معرفی
Thomas Goldstein is a Professor of Computer Science at the University of Maryland, with appointments in Applied Mathematics and Electrical and Computer Engineering. He leads the Maryland Center for Machine Learning and focuses on developing responsible AI systems through the integration of optimization, signal processing, and applied mathematics. His work emphasizes efficient hardware utilization and theoretical foundations to improve model training, robustness, and scalability. Goldstein holds a PhD in Mathematics from UCLA and has held research positions at Rice University and Stanford University. He is a recipient of the SIAM DiPrima Prize, a DARPA Young Faculty Award, and a Sloan Fellowship.
Education:
- PhD in Mathematics, UCLA
- Research Scientist, Rice University
- Research Scientist, Stanford University
Research Interests:
Goldstein's research bridges theory and systems, addressing challenges in large-scale machine learning, distributed optimization, and AI safety. His lab develops novel architectures like recurrent depth models for latent reasoning and explores watermarking techniques to protect AI-generated content. Recent work includes the Huginn-3.5B model, which achieves superior performance on GSM8K tasks using latent reasoning without explicit CoT training.
Grants & Awards:
- Sloan Fellowship
- DARPA Young Faculty Award
- SIAM DiPrima Prize
Labs & Teams: The Maryland Center for Machine Learning (MCML) serves as a hub for interdisciplinary AI research, focusing on ethical AI development and scalable systems. Goldstein collaborates closely with industry partners and maintains an active open-source initiative, releasing tools like the Common Pile dataset and Neftune optimization frameworks.
