Summary of Just Say the Name: Online Continual Learning with Category Names Only Via Data Generation, by Minhyuk Seo et al.
Just Say the Name: Online Continual Learning with Category Names Only via Data Generation
by Minhyuk Seo, Seongwon Cho, Minjae Lee, Diganta Misra, Hyeonbeom Choi, Seon Joo Kim, Jonghyun Choi
First submitted to arxiv on: 16 Mar 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 proposes a new approach to continual learning, called Generative name only Continual Learning (GenCL), which uses generative models to provide the names of new concepts without requiring extensive human supervision. The authors address limitations in previous approaches that rely on web-scraped data or human supervision by introducing two key components: HIerarchical Recurrent Prompt Generation (HIRPG) and COmplexity-NAvigating eNsembler (CONAN). These methods generate diverse prompts for the generative models, which outperform prior arts in tasks such as image recognition and multi-modal visual reasoning. The proposed approach is validated through experiments on various datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper solves a big problem with learning new things without human help. Usually, this requires lots of supervision, but that can be expensive and impractical. Instead, the authors propose a way to learn by just getting the name of what’s new, without needing examples or training data. They use special models called generative models, which can create new data that looks like real data. To make sure this new data is diverse and good quality, they introduce two new methods: one generates prompts for the models, and another picks out the best samples from multiple models. The authors show that their approach works better than others in tasks like recognizing images and understanding text. |
Keywords
* Artificial intelligence * Continual learning * Multi modal * Prompt
