最近LMAの由来を考えています。in the case of minimizing the sum of squared errors of an approximation to a function over a training set, a step delta is made for the parameter vector, where delta is a vector. Apparently, the direction of de…