Anomaly Attribution

Anomaly attribution aims to identify the root causes of unexpected deviations in model predictions or system behavior, providing explanations for why an anomaly occurred. Current research focuses on developing methods, including generative models like variational autoencoders and likelihood-based approaches, that can handle both black-box models and noisy data, often addressing the limitations of existing explainable AI techniques. This field is crucial for improving the reliability and trustworthiness of AI systems across various domains, from engineering and scientific instrumentation to building energy management, by enabling better understanding and mitigation of anomalous events.