Luiza Pozzobon

I am a PhD student at the University of Washington where I am luckily advised by Noah Smith and Luke Zettlemoyer. Before UW, I completed my masters at Unicamp with professors Paula Dornhofer and Eduardo Valle. I have also been a research scholar at Cohere Labs where I worked with Sara Hooker, Beyza Ermis, and Patrick Lewis.

I am interested in learning how systems learn through an information-theoretic standpoint. Some of the questions I’ve been thinking about are in the lines of:

• How is information compressed in the weights of a model?
• What is the interplay of the information contained in data and what can be extracted by a compute-bounded observer?
• Can we quantify the “capacity” of a model and how different data distributions might make use of that capacity?

Due to the broad nature of these questions, I also dabble with ideas from the representation learning, learning dynamics, optimization, algorithmic information theory and (mechanistic?) interpretability literature.