Summary of Understanding Likelihood Of Normalizing Flow and Image Complexity Through the Lens Of Out-of-distribution Detection, by Genki Osada et al.
Understanding Likelihood of Normalizing Flow and Image Complexity through the Lens of Out-of-Distribution Detection
by Genki Osada, Tsubasa Takahashi, Takashi Nishide
First submitted to arxiv on: 16 Feb 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 This paper investigates the mystery behind deep generative models (DGMs) assigning higher likelihoods to unknown out-of-distribution (OOD) inputs than their known training data. By proposing a hypothesis that less complex images concentrate in high-density regions in the latent space, leading to higher likelihood assignments in Normalizing Flows (NF), this research sheds light on the underlying mechanism driving this phenomenon. The study experimentally demonstrates the validity of this hypothesis across five NF architectures and shows that treating image complexity as an independent variable can alleviate this issue. Furthermore, the authors provide evidence for the applicability of their findings to another DGM, PixelCNN++. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us understand why deep generative models are sometimes bad at figuring out when new pictures don’t belong in a group. It seems that these models think simple pictures are more likely to be part of this group than they really are. The researchers propose an idea that explains why this happens and test it with several types of these models. They find that by treating picture simplicity as something separate from the original model, we can make these models better at identifying new pictures that don’t belong. |
Keywords
* Artificial intelligence * Latent space * Likelihood
