Protocols and Structures for Inference: A RESTful API for Machine Learning Services

The last few years have seen rapid growth in the online delivery of machine learning services by a variety of both established and new companies. Each service offers its own, typically RESTful, API and a subset of machine learning techniques. Composing these services is difficult and extending their learning technique offerings impossible for anyone outside their respective development teams. We offer an alternative approach that is flexible and federated: data and learning algorithm providers are given considerable freedom in what they offer, while learning service consumers have the power to easily combine different services to produce new and powerful inference tools. We will demonstrate the features of our RESTful API for machine learning across a number of different example web services.