Home

Gossip Learning Framework

Machine learning in a fully distributed network environment (i.e. without any central control or synchronization) is more and more important topic and research field. In this model we have a network environment in which the nodes stores data records and can communicate to each other (if they know the ID-s their partners). The task is to build machine learning models which can efficiently predicate some property of the data records (classification problem) satisfying the following restrictions:

• the data records cannot be moved (due to privacy issues)
• during the learning phase there is NO * synchronization * guarantee on package loss * guarantee on package delay
• user can join and leave the network (churn)
• the communication cost (number of messages and message size) should be kept low

In some scientific papers we proposed a generic approach which is based on the following idea: multiple models taking random walks over the network in parallel, while applying an online learning algorithm to improve themselves and getting combined via local and global ensemble learning methods. This project is an implementation of the above introduced generic framework called Gossip Learning Framework (GoLF) based on the PeerSim P2P simulator.