PhD in the Department of Applied Mathematics and Theoretical Physics (DAMTP) as part of Machine learning and AI for medicine group and supervised by Prof. Mihaela van der Schaar. Publications. I have published 13 papers NeurIPS/ICML/ICLR, the most prestigious machine learning conferences. Synopsis. We are gradually entering in a phase where humans will increasingly interact with AIs, hence forming human-AI teams. I see an immense potential in these teams to approach cutting-edge scientific and medical problems. My current research focuses on making these teams more efficient by improving the information flow between complex ML models and human users. This touches upon various fascinating question including:· Interpretability. How to interpret the predictions of the model? Can we trust the model?· Representation Learning. How to learn efficient representations of the data? Can we learn representations that are easier for humans to understand?· Data-Centric AI. Can we understand the model's failure modes by analyzing the training data? What do we mean really by "good data"· Generative AI. How to generate new data in a way that respects a list of desiderata? How can the above fields help with that? For more details, please check my attached presentations bellow.

5 months internship in the Machine Learning Research team, where I studied the robustness of state-of-the-art multimodal machine learning models. We published the results in a paper in which we discover outlier features and high polysemanticity in zeroshot CLIP models. We show that only 20% of the latent space directions contribute to the model effective robustness.

3 months internship in the AI4Science team, where I contributed to the development of new machine learning approaches for material discovery. My results will be published as part of a big paper.

10 weeks of summer internship in machine learning applied to quantitative finance.

1 year of full-time research in black-holes physics and quantum field theory. Working on such an abstract subject developed my ability to think about problems in a conceptual way. I also realised that this conceptual thinking can serve more concrete fields to have a real-world impact.

1 year of full-time research in black-holes physics and quantum field theory. Working on such an abstract subject developed my ability to think about problems in a conceptual way. I also realised that this conceptual thinking can serve more concrete fields to have a real-world impact.

Part-time research internship in the van der Schaar Lab. This was my first exposure to concrete machine learning research. I managed to develop a new interpretability method to translate black-box machine learning models into closed-form symbolic expressions. This led to a paper published in NeurIPS 2020 and a PhD offer that I accepted.

4 months of full-time research in Modified Theories of Gravitation. This project was a good mix between theoretical and experimental work. I had to study pathological behaviours of gravity theories and assess my findings through numerical simulations. During this first experience as a researcher, I have also learned the exploration-exploitation trade-off familiar to all researchers.