zeroth-order methods

New meta-learning algorithm tackles AI optimization challenges

Researchers have introduced a novel meta-learning algorithm called Greedy dynamical meta-learning, designed to overcome limitations of existing optimization methods for large AI models. The algorithm employs a two-loop …