Jerry Zhang
student., occasional writer.
About
I study physics and statistics at Harvard (class of 2027). I'm drawn to statistical physics and dynamics, materials and energy, and the information-theory and interpretability side of machine learning. After graduation I'm joining Hudson River Trading as an algorithm developer.
Along the way I've simulated ion transport in battery electrolytes at NASA Langley, used neural networks to study quantum magnets in the Kaxiras group, and interned in quantitative research at Citadel Securities. At Harvard I've been a course assistant for physics and helped run HackHarvard, including its events abroad.
I grew up in Minnesota doing math and physics competitions, and I speak English, Chinese, and a little German.
Projects
- Latent structure in protein sequences via HMMs 2026
How much of a protein’s 3D organization can a simple, interpretable model recover? Unsupervised hidden Markov models find recurring biochemical patterns in amino-acid sequences, and I compare their latent states against hydrophobicity, secondary structure, and solvent accessibility, alongside composition-based and supervised baselines.
- Optimization and sampling in spin glasses 2026
Competing interactions give spin glasses rugged energy landscapes full of local minima. This project measures how temperature, disorder, and interaction structure affect discrete and gradient-based algorithms across four canonical spin models, and when each one gets trapped.
- Multiple descent for minimum ℓ₁-norm interpolators 2025
Heavily overparameterized models can keep improving well past the interpolation threshold. I reproduce prior multiple-descent results for Gaussian designs and extend them to study how signal sparsity and feature distributions drive a further drop in test error.