Summary of Approximations to the Fisher Information Metric Of Deep Generative Models For Out-of-distribution Detection, by Sam Dauncey et al.
Approximations to the Fisher Information Metric of Deep Generative Models for Out-Of-Distribution Detection
by Sam Dauncey, Chris Holmes, Christopher Williams, Fabian Falck
First submitted to arxiv on: 3 Mar 2024
Categories
- Main: Machine Learning (stat.ML)
- Secondary: Computer Vision and Pattern Recognition (cs.CV); 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 This paper addresses the open problem of out-of-distribution (OOD) detection using likelihood-based deep generative models, specifically score-based diffusion models and variational autoencoders. The authors reproduce seminal work showing that these models consistently infer higher log-likelihoods for OOD data than training data. They propose a new method based on the gradient of a data point with respect to the model parameters, motivated by the intuition that OOD data should have larger gradient norms. This is formalized using the Fisher information metric, which has large absolute diagonal values, leading to chi-square distributed layer-wise gradient norms as features. The authors combine these features for OOD detection and demonstrate their method outperforms the Typicality test for most deep generative models and image dataset pairings. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about how we can use special kinds of computer programs called machine learning models to figure out when new data doesn’t fit with what we’ve seen before. These models are really good at guessing what data might look like, but they have a problem: they tend to think new, weird data is actually part of the group they’re used to seeing. The researchers in this paper want to fix that by looking at how the model’s “brain” changes when it sees new data. They came up with a simple way to do this using something called the Fisher information metric. It turns out this works really well, and their method is better than other ways people have tried to solve this problem. |
Keywords
* Artificial intelligence * Likelihood * Machine learning
