Hung-Hsu Chou
Assistant Professor of Mathematics
University of Pittsburgh
I study how machine learning models learn, why they select particular solutions, and what we can do to create safe, reliable, interpretable models.
RESEARCH
How does structure emerge during training?
I study staged learning, low-rank structure, and the geometry of learned representations.
Which solution does training select?
I investigate how initialization, parameterization, and optimizer influence the outcome.
How can us interpret and use learned structure?
I design anomaly detection that filters unsuitable inputs and metric for semantic accessibility of generative models.
SELECTED WORK
ICLR · 2026
Using the organization of network gradients to identify unfamiliar inputs.
Advances in Computational Mathematics · 2026
Connecting model parameterization with the solutions selected by training.
Preprint · 2026
Studying when semantic properties become predictable during generation.
From calculus to high-dimensional probability, I help students connect intuition, rigorous reasoning, and applications.
Optimization · Probability · Mathematics of machine learning
Spring 2027
Topics in Math Probability
Fall 2027
Mathematical Optimization for Data Science