Ann Huang

Hi there! I’m Ann, a fourth-year PhD student in computational neuroscience and AI at Harvard University and the Kempner Institute, supported by the Kempner Graduate Fellowship.

My research lies at the intersection of representation learning, interpretability, and dynamical systems theory, guided by three recurring themes. (1) System identifiability: I develop methods and theory to identify and steer the computations implemented by artificial and biological neural networks. In particular, I study under-specification: when a task or training objective admits many models with the same performance but different internal mechanisms. (2) Representational alignment: I design metrics that compare computations across systems by their dynamics — how internal states evolve in time and how they respond to inputs. (3) Training dynamics: I combine theory and experiment to study how curriculum learning speeds up learning and steers networks away from bad minima, and how training shapes models’ representations and selects among the many solutions a task admits. Ultimately, I aim to translate our mechanistic understandings of deep learning models into more efficient training and safer models that are robust to adversarial attacks and reward hacking.

Outside research, I enjoy skiing, hiking, climbing, reading, travelling, listening to rock music and attending concerts.

I’m always happy to chat about research and life, brainstorm, and collaborate. Reach me at annhuang@g.harvard.edu. If you are an undergrad seeking advice on grad school applications, navigating research opportunities, or figuring out whether a PhD is right for you, feel free to drop me an email too.