Tom Goldstein

Tom Goldstein is the Volpi-Cupal Endowed Professor of Computer Science at the University of Maryland, with additional appointments in Applied Mathematics and Electrical and Computer Engineering. He serves as Director of the Maryland Center for Machine Learning, where he leads research at the intersection of theoretical foundations and practical systems development.

His research interests span multiple areas of artificial intelligence, with particular focus on responsibly building AI systems that leverage mathematical foundations and efficient hardware. His work addresses critical challenges in the training, fine-tuning, benchmarking, and robustness of large models, drawing on expertise from signal processing and applied mathematics.

Goldstein's recent publications demonstrate a strong trend toward developing more efficient and responsible AI systems, with significant contributions in reasoning architectures (Huginn-3.5B), safer diffusion models, and efficient graph algorithms. His work on latent reasoning shows how recurrence can dramatically improve reasoning capabilities without requiring special chain-of-thought training data.

• SIAM's DiPrima Prize
• DARPA Young Faculty Award
• Sloan Fellowship

As an academic advisor, Goldstein mentors PhD students including jwkirchenbauer at the University of Maryland. His research has been supported by significant grants including Amazon Research Awards in 2021 (for automated graph algorithms) and 2023 (for building safer diffusion models). He collaborates extensively with researchers across academia and industry, particularly through Amazon's research programs.

Goldstein leads research efforts at the Maryland Center for Machine Learning, where his team focuses on the boundary between theoretical mathematics and practical AI systems implementation. His lab has recently developed innovative approaches to reasoning models that demonstrate superior performance with fewer parameters and training tokens compared to conventional approaches.