Polina Kirichenko: Anomaly Detection via Generative Models

Data Fest Online 2020 Uncertainty Estimation in ML track Speaker: Polina Kirichenko, New York University In this video, we will talk about how we can use deep generative models (DGMs) in out-of-distribution detection. We will discuss challenges of likelihood based anomaly detection which arise when modeling the distribution of natural images and understand the reasons why DGMs may assign higher likelihood to anomalous data. Finally, we will talk about several recent state-of-the-art approaches which overcome these challenges and apply DGMs to supervised and unsupervised anomaly detection. Register and get access to the tracks: Join the community: