Summary of Prototypical Distillation and Debiased Tuning For Black-box Unsupervised Domain Adaptation, by Jian Liang and Lijun Sheng and Hongmin Liu and Ran He
Prototypical Distillation and Debiased Tuning for Black-box Unsupervised Domain Adaptation
by Jian Liang, Lijun Sheng, Hongmin Liu, Ran He
First submitted to arxiv on: 30 Dec 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: 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 Unsupervised domain adaptation is a machine learning problem where knowledge from a labeled source domain is transferred to an unlabeled target domain. Recently, researchers have explored source-free domain adaptation, which uses only a pre-trained model without the labeled source data. This paper introduces black-box domain adaptation, where the source model is accessible through an API providing predicted labels and confidence values. The authors propose a two-step framework called ProDDing, which distills a customized target model using raw predictions from the source model and prototypes derived from the target domain as teachers. In the second step, ProDDing fine-tunes the distilled model by penalizing biased logits toward certain classes. Experimental results across multiple benchmarks show that ProDDing outperforms existing black-box domain adaptation methods, particularly in hard-label black-box domain adaptation where only predicted labels are available. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine you have a special tool that can learn from one job and then apply it to another job without needing new instructions. This paper is about making this tool better by letting it learn from a similar job, but without giving it all the details of that job. The authors created a new way for this tool to adapt to the new job, which they call ProDDing. They tested this method and found that it works better than other methods in certain situations. This could be useful for things like recognizing images or understanding speech. |
Keywords
» Artificial intelligence » Domain adaptation » Logits » Machine learning » Unsupervised
