Data-Driven Anomaly Detectors for Time Series Data and Big Data (DD-ANDET)

The project focuses on developing data-driven anomaly detectors for time series data and big data, using deep learning and machine learning techniques. It addresses the challenge of detecting anomalies in various applications, including predictive maintenance for industrial devices and dynamic gesture recognition. The project aims to improve fault detection, enhance cybersecurity, and advance medical applications by creating robust models that can adapt to real-world environments and handle multimodal data. The project involves international collaboration and contributes to the development of new anomaly detection techniques and algorithms.

Goals:
To develop a method using contrastive learning-based models suitable for anomaly detection from time series.
To develop a method using dynamic classification/clustering models based on concept drift from time series.
To develop a method to correctly perform domain adaptation in the presence of label shifts from time series.
To develop a method based on deep learning to engineer meaningful features by combining multimodal sensor data to handle anomaly detection problems from time series.
To perform an exploratory analysis to detect anomalies in dynamic body gestures.
To create a large set of experimental databases of time series that will serve as a test bed for the previous anomaly detectors.

Research lines:
Control engineering and automation technologies