معرفی
Micah Goldblum is an Assistant Professor at Columbia University, focusing on machine learning research with emphasis on AI safety, automated data science, and large-scale model training strategies. His work explores foundational topics like generalization theory, Bayesian inference, and algorithmic reasoning. He received a Ph.D. in Mathematics from the University of Maryland, advised by Tom Goldstein and Wojciech Czaja, followed by a postdoctoral fellowship at New York University under Yann LeCun and Andrew Gordon Wilson.
His research portfolio includes developing robust benchmarks for large language models (e.g., LiveBench), analyzing inductive biases in ML systems via Kolmogorov complexity, and advancing adversarial detection techniques like Binoculars. Key contributions span LLM evaluation, diffusion model theory, and fairness in computer vision systems. Goldblum’s work on Bayesian model selection and compression bounds for large language models has provided critical theoretical insights.
Goldblum has been recognized with the 2022 ICML Outstanding Paper Award for his work on Bayesian model selection. His research also addresses societal challenges, such as protecting privacy in facial recognition systems (LowKey) and mitigating biases in neural architectures. Collaborations include leading projects like the Battle of the Backbones vision benchmark and exploring adversarial robustness in federated learning.
His labs and teams focus on interdisciplinary applications of ML, including automated benchmarking, secure AI deployment, and ethical alignment of LLMs. Goldblum’s publications consistently bridge theory and practice, with recent breakthroughs in diffusion model design, continual learning algorithms, and understanding the intrinsic properties of data distributions.
