Improving Machine Learning Models with Swarm Learning
Training Machine Learning models requires massive amounts of data. When those models are deployed in multiple remote locations, it can be challenging to aggregate the data back to a central repository and then redistribute updates back out to remote sites. HPE has come up with a new approach to training machine learning models call Swarm Learning. In this new approach, remote sites get an initial set of parameters and then use compute at the location to improve the model. The models all communicate in a "swarm", using blockchain they nominate a single node as the source of truth and then improve the model based on information from all locations. In this interview from HPE Discover in Las Vegas, Prasad Shastry explains how swarm learning works and provides some example use cases where there are significant benefits in improving machine learning.