Summary of Hilo: a Learning Framework For Generalized Category Discovery Robust to Domain Shifts, by Hongjun Wang et al.
HiLo: A Learning Framework for Generalized Category Discovery Robust to Domain Shifts
by Hongjun Wang, Sagar Vaze, Kai Han
First submitted to arxiv on: 8 Aug 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
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 introduces Generalized Category Discovery (GCD), a challenging task where models categorize unlabelled instances from different domains, not just labelled categories. The authors propose HiLo' networks that extract high-level semantic and low-level domain features to minimize mutual information between representations. They extend their method with a domain augmentation technique and curriculum learning approach. A benchmark is constructed using corrupted fine-grained datasets and large-scale evaluation on DomainNet with real-world domain shifts, re-implementing GCD baselines. The proposedHiLo’ networks outperform state-of-the-art (SoTA) category discovery models by a significant margin. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper solves a problem called Generalized Category Discovery. Imagine you have some pictures and some of them are labelled, but the rest are not. You need to put all the unlabelled pictures into groups based on what’s in them. The challenge is that these unlabelled pictures might come from different places (like cats, dogs, and cars) or new categories we’ve never seen before. The authors propose a new way to solve this problem called `HiLo’ networks. They also test their method on real-world data and show it works better than other methods. |
Keywords
» Artificial intelligence » Curriculum learning
