Summary of Palm: Pushing Adaptive Learning Rate Mechanisms For Continual Test-time Adaptation, by Sarthak Kumar Maharana et al.
PALM: Pushing Adaptive Learning Rate Mechanisms for Continual Test-Time Adaptation
by Sarthak Kumar Maharana, Baoming Zhang, Yunhui Guo
First submitted to arxiv on: 15 Mar 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG)
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 A real-world vision model’s performance decreases when facing rapid changes in domain distributions in dynamic environments. To address this issue, a continuous test-time adaptation (CTTA) approach adjusts a pre-trained source discriminative model using unlabeled test data. A highly effective CTTA method applies layer-wise adaptive learning rates to selectively adapt pre-trained layers. However, it has limitations due to poor domain shift estimation and pseudo-label inaccuracies. This work aims to overcome these by identifying adaptation layers based on model prediction uncertainty without relying on pseudo-labels. We utilize gradient magnitude as a metric to select layers for further adaptation and evaluate their sensitivity to approximate the domain shift. Our method outperforms prior approaches in extensive image classification experiments on CIFAR-10C, CIFAR-100C, and ImageNet-C. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary When pictures change quickly, like when lighting or color changes, a computer model’s ability to recognize them drops. A new way to adjust the model to these changing conditions uses information from the test images without needing labels. The method picks which parts of the model to update based on how certain it is about its predictions. This helps improve the accuracy of recognizing pictures in different environments. We tested this approach on several datasets and found that it works better than previous methods. |
Keywords
* Artificial intelligence * Discriminative model * Image classification
