Summary of A Layer Selection Approach to Test Time Adaptation, by Sabyasachi Sahoo et al.
A Layer Selection Approach to Test Time Adaptation
by Sabyasachi Sahoo, Mostafa ElAraby, Jonas Ngnawe, Yann Pequignot, Frederic Precioso, Christian Gagne
First submitted to arxiv on: 4 Apr 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper proposes a novel approach to test time adaptation (TTA), a technique used to adapt a pre-trained model to a new domain during inference. The method, called GALA, selects the most beneficial layers for update during TTA based on misaligned gradients and filters out unreliable samples with noisy gradients. This approach prevents performance degradation and focuses adaptation on trainable layers while preserving pretrained features. The proposed framework is demonstrated to improve the performance of existing TTA approaches across multiple datasets, domain shifts, model architectures, and TTA losses. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper solves a problem in machine learning called distribution shift, where a pre-trained model doesn’t work well when used with new data it hasn’t seen before. The authors found that not all parts of the model are equally good at adapting to this new data. They created a new way to choose which parts of the model to update during adaptation, called GALA. This helps the model avoid getting worse and instead focuses on improving the parts that can be changed. The paper shows that this approach works well across different types of datasets, models, and ways of adapting. |
Keywords
* Artificial intelligence * Inference * Machine learning
