
CoCoA - Communication-Efficient Distributed Dual Coordinate Ascent
A framework for distributed training of machine learning models for classification and regression.
- paper appearing at ICML 2015 arxiv.org/abs/1502.03508 (general convergence proof, and better scalability to larger numbers of machines)
- paper appeared at NIPS 2014 arxiv.org/abs/1409.1458
Code for both algorithms is available as open-source on github.com/gingsmith/cocoa, running on Apache Spark
this is a joint project with Virginia Smith, Martin Takáč, Jonathan Terhorst, Sanjay Krishnan, Thomas Hofmann, Michael I. Jordan