I am a Master's student in Artificial Intelligence at the University of Florence and a Researcher at the Media Integration and Communication Center (MICC).
My work bridges the gap between Generative Computer Vision and Engineering Rigor. I specialize in detecting anomalies in visual media (DeepFakes) by modeling authentic signals using Autoencoders, Wavelets, and Transformers. I also care deeply about computational efficiency and sustainable AI ("Green AI").
- Generative Anomaly Detection: Leveraging Masked Autoencoders (MAE) and Variational Bottlenecks (VAE) to learn robust representations of authentic data.
- Frequency Domain Analysis: utilizing Wavelet transforms to detect high-frequency artifacts in synthesized media.
- Multimodal Learning: fusing RGB and Event-stream data for robust object tracking.
Accepted to ACMM '25
The largest drone detection dataset for tracking and forecasting.
- Data: 14+ hours of annotated RGB and Event-stream data across 5 drone models.
- Contribution: Automated annotation pipeline using MetaVision Spatter-tracker to minimize human labeling.
- Benchmark: Validated with YOLO, RT-DETR, and ByteTrack.
Accepted to GREENS '25
A comprehensive study on the environmental impact of programming languages in ML.
- Study: Benchmarked 7 algorithms across C++, Java, Python, MATLAB, and R.
- Tools: Integrated
CodeCarbonto measure energy consumption and CO₂ emissions per run. - Finding: Compiled languages (C++, Java) can be up to 54x more efficient than interpreted ones, but implementation quality matters most.
Languages
Deep Learning & Data
DevOps & Tools
