · similarity 0.90)
- Towards Real-World Test-Time Adaptation: Tri-net Self-Training with Balanced Normalization (AAAI 2024 · similarity 0.90)
- ReservoirTTA: Prolonged Test-time Adaptation for Evolving and Recurring Domains (NeurIPS 2025 · similarity 0.89)
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.
