Research map / Label-efficient and robust learning / Test-time adaptation
Rapid Network Adaptation: Learning to Adapt Neural Networks Using Test-Time Feedback
Teresa Yeo, Oğuzhan Fatih Kar, Zahra Sodagar, Amir Zamir
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 |
|---|---|---|---|
| replace test-time optimization with learned amortized adaptation (RNA) | depth adaptation · Taskonomy depth | relative runtime: 14.85 -> 1.56 | Improved |
| use RNA instead of TTO (episodic) | depth Taskonomy-CC · Taskonomy-CC | l1 error x100: 2.99 -> 1.68 | Improved |
| RNA vs online TTO on Hypersim cross-dataset shift | depth · Hypersim | l1 error x100: TTO online 14.17 vs RNA 16.17 | Hurt |
| train h jointly with f instead of frozen f (ablation) | RNA · Taskonomy-CC depth | l1 error x100: 1.68 -> 1.50 | Improved |
| use FiLM-f controller instead of FiLM-x or HyperNetwork-x (ablation) | RNA · Taskonomy-CC depth | l1 error: 0.039 / 0.041 -> 0.024 | Improved |
| feed concatenated RGB and sparse supervision to a single model | multi-modal baseline · Taskonomy-CC depth | error vs RNA: 42.5% worse than RNA | Hurt |
| use coarse WordNet labels with RNA | ResNet50 · ImageNet clean | error (%): TTO 24.72 -> RNA 16.72 | Improved |
| use noisy DINO k-NN coarse labels with RNA | ResNet50 · ImageNet-C | error (%): TENT 46.19 vs RNA 54.86 | Hurt |
| control for parameter count across methods (ablation) | RNA · Taskonomy-CC depth | l1 error x100: TTO 0.019 -> RNA 0.018 | Improved |
| apply RNA to DPT and ConvNext main networks (ablation) | DPT, ConvNext · Taskonomy-CC / ImageNet-C | error: DPT depth 2.61 -> 1.56; ConvNext 41.44 -> 38.04 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- TENT — Tent: Fully Test-Time Adaptation by Entropy Minimization (ICLR 2021) · better on 1 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
- Adaptive Neural Networks for Efficient Inference (ICML 2017 · similarity 0.90)
- A Layer Selection Approach to Test Time Adaptation (AAAI 2025 · similarity 0.87)
- From Kernels to Features: A Multi-Scale Adaptive Theory of Feature Learning (ICML 2025 · similarity 0.86)
- Noise Hypernetworks: Amortizing Test-Time Compute in Diffusion Models (NeurIPS 2025 · similarity 0.86)
- Towards Real-World Test-Time Adaptation: Tri-net Self-Training with Balanced Normalization (AAAI 2024 · similarity 0.86)
- Test-Time Model Adaptation with Only Forward Passes (ICML 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.
