Schuyler Moss
Hi, I’m Schuyler! I am a PhD student at the University of Waterloo where I work with Roger Melko and the Perimeter Institute Quantum Intelligence Lab, or the PIQuIL. Previously, I attended the University of North Carolina at Chapel Hill where I studied physics and mathematics, earning Highest Honors for the defense of my senior honors thesis. Though my undergraduate research experiences span a number of fields, I mostly worked with Joaquín Drut and his Computational Quantum Matter group.
Research and Publications
All of my research has focused on training neural networks to represent quantum many-body wavefunctions. Neural networks trained to accomplish this task are known as neural quantum states (NQS).
My first projects as a graduate student focused on a new method for training NQS, which we called “data-enhanced variational Monte Carlo”. The method involves pretraining an NQS using data from a quantum device and then performing variational Monte Carlo. Our first paper introducing the method was published in Physical Review B. Our second work, which was my first first-author paper, was published in Physical Review A. I presented this work at the Toulouse School for Machine Learning and Quantum Many Body Physics in April 2022, at the IAIFI PhD Summer School in August 2022, and at the 21st annual International Conference on Recent Progress in Many-Body Theories, where I was recognized as one of the best student presentations.
Our research group has done a lot of work using RNN Wavefunctions. In fact, we used RNN wavefunctions in the two works mentioned above. Recently, we used RNN wavefunctions to perform large-scale simulations for two paradigmatic models in quantum many-body physics: the square-lattice antiferromagnetic Heisenberg model (SLAHM) and the triangular-lattice antiferromagnetic Heisenberg model (TLAHM). RNN wavefunctions have special properties that allowed us to perform these large-scale simulations without a prohibative demand for computational resources. We believe that this type of NQS architecture is especially suited for extracting properties of quantum many-body systems in the thermodynamic limit.
For NQS, our learning task is learning a quantum wavefunction. I am generally interested in understanding how different this task is from standard machine learning tasks like computer vision and natural language modelling. Machine learning theory is developing, but it is lagging behind the ground-breaking-heuristic-observation type results. As such, I have started to approach this question (“How different is the NQS setting from standard machine learning?”) by investigating whether the ground-breaking-heuristic-observation type results hold in the NQS setting. Recently, my collaborators at the Flatiron Institute’s center for computational quantum physics and I, observed Double Descent for the first time in the NQS setting. I am also interested in investigating whether neural scaling laws appear in our Heisenberg model simulation data (above projects).
Non-Physics Fun
When I am not working, I’m likely sitting on my front porch reading, tracking down the best cup of coffee (currently tea), KNITTING (!), or partaking in some form of cardio. I’ve never met a body of water I didn’t like, I’m a competitive but indcredibly mediocre chess player (and the same holds generally for other games and puzzles), and I love talking to strangers. I have a lot of hobbies and I will probably try to talk to you about them.
Film Photos
I have taken lots of photos with a camera that my dad used when he was in college. It’s a 1980’s era Nikon FG that shoots 35mm. Here are some of the shots that I am particularly fond of.