This operation’s output is estimated by simulating trajectories in the REINFORCE algorithm, one of a class of methods named for “policy” and this operation. Pseudo-residuals (-5[1])are fit using “weak learners” such as decision trees in a method contrasted with random forests named for this operation’s (-5[1])“boosting.” (10[1]-5[1])Since (10[1])the sigmoid (-5[1])activation function saturates at extreme values, it is susceptible to a problem in which this operation “vanishes.” The negative learning rate scales the result of this operation applied to the loss associated (10[1])with a single data (-5[1])point in the update step of a stochastic algorithm named for this operation. An optimization algorithm that takes steps in the opposite direction to this operation is named for this operation’s “descent.” For 10 points, (-5[1])name this operation that outputs a vector of partial derivatives. (10[1])■END■ (10[5]0[4])