About
Edward Hu is a groundbreaking AI researcher and inventor best known for pioneering Low-Rank Adaptation (LoRA), the revolutionary parameter-efficient fine-tuning technique that has become ubiquitous for customizing massive language models and diffusion systems with minimal compute—reducing trainable parameters by orders of magnitude while matching full fine-tuning performance—and μTransfer (also stylized as UTransfer), a zero-shot hyperparameter transfer method that stabilizes and accelerates training at extreme scales by leveraging maximal update parametrization (μP). Developed during his time as a researcher at Microsoft (collaborating closely with OpenAI on GPT-3), these innovations have powered countless downstream applications and underpinned training of the world's largest models. Hu later contributed at OpenAI, pursued PhD research under Yoshua Bengio at Mila (focusing on generative models and robust reasoning), and is now a founding partner in a stealth AI company in Woodside, CA, continuing to push boundaries in efficient, scalable intelligence.