Meta-Learning Mini-Batch Risk Functionals

Jacob Tyo, Zachary C. Lipton

Supervised learning typically optimizes the expected value risk functional of the loss, but in many cases, we want to optimize for other risk functionals. In full-batch gradient de…