A recent issue of Quanta Magazine featured a story spotlighting new research by Assistant Professor Benjamin Grimmer. Grimmer’s study suggests that a basic assumption about gradient descent, a long-relied on optimization method used in many machine learning programs, may be wrong.