Marcella ASTRID

Marcella ASTRID is a Researcher at the University of Luxembourg's Interdisciplinary Centre for Security, Reliability and Trust (SnT), specifically within the CVI2 department. Her research focuses on advanced machine learning techniques applied to cybersecurity and computer vision challenges, particularly in the detection of synthetic media (deepfakes) and anomaly detection systems. She specializes in developing robust models that address vulnerabilities in adversarial scenarios and improve generalizability across diverse datasets.

Key research areas include deepfake detection through spatio-temporal analysis, anomaly detection via novel training paradigms, and efficient neural network compression for resource-constrained environments. Her work bridges theoretical advancements in machine learning with practical applications in surveillance systems, autonomous robotics, and cybersecurity infrastructure. Recent contributions emphasize leveraging autoencoder weaknesses, pseudo anomaly generation, and localized attention mechanisms to enhance detection accuracy and model robustness.

Marcella’s research also explores cross-modal learning (audio-visual synchronization) and semi-supervised methodologies to tackle data scarcity issues in critical domains like battery thermal imaging and multi-camera traffic classification. She actively publishes in top-tier venues and collaborates on interdisciplinary projects addressing real-world security and reliability challenges.