Don't Just Impute, Detect: Task-Aware Diffusion Pipelines for Edge Anomaly Detection

The Task-Aware Diffusion Refiner (TADR) is proposed for Industrial Internet of Things (IIoT) edge pipelines. TADR is a conditional diffusion refiner trained against the loss landscape of a frozen downstream detector. TADR raises F1 by $0.338$ over the unrefined detector on SMD with a paired significance test where $p = 0.0015$. The anomaly enhancement objective alone contributes $0.463$ in ablation.