Gradient descent
Revision as of 13:18, 6 June 2011 by Zeno Gantner (talk | contribs)
Gradient descent (GD) is a general optimization algorithm, can be used to find a (possibly local) minimum of a differentiable function. Stochastic gradient descent (SGD) performs updates for single data points (or batches), whereas complete GD computes the complete gradient and then performs an update. In recommender systems, methods based on gradient descent are popular for fitting the parameters of a prediction model, e.g. matrix factorization models.
