Summary of Weiper: Ood Detection Using Weight Perturbations Of Class Projections, by Maximilian Granz et al.
WeiPer: OOD Detection using Weight Perturbations of Class Projections
by Maximilian Granz, Manuel Heurich, Tim Landgraf
First submitted to arxiv on: 27 May 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 Recent advances in out-of-distribution (OOD) detection on image data have shown that pre-trained neural network classifiers can effectively separate in-distribution (ID) from OOD data. Methods have been proposed that leverage the class-discriminative ability of the model itself, either using logit information directly or processing the model’s penultimate layer activations. With WeiPer, we introduce perturbations of the class projections in the final fully connected layer, creating a richer representation of the input. We demonstrate that this simple trick can improve OOD detection performance for various methods and propose a distance-based method that leverages the properties of the augmented WeiPer space. Our results achieve state-of-the-art OOD detection across multiple benchmarks of the OpenOOD framework, particularly in challenging settings where OOD samples are positioned close to the training set distribution. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A new way to detect when images don’t belong has been discovered! Researchers found that using special tricks on pre-trained neural networks can help identify images that don’t fit with a group. They came up with an idea called WeiPer, which adds extra information to the last layer of the network. This helps the network do a better job of recognizing when an image doesn’t belong. The team tested this method and showed it works well on different types of images. They even came up with a new way to measure how well it does! |
Keywords
» Artificial intelligence » Neural network
