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. (10[2])Pseudo-residuals (-5[1])are fit using “weak (-5[1])learners” such as decision trees in a method (10[1]-5[1])contrasted (10[1]-5[1])with random forests named for this operation’s (-5[2])“boosting.” (10[9]-5[6])Since (10[2])the sigmoid (10[5]-5[2])activation (10[1])function saturates at extreme values, (10[1])it is susceptible to a problem (10[1])in which this operation “vanishes.” (10[8]-5[1])The negative (10[2])learning (-5[1])rate scales (10[1])the result (10[1])of this operation (-5[2])applied to the loss associated (10[1])with a (10[1])single data (-5[1])point (-5[1])in the update (10[1])step of a stochastic algorithm named (10[1])for this operation. (10[2])An optimization algorithm that takes steps in the opposite (10[1])direction (-5[1])to this operation (10[2]-5[1])is named for this operation’s “descent.” (10[8])For 10 (10[2])points, (10[7]-5[1])name (10[1])this (10[2])operation (10[2]-5[1])that (10[1])outputs (10[1])a vector of partial derivatives. (10[4])■END■ (10[18]0[15])