Summary of Future-proofing Class-incremental Learning, by Quentin Jodelet et al.
Future-Proofing Class-Incremental Learning
by Quentin Jodelet, Xin Liu, Yin Jun Phua, Tsuyoshi Murata
First submitted to arxiv on: 4 Apr 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 This research proposes a novel approach to exemplar-free class incremental learning, a challenging setting where replay memory is unavailable. The method uses a pre-trained text-to-image diffusion model to generate synthetic images of future classes, which are then used to train the feature extractor. This approach improves state-of-the-art methods and achieves higher performance than using real data from different classes. Experiments on CIFAR100 and ImageNet-Subset demonstrate the effectiveness of this method, especially in settings where only a few classes are available during the first incremental step. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper explores ways to improve class incremental learning without needing old examples of previous classes. The current methods that work well use frozen feature extractors, but they’re not perfect and can struggle when there’s not enough data from new classes. To solve this problem, researchers propose using a special model that generates fake images of future classes. This helps train the feature extractor better. Results show that this approach beats existing methods and is especially helpful when only a few new classes are available. |
Keywords
* Artificial intelligence * Diffusion model
