SKOLKOVO INNOVATION CENTER, MOSCOW, RUSSIA
APRIL 16, 2019
SKOLKOVO INNOVATION CENTER, MOSCOW, RUSSIA | APRIL 16, 2019

Anomaly Detection based on Zero-Shot Outlier Generation

Mark Skanavi
10:00 10:30

Anomaly detection suffers from unbalanced data since anomalies are quite rare. Generating artificial anomalies is a solution to such ill or not fully defined data. However, synthesis requires an expressive representation to guarantee the quality of the generated data. In this talk, I will present a two-level hierarchical latent space representation that distills inliers’ feature-descriptors into more robust representations based on a variational family of distributions for zero-shot anomaly generation. From the learned latent distributions of inliers, we select those that lie on the outskirts of the training data as synthetic-outlier generators, and we synthesize from them, i.e., generate negative samples without seeing them before, to train binary classifiers. We found that the proposed method allows us to synthesize pseudo outlier samples effectively, and in turn, train robust binary classifiers for outlier detection (without the need for real outliers during training)

Следующее мероприятие в этом зале

Robots that can dream of new molecules

Mark Skanavi
10:30 11:00

Speaker

Associate Professor,
Innopolis University