OCSVM-Guided Representation Learning for Unsupervised Anomaly Detection

Name: OCSVM-Guided Representation Learning for Unsupervised Anomaly Detection
Uploaded: 2026-10-09T01:54:10.811633+00:00
Duration: 436 s
Description: This method combines autoencoders with a One-Class SVM to find rare anomalies in unlabeled data. It uses a custom loss function that guides the encoder to create latent representations perfectly aligned with the OCSVM's decision boundary. This tight coupling improves detection accuracy, especially for subtle medical lesions, by ensuring normal data is well-clustered.

This method combines autoencoders with a One-Class SVM to find rare anomalies in unlabeled data. It uses a custom loss function that guides the encoder to create latent representations perfectly aligned with the OCSVM's decision boundary. This tight coupling improves detection accuracy, especially for subtle medical lesions, by ensuring normal data is well-clustered.

Open the full episode