I am a postdoc working with Professor Patrick Rebeschini. My research focuses on machine learning theory and statistical learning, with particular emphasis on heterogeneous data settings.

I have recently been working on two main problems: (i) designing and analysing algorithms for online learning where data arrives sequentially and evolves over time; (ii) I study the training dynamics of popular machine learning architectures such as neural networks and attention mechanisms, particularly how they behave under challenging conditions like non-stationarity, contamination, and distribution shifts. I am especially interested in understanding phenomena such as benign overfitting, early stopping, and in-context learning with data heterogeneity.

I completed my PhD at Cambridge in 2024, where I was advised by Professor Ramji Venkataramanan. My PhD work centered on information theory and statistical learning, motivated by fundamental questions such as: Given a complex statistical estimation problem, what is the minimal amount of data we need to estimate the underlying signal? Can we design polynomial-time, mathematically-principled algorithms that approach the minimum?

• Data heterogeneity: non-stationarity, contamination, and distribution shifts
• General first-order methods, e.g., gradient descent, approximate message passing
• Online learning and bandits
• Information theory and communication systems

I completed a DPhil in Genomic Medicine and Statistics at the University of Oxford in 2024, where I applied statistical genetics to study metabolic and endocrine diseases in large-scale population biobanks such as UK Biobank. Prior to coming to Oxford, I did an undergraduate degree in Molecular Biology (with a minor in Computer Science) at Princeton University, USA.

I currently work on triangulating statistical and machine learning methods to interpret the effects of non-coding genetic variation on both molecular and systemic human phenotypes.