Summary of Multiood: Scaling Out-of-distribution Detection For Multiple Modalities, by Hao Dong et al.
MultiOOD: Scaling Out-of-Distribution Detection for Multiple Modalities
by Hao Dong, Yue Zhao, Eleni Chatzi, Olga Fink
First submitted to arxiv on: 27 May 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); 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 The paper proposes a novel approach to detecting out-of-distribution (OOD) samples in multimodal scenarios, which is essential for deploying machine learning models in safety-critical applications such as autonomous driving and robot-assisted surgery. The authors introduce the MultiOOD benchmark, characterized by diverse dataset sizes and varying modality combinations, and evaluate existing unimodal OOD detection algorithms on it. They observe that the mere inclusion of additional modalities yields substantial improvements, underscoring the importance of utilizing multiple modalities for OOD detection. The proposed Agree-to-Disagree (A2D) algorithm encourages Modality Prediction Discrepancy between in-distribution and OOD data during training, which is strongly correlated with OOD performance. Additionally, a novel outlier synthesis method, NP-Mix, explores broader feature spaces by leveraging information from nearest neighbor classes and complements A2D to strengthen OOD detection performance. Extensive experiments on MultiOOD demonstrate that training with A2D and NP-Mix improves existing OOD detection algorithms by a large margin. The authors make their source code and benchmark available for further research. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary For out-of-distribution (OOD) sample detection, machine learning models are crucial in safety-critical applications like self-driving cars or robotic surgery. Currently, most studies focus on image data alone. But real-world situations involve multiple types of information. This paper develops a new approach to detect OOD samples in these scenarios. It creates a benchmark called MultiOOD that includes different amounts and types of data. Researchers tested existing methods on this benchmark and found that using more types of information significantly improved performance. The authors suggest two ways to make their method work better: A2D, which helps the model learn from differences between expected and unexpected data; and NP-Mix, which generates new examples by combining features from similar classes. By training with these techniques, existing methods were greatly improved. The researchers share their code and benchmark so others can build on this work. |
Keywords
» Artificial intelligence » Machine learning » Nearest neighbor
