Linked from

The 26 pages that link to Gradient descent, each with the reason it gives.

• Machine learningRelated: It is a common way to adjust model parameters to reduce training loss.

• GradientRelated: It uses the gradient to choose the direction of fastest local decrease.

• Euclidean distanceRelated: Squared Euclidean error gives a simple, differentiable objective for many such updates.

• Chain ruleRelated: Gradients of nested model computations are obtained through chain-rule propagation.

• DifferentiabilityRelated: Its standard update requires differentiability of the objective with respect to its parameters.

• Convex optimizationRelated: It solves smooth unconstrained convex problems through repeated local steps.

• StandardizationRelated: Scaling predictors similarly can improve its numerical conditioning and convergence.

• Euclidean normRelated: The Euclidean norm can measure gradient size and parameter-step length.

• Objective functionRelated: For differentiable objectives, its gradients guide successive improvements.

• Neural networkRelated: Training commonly uses it to reduce a network’s prediction error.

• Parameter estimationRelated: It can find parameter values that minimize loss or maximize likelihood.

• BackpropagationRelated: It uses gradients computed by backpropagation to reduce the network’s loss.

• Deep learningRelated: Training uses it to reduce prediction error by changing network parameters.

• Convolutional neural networkRelated: It is commonly used to fit the network’s filter weights.

• Mathematical optimizationRelated: It is a standard approach for minimizing differentiable objectives.

• Training dataRelated: It uses errors measured on training examples to update model parameters.

• Optimization problemRelated: It searches for local minima when the objective is differentiable.

• Sign functionRelated: For absolute-value objectives, sign values determine the update direction away from zero.

• Saddle pointRelated: A saddle can slow or redirect gradient descent despite not being a local minimum.

• Differentiable functionRelated: Differentiable objective functions provide gradients that guide local improvement.

• Numerical optimizationRelated: It is the basic derivative-based route toward a local minimum.

• Multivariable calculusRelated: It uses multivariable derivatives to seek low values of functions.

• Full waveform inversionRelated: Its gradient-based updates refine subsurface parameters to lower waveform misfit.