Summary of Gps-ssl: Guided Positive Sampling to Inject Prior Into Self-supervised Learning, by Aarash Feizi et al.
GPS-SSL: Guided Positive Sampling to Inject Prior Into Self-Supervised Learning
by Aarash Feizi, Randall Balestriero, Adriana Romero-Soriano, Reihaneh Rabbany
First submitted to arxiv on: 3 Jan 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 A novel approach to self-supervised learning (SSL) is proposed, which leverages guided positive sampling to incorporate prior knowledge into the selection of positive samples. This method, called GPS-SSL, uses a metric space where Euclidean distances serve as a proxy for semantic relationships, allowing for the generation of positive samples from nearest neighbor sampling. Unlike current SSL methods that rely on data augmentations (DA) for generating positive samples, GPS-SSL decouples prior knowledge from DA, making it applicable to any SSL method. The proposed approach outperforms baseline methods on various downstream datasets from different domains, even when using minimal or strong DAs. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary GPS-SSL is a new way to help machines learn without labeled data. It lets you give the machine some extra information about what you want it to learn, and then uses that information to pick the best examples for learning. This makes it easier for the machine to learn and understand new things. The method works by creating a special kind of space where distances between things are meaningful, and then using that space to find the most important examples. GPS-SSL is simple to use and can be applied to many different machine learning methods. It even works well when there’s not much data or when the data is tricky. |
Keywords
* Artificial intelligence * Machine learning * Nearest neighbor * Self supervised
