Daniel Pfrommer

Daniel Pfrommer is a PhD candidate in electrical engineering and computer science whose research seeks to advance the theoretical analysis and algorithmic development of high- dimensional control systems. As a MathWorks Fellow, Daniel will focus on the approximation of classical control algorithms by neural network-based models. His current work aims to develop new methods for learning latent state representations in high-dimensional, partially observed dynamical systems, performing filtering over the learned representation, and finally, developing new algorithms for learning control policies based on these methods. Daniel’s broader goal is to create a general filtering algorithm capable of performing inference over learned high-dimensional state representations directly from video data. This could yield a tractable approach for mining an entire “world model” from large quantities of unsupervised video data in a scalable manner. His work, which draws heavily on MathWorks products, has implications for the field of machine learning, enabling more efficient compression algorithms, new algorithms for AI-based content editing, and better model-based reinforcement learning algorithms. These capabilities could advance work in many domains, from the deployment of robotic systems to economic and social policy decision-making.