Manuel Madeira
PhD student in Artificial Intelligence at EPFL, advised by Pascal Frossard and Dorina Thanou
My research focuses on the interplay between generative models, in particular diffusion and flow matching, and data structure, often using graph generation and language modeling as testbeds. I explore both how these models exploit inherent data structure and how to explicitly encode this structure into them. I also apply these methods to scientific problems, including molecular generation and digital pathology.
During my PhD, I interned at Apple MLR, where I investigated training-inference alignment in masked diffusion language models, and at Amazon, where I developed graph-based deep learning methods for supply-chain planning. Before starting my PhD, I was a Machine Learning Researcher at Inductiva, a startup where I worked on physics-informed neural networks for heat diffusion and coastal dynamics.
Prior to that, I completed my BSc and MSc in Biomedical Engineering from Instituto Superior Técnico. I was also a visiting student at Tsinghua University. For my MSc thesis, I worked on variance reduction for stochastic convex optimization with Renato Negrinho, Pedro Aguiar, and João Xavier.
For a complete list of publications, check my Google Scholar page.
