Summary of Class Incremental Learning with Task-specific Batch Normalization and Out-of-distribution Detection, by Xuchen Xie et al.
Class Incremental Learning with Task-Specific Batch Normalization and Out-of-Distribution Detection
by Xuchen Xie, Yiqiao Qiu, Run Lin, Weishi Zheng, Ruixuan Wang
First submitted to arxiv on: 1 Nov 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 paper explores incremental learning for image classification, focusing on reducing catastrophic forgetting when old data is restricted. The challenge lies in balancing plasticity (learning new knowledge) and stability (retaining old knowledge). Incremental learning is divided into two paradigms: task incremental learning (TIL), which has access to task-identifiers, and class incremental learning (CIL), where the task-identifier is unavailable. The authors propose an extension to their previous TIL method by predicting task-identifiers through an “unknown” class added to each classification head. This approach enables task-ID prediction, making it applicable to CIL. Task-specific batch normalization modules are used to adjust output feature maps across tasks, enhancing the model’s plasticity. The paper introduces task-specific batch normalization into CIL for the first time and achieves state-of-the-art performance on multiple datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This study looks at how machines can learn new things without forgetting what they already know. When we don’t have access to old information, it’s hard for computers to remember what they learned before. The researchers found a way to balance two important things: learning new things (plasticity) and remembering old things (stability). They divided this process into two types: task incremental learning, where the computer knows what task it’s doing, and class incremental learning, where it doesn’t know. To make their method work in both cases, they added a special “unknown” class to each classification head. This helps the machine figure out which task it’s working on. The researchers also used something called batch normalization to help the computer remember what it learned before. |
Keywords
» Artificial intelligence » Batch normalization » Classification » Image classification
