Summary of Learning Label Hierarchy with Supervised Contrastive Learning, by Ruixue Lian et al.
Learning Label Hierarchy with Supervised Contrastive Learning
by Ruixue Lian, William A. Sethares, Junjie Hu
First submitted to arxiv on: 31 Jan 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
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 The paper introduces a family of Label-Aware Supervised Contrastive Learning (LASCL) methods that incorporate hierarchical information to improve the performance of contrastive learning frameworks in scenarios where label hierarchy exists. The proposed approach adjusts the distance between instances based on measures of class proximity and introduces an instance-center-wise contrastive to move within-class examples closer to their centers, represented by learnable label parameters. These learned label parameters can be used as a nearest neighbor classifier without further fine-tuning, generating a better feature representation with improved intra-cluster compactness and inter-cluster separation. The proposed LASCL is evaluated on three datasets and outperforms baseline supervised approaches in text classification tasks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper creates a new way to learn from data that takes into account how classes are related to each other. This helps when you have lots of similar classes, like different breeds of dogs, but also some very different ones, like dogs and cats. The method adjusts how close or far apart different instances are based on how close their class labels are. It then uses this information to make the features learned from the data more useful for classification tasks. This new approach is tested on three different datasets and does better than other methods in a specific task. |
Keywords
* Artificial intelligence * Classification * Fine tuning * Nearest neighbor * Supervised * Text classification
