Multi-view Anomaly Detection via Robust Probabilistic Latent Variable Models

W e propose probabilistic latent variable models for multi-view anomaly detection, which is the task of finding instances that have inconsi stent views given multi-view data. With the proposed model, all views of a non-anomalous instance are assumed to be generated from a single latent vector. On th e other hand, an anomalous instance is assumed to have multiple latent vecto rs, and its different views are generated from different latent vectors. By infer ring the number of latent vectors used for each instance with Dirichlet process p riors, we obtain multi-view anomaly scores. The proposed model can be seen as a robus t extension of probabilistic canonical correlation analysis for noisy mu lti-view data. W e present Bayesian inference procedures for the proposed model based on a stochastic EM algorithm. The effectiveness of the proposed model is demon strated in terms of performance when detecting multi-view anomalies.