Summary of A Fresh Look at Generalized Category Discovery Through Non-negative Matrix Factorization, by Zhong Ji et al.
A Fresh Look at Generalized Category Discovery through Non-negative Matrix Factorization
by Zhong Ji, Shuo Yang, Jingren Liu, Yanwei Pang, Jungong Han
First submitted to arxiv on: 29 Oct 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 proposes a novel framework for Generalized Category Discovery (GCD), which aims to classify both base and novel images using labeled base data. The existing approaches fall short in optimizing the co-occurrence matrix based on cosine similarity, leading to inadequate sparsity in base and novel domains. To address this, the authors introduce Non-Negative Generalized Category Discovery (NN-GCD), a framework that employs Symmetric Non-negative Matrix Factorization (SNMF) to prove the equivalence of optimal K-means with optimal SNMF. The optimization problem is reframed as non-negative contrastive learning (NCL) optimization, and a GELU activation function and NMF NCE loss are proposed to ensure convergence. A hybrid sparse regularization approach is also introduced to impose sparsity constraints on the co-occurrence matrix. Experimental results show that NN-GCD outperforms state-of-the-art methods on GCD benchmarks, achieving an average accuracy of 66.1% on the Semantic Shift Benchmark, a 4.7% improvement over prior counterparts. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine you have some pictures and you want to group them into categories based on what’s in the pictures. But it’s not just about grouping similar pictures together – you also need to make sure that new, unseen pictures can be put into the right category too. The problem is that current methods don’t do this very well, especially when there are a lot of different things going on in each picture. To solve this problem, researchers have developed a new way of organizing and categorizing images called Non-Negative Generalized Category Discovery (NN-GCD). This method uses some fancy math to make sure that the categories are correct and that new pictures can be added without messing up the system. The results show that NN-GCD works much better than previous methods, with an accuracy rate of 66.1% on a benchmark test. |
Keywords
» Artificial intelligence » Cosine similarity » K means » Optimization » Regularization
