About me

I’m currently a second year PhD student at Politecnico di Milano advised by Prof. Matteo Matteucci.

My background is specifically into Signal Processing and Machine Learning. At the beginning of my studies I was mainly interested in audio signal processing, such as unsupervised techniques for blind source separation. Then, I surrendered to the charm of Deep Learning and during my MSc thesis I worked on a recurrent model for Semantic Segmentation.

My research focuses on Stability and Robustness of Deep Neural Networks (DNNs), especially to adversarial attacks. Neural Networks are powerful and fascinating models, that are incredibly good at generalize to new examples, but they have been proved to be also particularly sensitive to imperceptible perturbations to the input. Much is still to be understood about their properties and this can be approached from different perspectives. I started analyzing robustness of DNNs from a dynamical system point of view, and I’m now digging also into Information Geometry and Optimization.

I’m also particularly interested into continual learning, meta-learning, model adaptation and I believe that Information Geometry can answer many open questions also in these fields.