Synthetic Outlier

Synthetic outlier generation is a rapidly developing field focused on creating artificial data points that represent anomalies or unusual instances, primarily to improve the robustness and performance of machine learning models. Current research emphasizes generating realistic synthetic outliers using diffusion models and other deep learning architectures, often tailored to specific data modalities (e.g., images, time series, graphs) and addressing challenges like maintaining data utility while protecting privacy. This work is crucial for enhancing anomaly detection, out-of-distribution detection, and the reliable deployment of machine learning systems in safety-critical applications, particularly where labeled outlier data is scarce or expensive to obtain.