Juraj Marusic
PhD Student in Statistics, Columbia University
Columbia University
New York, NY, USA
I am a third-year PhD student in the Statistics Department at Columbia University, advised by John P. Cunningham and Marco Avella Medina.
My work sits at the intersection of formal statistics and machine learning. In Bayesian deep learning, I study how the implicit bias of optimization can take the place of hand-crafted priors over network weights, and how choices such as the learning rate shape the function that training converges to. In robust Bayesian inference, I study how posteriors behave when the data are contaminated or the model is misspecified.
Before coming to Columbia, I completed Part III of the Mathematical Tripos at the University of Cambridge, and earned my undergraduate degree in mathematics at the University of Zagreb in Croatia.
You can reach me at juraj.marusic [at] columbia [dot] edu
Publications * denotes co-lead author
A full list of publications can be found on my Google Scholar profile.
Variational Deep Learning via Implicit Regularization
Characterizing the Edge of Stability in Variational Training Without Priors
A Theoretical Framework for M-posteriors: Frequentist Guarantees and Robustness Properties
Differentially Private Hyperparameter Tuning using Local Bayesian Optimization