Sara Dragutinovic

I'm a machine learning PhD student at the NYU Center for Data Science, advised by Prof. Rajesh Ranganath. I study how the choices we make in training (the optimizer, the architecture, the structure of the data) shape what neural networks end up learning. Right now I'm focused on LLMs. I'm trying to understand why the Muon optimizer works so well and what it costs, and I'm designing transformer architectures that aim to be easier to interpret.

I studied Mathematics and Computer Science at the University of Oxford. I did my Master's thesis at the Gatsby Computational Neuroscience Unit with Aaditya Singh and Prof. Andrew Saxe, on how transformers learn in context.

I'm originally from Serbia. Outside research, I love Latin dancing, I take photos, and I enjoy concerts and theatre.

Email · GitHub · LinkedIn · Twitter/X · Google Scholar