Summary of Label Distribution Shift-aware Prediction Refinement For Test-time Adaptation, by Minguk Jang et al.
Label Distribution Shift-Aware Prediction Refinement for Test-Time Adaptation
by Minguk Jang, Hye Won Chung
First submitted to arxiv on: 20 Nov 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 This paper introduces a novel approach to test-time adaptation (TTA) called Label Distribution shift-Aware prediction Refinement for Test-time adaptation (DART). The authors first analyze existing TTA methods and identify that they often suffer from performance drops when facing additional class distribution shifts. To address this issue, DART trains a prediction refinement module during an intermediate time by exposing it to diverse class distributions using the training dataset. This module is then used during test time to detect and correct class distribution shifts, leading to improved pseudo-label accuracy for test data. The authors demonstrate the effectiveness of DART on various benchmarks, including CIFAR-10C, PACS, OfficeHome, and ImageNet, achieving 5-18% gains in accuracy under label distribution shifts without any performance degradation when there is no shift. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps fix a big problem with machine learning models. When these models are used on new data that’s different from the training data, they can make mistakes. The authors create a new way to help these models work better in this situation. They call it DART (Label Distribution shift-Aware prediction Refinement for Test-time adaptation). It works by teaching the model to recognize when it’s making mistakes and correct them. This helps the model be more accurate on new data. The authors test their method on many different datasets and show that it makes a big difference. |
Keywords
* Artificial intelligence * Machine learning
