Summary of Your Classifier Can Be Secretly a Likelihood-based Ood Detector, by Jirayu Burapacheep et al.
Your Classifier Can Be Secretly a Likelihood-Based OOD Detector
by Jirayu Burapacheep, Yixuan Li
First submitted to arxiv on: 9 Aug 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: 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 proposed Intrinsic Likelihood (INK) method offers a rigorous likelihood interpretation for discriminative-based classifiers, enabling effective out-of-distribution (OOD) detection. Building on the constrained latent embeddings of these models, INK scores operate by modeling mixtures of hyperspherical embeddings with constant norm. This connection allows for optimization in modern neural networks and achieves state-of-the-art performance on various OOD detection setups, including far-OOD and near-OOD scenarios. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper proposes a new method called Intrinsic Likelihood (INK) to help machines better detect when they are shown data that is outside what they have learned. This is important because machines often make mistakes when faced with unknown or unexpected information. INK works by using the internal workings of these machines, specifically how they represent and understand different types of data. By doing so, INK can accurately identify when new data is out-of-distribution, helping to prevent mistakes. |
Keywords
» Artificial intelligence » Likelihood » Optimization
