A machine should be able to ask: “why should I do that?” Most systems learn only from a score saying how well they did. I work on whether they can learn from an explanation instead, e.g., why one action is better than another, which takes me into robot learning, explainability, and how systems plan and decide under uncertainty in human-populated environments.
I am drawn to machine learning that has to survive contact with the real world: robots, physiological signals, and more recently immunology.
I am open to new industry collaborations, and to research visits in other machine learning and robotics groups. I am happy to hear from prospective M2 and PhD students interested in co-supervision. My curriculum vitae is here (PDF).
From September 2025, I joined ISIR, Polytech – Sorbonne University as an Associate Professor! I have been co-teaching Data Science for Public Policy at Sciences Po Paris since 2021.
I contributed to yomix, an interactive tool to explore low-dimensional embeddings of omics data. To keep my feet on the ground and see how things move from academia to industry, I review startup project proposals in my domain for the European Union and collaborate with deeptech companies. Previously, as Head of AI Research at habs.ai, I worked on multimodal machine learning to leverage biometric data for diverse applications, including pain monitoring.
Research Interests (among others):
- Reinforcement Learning and Inverse Reinforcement Learning
- Explainable AI in Sequential Decision-Making
- Interactive Robot Learning (I keep working on a graduate-level educational module)
- Multi-agent multimodal systems for physiological data modeling
You can follow my updates on Github, GoogleScholar, Linkedin. I am co-hosting seminars and podcast sessions on Talking Robotics.
Contact me: tulli[at]isir.upmc.fr.
Click here to know more about me ↓
I graduated in March 2023 with a PhD from Instituto Superior Técnico in Portugal with a thesis focused on the use of explanations, e.g., information about contrastive examples, of AI agents as learning signals. During my doctorate, thanks to one of my thesis supervisors, Francisco S. Melo, I was introduced to inverse reinforcement learning approaches. I then began to wonder how to integrate information about causal relationships into learning by demonstration approaches. The goal was to develop a form of learning that could resemble “why should I do that” rather than simply learning a mapping or a reward function that describes the relationship between states and actions. My idea was to use explanation as a learning signal for agents. I therefore studied both how to generate these explanations, for example through explainable AI, and how to leverage them in robotic learning.
Despite the pandemic, the PhD was an incredible experience, as it was in collaboration with many universities in Europe through a Marie Skłodowska-Curie fellowship. After that, I completed a Postdoc at the Institute of Intelligent Systems and Robotics in Paris, supervised by Mohamed Chetouani. During this time, I initiated two HumanE-AI NET micro-projects and taught many cool subjects at Sorbonne University as an Assistant Professor. Before my PhD, I had the opportunity to complete a bachelor’s thesis on DIY air pollution sensors and georeferenced interactive visualization, and my master’s thesis with Nicu Sebe at the University of Trento. To support my studies, I did various jobs (pizza delivery too!) and completed some internships in both industry and academia, including work on a real-time interface for a remotely operated underwater drone at Witted Srl and serious games for mild cognitive impairments at ISTI-CNR in Pisa.
