Julia Balla
I am a third-year PhD student at MIT EECS co-advised by Professors Tess Smidt and Tommi Jaakkola. My research focuses on understanding how structure in data should inform the design of machine learning methods, particularly in the context of generative modeling and AI for scientific discovery. I am supported by the NDSEG Fellowship.
Previously, I completed my M.S. in Computer Science at the University of Oxford as a DeepMind scholar and my B.S. in Mathematics with Computer Science at MIT.
Our new blog post on tokenization for non-sequential data was accepted to the ICLR 2026 Blog Post Track!
C14311: Minecraft Fires, Social Networks, and Quantum Complexity
I co-taught a class on graph theory and complex systems science to high schoolers at MIT Splash.
[slides]
Square Peg, Round Hole: Plugging Non-Sequential Data into Sequential Language Models
Exploring the emerging landscape of techniques for turning non-sequential data into 1D sequences, which autoregressive models are designed to process.
ICLR Blog Post Track 2026
[post]
What’s the Erdős Number of an LLM?
Algorithmic and mathematical discovery with machine learning.
[post]
Over-squashing in Graph Neural Networks
Intro to over-squashing, its relation to other measures of information loss in GNNs, and some proposed solutions.
[post]
Ramsey Theory
Final paper for MIT 18.204: Seminar in Discrete Mathematics.
[paper]