Results for "partial derivatives"

Matrix of second derivatives describing local curvature of loss.

Matrix of first-order derivatives for vector-valued functions.

Search algorithm for generation that keeps top-k partial sequences; can improve likelihood but reduce diversity.

A narrow minimum often associated with poorer generalization.

Direction of steepest ascent of a function.

Matrix of curvature information.

Gradients shrink through layers, slowing learning in early layers; mitigated by ReLU, residuals, normalization.