Summary of Ovor: Oneprompt with Virtual Outlier Regularization For Rehearsal-free Class-incremental Learning, by Wei-cheng Huang et al.
OVOR: OnePrompt with Virtual Outlier Regularization for Rehearsal-Free Class-Incremental Learning
by Wei-Cheng Huang, Chun-Fu Chen, Hsiang Hsu
First submitted to arxiv on: 6 Feb 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 A recent breakthrough in class-incremental learning (CIL) has shown that using large pre-trained models with learnable prompts can outperform traditional rehearsal-based methods. However, these rehearsal-free CIL methods struggle to distinguish between classes from different tasks, as they are not trained together. To address this issue, we propose a novel regularization method based on virtual outliers that tightens the decision boundaries of the classifier, reducing confusion among classes. Our simplified prompt-based method eliminates the need for a pool of task-specific prompts, achieving comparable results to state-of-the-art methods while using fewer learnable parameters and lower inference costs. We demonstrate the effectiveness of our approach on the ImageNet-R and CIFAR-100 benchmarks, boosting the accuracy of previous SOTA rehearsal-free CIL methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine you’re learning new skills, but each time you learn something new, it’s hard to remember what you learned before. This is a problem that computers face too! They struggle to keep track of all the things they’ve learned, especially when they need to learn something completely new. A team of researchers has found a way to make computers better at learning in this way. They came up with a clever trick called “virtual outliers” that helps computers make better decisions and avoid getting confused between different types of tasks. This means that computers can learn faster and more accurately, which is really important for things like recognizing pictures or understanding what people are saying. |
Keywords
* Artificial intelligence * Boosting * Inference * Prompt * Regularization
