Generalization Properties and Implicit Regularization for Multiple Passes SGM
Copyright (C) 2016, Laboratory for Computational and Statistical Learning (IIT@MIT). All rights reserved.
By Junhong Lin, Raffaello Camoriano, and Lorenzo Rosasco
Contact: raffaello.camoriano@iit.it
Please check the attached license file.
This Matlab package replicates the results of the Multiple passes Stochastic Gradient Method experiments presented in the following work:
Lin, J., Camoriano, R. and Rosasco, L., 2016, June. Generalization properties and implicit regularization for multiple passes SGM. In International Conference on Machine Learning (pp. 2340-2348). http://proceedings.mlr.press/v48/lina16.html
We study the generalization properties of stochastic gradient methods for learning with convex loss functions and linearly parameterized functions. We show that, in the absence of penalizations or constraints, the stability and approximation properties of the algorithm can be controlled by tuning either the step-size or the number of passes over the data. In this view, these parameters can be seen to control a form of implicit regularization. Numerical results complement the theoretical findings.