I am a PhD student at MIT CSAIL, where I work with Bill Freeman and Josh Tenenbaum. Before MIT, I studied EECS at Berkeley and worked on computational imaging with Laura Waller and quantum computing with Alp Sipahigil.
I study visual perception and physical reasoning, using probabilistic programming and machine learning to construct structured 3D representations of the physical world that support boundedly rational decision-making. My interests are in understanding how human-level visual scene understanding emerges from the combination of rich symbolic structure and learned features, and in formalizing the computational principles that govern how agents perceive, reason, and act in the physical world.
If you're also interested in visual perception, physical reasoning, probabilistic programming, or Bayesian modeling, please send me an email! If you're considering grad school and aren't sure about how the process works, or just want to talk to a current grad student to see if it's right for you, consider checking out the MIT EECS Graduate Application Assistance Program (GAAP).
Selected papers, ordered roughly by relevance to my current work.
A generative model of moving matter that groups motion and appearance cues into particles and clusters, supporting motion-based perception across random dots, camouflaged structure-from-motion scenes, and naturalistic RGB video.
An uncertainty-aware perception system for structured 3D scenes, using a hierarchical Bayesian model and GPU-accelerated inference to learn novel objects from a few views and infer object pose and scene composition in clutter.
A learned domain-randomization method for robotic reinforcement learning, using normalizing flows to model flexible simulation distributions that train robust skills and serve as uncertainty-aware artifacts for multi-step manipulation planning.
A computational imaging study of aberration recovery in Fourier ptychographic microscopy, showing where software-only self-calibration breaks down and how a simple diffuser calibration can extend accurate reconstruction to larger optical errors.
Identifies interface piezoelectricity as a decoherence channel in silicon-based superconducting quantum processors, showing that aluminum-silicon interfaces can convert microwave energy into acoustic loss and limit qubit quality factors.
Uses phononic-bandgap engineering to protect a superconducting qubit from defect-mediated phonon loss, reshaping its dissipative environment and pointing toward smaller, longer-lived qubits for quantum processors.