Both shrink the step toward a target as you get closer. Binary search halves the search interval each iteration; learning-rate decay shrinks the gradient step as training approaches a minimum. Same intuition: take big steps when far, small steps when near.