Marco Milanta

Working on an NLP startup focusing on sentiment analysis. First hand experience with large language models.

Worked on deep learning techniques to forecast the effect of droughts 🌎.

I have worked as a TA for Probabilistic Artificial Intelligence, a class by Andreas Krause at ETHz. I held some tutorials and worked behind the scenes to prepare homework and slides.

At Polimi I got in touch with the beautiful world of mathematics. I learned a lot of theorems from all over mathematics, but, even more important, I learned how to come up with extensive and original proofs. I also learned how to apply all of this knowledge to practical problems: from applied statistics to PDE’s numerical calculus.

In this Master’s, I dived into the world of machine learning. Starting from theory, all the way to real-world implementation.

Counterintutively, many very large models perform well even when they interpolate the training data. However, there are still many exceptions in which this phenomenon fails. We investigate how inductive bias plays a crucial role. Paper available here.

We worked on a deep learning model to forecast how satellite images in a fixed location evolve through time, conditioned on weather data. We implemented a convolutional version of s LSTM network to encode prior knowledge of both temporal and spatial dependencies. The preprint available here, code is available in this repo.