My main focus is finding strategies to optimize highly non-convex problems that arise in data science and biophysics. 

My recent works focus on inferring the structure of data with unsupervised neural networks, trying to simplfy modern architertures to understand deep learning with a physicist's perspective. In particular, we found that simple Hopfield networks can recognise examples from a test dataset by exploiting the spurious mixtues in the energy landscape.

During my PhD I also studied local-entropy-based algorithms to train neural networks, which find rare solutions with better generalization properties. I applied the same kind of algorithms to characterize pre-biotic proteins and possible connections with the neutral theory of evolution.

You can find more here.