Georgiou received his Diploma (5-year integrated Master’s) in Computer Science and Engineering from the University of Patras, Greece in 2019.

Georgiou’s research investigates how masking-based strategies can serve as a unifying principle for learning more robust, generalizable, and interpretable deep learning representations across diverse domains. His dissertation, titled “Learning from What’s Missing: Masked Approaches for Robust and Interpretable Deep Models,” spans three application areas. In remote sensing, he develops cross-scale masked autoencoders that align representations across different satellite image resolutions, enabling effective transfer to multi-resolution real-world data. In general-purpose computer vision, he proposes a dual-space masked distillation framework that combines pixel-level reconstruction with semantic-level teacher alignment, achieving competitive performance on ImageNet-1K. In clinical AI, he applies random feature masking to longitudinal health records for early dementia detection, improving model robustness to missing data while producing more interpretable predictions. Across all three domains, the central insight is that partial observation is not merely a data limitation but a powerful learning signal that forces models to capture broader structure and avoid over-reliance on narrow cues.