About me
Aaron Defazio is a Research Scientist at Meta on the Fundamental AI Research Team, specializing in the field of optimization algorithms for machine learning. Aaron holds a PhD in Computer Science from ANU (Australian National University) and has a rich background in research, having contributed to projects such as the development of deep neural network reconstructions for high acceleration rates in MRI imaging. He is also known for the development of the SAGA algorithm for variance-reduced finite sum optimization, and the D-Adaptation method for learning rate adaptation, which was awarded the 2023 ICML Outstanding Paper award.
