Current Research

Anomaly detection in 3D printing

3D printing failures incur a significant economic cost upon manufacturers, but can be challenging to quickly detect. While some machine-learning models have already shown promise in quickly exposing failed runs, training such models on real-world printing setups is expensive. By using numerical methods to simulate 3D printing, we can quickly iterate models and explore efficient solutions. 

Reinforcement learning for soliton nucleation/control

Magnetic solitons such as skyrmions and hopfions represent energy efficient and stable phenomena that allow for the encoding and transfer of information in magnetic materials. While computing paradigms based on solitons are promising, efficient production and control of solitons in general remain to be fully explored. We aim to evaluate the efficacy of methods such as reinforcement learning and genetic algorithms in this context. 

Interpreting Speech Models (new!)

Speech recognition models turn raw audio into internal representations that are useful, but hard to explain. It is usually unclear what features of speech they actually capture, such as phoneme identity or speaker voice characteristics. By skillfully analyzing these representations, we hope to both develop a deeper understanding of speech itself as well as improving the foundations of speech therapy.

Potential Fall 2026 project Themes

Simulated robotics 

Strongly correlated electron spin systems

Modeling and data analysis for wet-lab collaboration