Riccardo Renzulli

Hello, World! 👋

I am fascinated by understanding and reasoning with deep neural networks and how we can enhance their cognitive plausibility. As different networks encode distinct assumptions, can we leverage these inductive biases to construct more robust models? For instance, is it possible to create networks that require less data and fewer parameters by modeling intricate objects as compositions of distinct parts?

To mention a bunch of keywords, I’m interested in representation learning, object-centric learning, prototypical learning, shortcut learning, robustness, interpretability and foundation models. Additionally, I’m keen on exploring diverse applications, such as medical image analysis and aerial/satellite imagery.

I received my PhD from University of Turin in 2023, advised by Prof. Marco Grangetto. My PhD research focused on capsule networks (check here the presentation of my defense), which are deep learning models that encode objects into vector representations, called capsules, that can be used to model hierarchical relationships better.

In 2022, I spent six months at Aalto University (Helsinki, Finland), supervised by Prof. Ville Kyrki and Francesco Verdoja, where I developed a system for non-GNSS visual localization of UAVs.

In 2018-2019, I also worked for some artificial intelligence companies as deep learning scientist.

I received my master’s and bachelor’s degrees at the Computer Science Department of the University of Turin. My previous research advisor was Prof. Valentina Gliozzi, who introduced me to description logics, non-monotonic reasoning and prototypical representations.

Collaborations! I am looking for collaborations, so if you are interested or you want to have a constructive discussion chat, please drop me an email (riccardo.renzulli@unito.it).

P.S. Make Earth Green Again 🌱