Summary of Semi-supervised Concept Bottleneck Models, by Lijie Hu et al.
Semi-supervised Concept Bottleneck Models
by Lijie Hu, Tianhao Huang, Huanyi Xie, Chenyang Ren, Zhengyu Hu, Lu Yu, Di Wang
First submitted to arxiv on: 27 Jun 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
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 Concept Bottleneck Models (CBMs) have gained popularity for providing explanations of deep learning models, while maintaining high accuracy using human-like concepts. However, current CBM training relies heavily on expert-annotated concept labels, which can be costly and require significant resources. Additionally, concept saliency maps often misalign with input saliency maps, causing irrelevant predictions. To address these limitations, we propose SSCBM (Semi-supervised Concept Bottleneck Model), a framework suitable for practical situations with scarce annotated data. Our approach leverages joint training on labeled and unlabeled data, aligning the latter at the concept level to solve annotation alignment issues. We proposed pseudo label generation and an alignment loss strategy. Experiments demonstrate the effectiveness and efficiency of SSCBM, achieving 93.19% concept accuracy and 75.51% prediction accuracy with only 20% labeled data. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about a new way to understand how computers learn from examples. Right now, these computers can make good predictions, but we don’t know why they’re making those predictions. The authors want to change that by creating a new model that explains its decisions using simple concepts. The problem is that making this model requires a lot of labeled data, which can be expensive and time-consuming. To fix this, the authors created a new model that can use some unlabeled data too. This helps the model learn more efficiently and make better predictions. The results show that this new model works well, even when we only have a little bit of labeled data. |
Keywords
» Artificial intelligence » Alignment » Deep learning » Semi supervised
