Explore a comprehensive video explanation of the paper "VOS: Learning What You Don't Know by Virtual Outlier Synthesis." Delve into the challenges of out-of-distribution detection in neural networks and discover how the VOS framework addresses these issues through adaptive synthesis of virtual outliers. Learn about the novel unknown-aware training objective that shapes the uncertainty space between in-distribution data and synthesized outlier data. Gain insights into the application of VOS in object detection and image classification models, and understand how it achieves state-of-the-art performance in improving out-of-distribution detection.
Learning What You Don't Know by Virtual Outlier Synthesis - Paper Explained
Overview
Syllabus
- Intro
- Sponsor: Assembly AI Link below
- Paper Overview
- Where do traditional classifiers fail?
- How object detectors work
- What are virtual outliers and how are they created?
- Is this really an appropriate model for outliers?
- How virtual outliers are used during training
- Plugging it all together to detect outliers
Taught by
Yannic Kilcher
