Summary of A Geometric Explanation Of the Likelihood Ood Detection Paradox, by Hamidreza Kamkari et al.
A Geometric Explanation of the Likelihood OOD Detection Paradox
by Hamidreza Kamkari, Brendan Leigh Ross, Jesse C. Cresswell, Anthony L. Caterini, Rahul G. Krishnan, Gabriel Loaiza-Ganem
First submitted to arxiv on: 27 Mar 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (stat.ML)
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 investigates the puzzling behavior of likelihood-based deep generative models (DGMs) when trained on complex datasets and confronted with out-of-distribution (OOD) data from simpler sources. Specifically, it explores why DGMs assign higher likelihood values to OOD data yet fail to generate such samples. The authors propose a novel method for OOD detection that leverages local intrinsic dimension (LID) estimation and pairs it with likelihood estimates obtained from pre-trained DGMs. This approach can be applied to various DGM architectures, including normalizing flows and score-based diffusion models, and achieves state-of-the-art results in OOD detection benchmarks using the same DGM backbones. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks into why deep generative models (DGMs) are good at recognizing what they haven’t seen before. When trained on a lot of data, these models get really good at making things that look like that data. But sometimes, they’re tricked into thinking something is likely when it’s actually not. The authors found out that this happens when the model sees new things that are similar to what it has learned before. They also discovered how to use this phenomenon to create a better way of figuring out whether new data is real or not. This method can be used with different types of DGMs and does a great job at telling apart real and fake data. |
Keywords
* Artificial intelligence * Likelihood
