libAGML : A C++ framework for Asynchronous Gossip Machine Learning in networks

Download demo and sources: agml-0.1.0-Linux.tar.gz

libAGML is a framework for large scale distributed Machine Learning. It is designed as an intermediary layer between abstract Pattern Recognition/Machine Learning models and available computing/storage resources, in a Cloud Computing spirit. Unlike MapReduce, MPI-based or other frameworks, libAGML works in a fully decentralized and asynchronous fashion, thanks to randomized Gossip protocols.

It comes with a bunch of standard ML algorithms (KMeans, Expectation-Maximization for Gaussian Mixture Models, Principal Components Analysis, Support Vector Machines and gradient-based convex optimization, etc).

For an overview of the underlying paradigm and motivations, take a look at our presentation slides.