ASTER: Latent Pseudo-Anomaly Generation for Unsupervised Time-Series Anomaly Detection

ASTER is a novel framework for unsupervised time-series anomaly detection that generates pseudo-anomalies directly in the latent space, circumventing the need for handcrafted anomaly injections and domain expertise. Utilizing a latent-space decoder to create tailored pseudo-anomalies and a pre-trained LLM to enhance temporal and contextual representations, ASTER demonstrates state-of-the-art performance across three benchmark datasets. This advancement is significant for practitioners as it provides a more efficient and effective method for detecting anomalies in complex time-series data without relying on labeled datasets.