Section: New Software and Platforms

Stochs

fast stochastic solvers for machine learning

Keyword: Machine learning

Functional Description: The stochs library provides efficient C++ implementations of stochastic optimization algorithms for common machine learning settings, including situations with finite datasets augmented with random perturbations (e.g. data augmentation or dropout). The library is mainly used from Python through a Cython extension. Currently, SGD, (S-)MISO and (N-)SAGA are supported, for dense and sparse data. See the following reference for details:

A. Bietti and J. Mairal. Stochastic Optimization with Variance Reduction for Infinite Datasets with Finite-Sum Structure. arXiv 1610.00970, 2017.

• Participants: Alberto Bietti and Julien Mairal

• Contact: Alberto Bietti

• Publication: Stochastic Optimization with Variance Reduction for Infinite Datasets with Finite-Sum Structure

• URL: https://github.com/albietz/stochs