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

This international research project, titled Data-Driven Anomaly Detectors for Time Series Data and Big Data (DD-AnDet), focuses on addressing the unsolved problem of anomaly detection in time series and Big Data. The issue arises because Deep Learning (DL) and Machine Learning (ML) models are prone to inaccuracies when inferring online data that may be corrupted or reveal anomalous behaviors, which is critical in applications like predictive maintenance and dynamic gesture recognition. The proposed solution involves utilizing advanced DL techniques, such as contrastive learning, multimodal learning, domain adaptation, and label shift, in addition to addressing concept drift via dynamical clustering. The project aims to create robust models capable of detecting known anomalies or deviations from a normal state, overcoming the limitations of classical approaches that rely on assumptions not always applicable to online data. Expected outcomes include the development of four indexed scientific articles in Scopus, the creation of experimental databases for testing, and contributions to AI and ML theory, positively impacting industrial efficiency and communication accessibility.

Goal:
The main goal of this project is to develop innovative and effective anomaly detection models that leverage cutting-edge deep learning techniques—including contrastive learning, multimodal learning, domain adaptation, and label shift—to identify anomalies across various applications such as fault detection in industrial devices and dynamic gesture recognition, thereby advancing the field through the contribution of new, multidisciplinary approaches to anomaly detection.

Research lines:
Control engineering and automation technologies