gradientinformed
Gradientinformed is an adjective used in mathematics, computer science, and related fields to describe methods, models, or decisions that incorporate gradient information—the derivatives of a target function—with the aim of improving efficiency, accuracy, or speed of optimization, estimation, or planning.
Applications include numerical optimization, where gradient information points toward directions of greatest improvement and is used
Advantages of gradient-informed approaches include faster convergence, improved sample efficiency, and better handling of difficult landscapes
See also: gradient descent, stochastic gradient, automatic differentiation, Bayesian optimization, Gaussian process, gradient-enhanced learning.