OCSVM-Guided Representation Learning for Unsupervised Anomaly Detection

The article presents a novel method for unsupervised anomaly detection (UAD) that integrates representation learning with an analytically solvable One-Class SVM (OCSVM) through a custom loss function. This approach addresses limitations of existing methods by aligning latent features with the OCSVM decision boundary, resulting in improved detection of small, non-hyperintense lesions in medical imaging tasks, as demonstrated on the MNIST-C benchmark and brain MRI datasets. The findings indicate enhanced robustness to domain shifts, making it significant for practitioners focused on UAD in real-world applications, particularly in healthcare.