Machine Learning as a Service: Protocols and Structures for Inference
James Montgomery (NICTA)
NICTA SML SEMINARDATE: 2012-10-04
TIME: 11:15:00 - 12:00:00
LOCATION: NICTA - 7 London Circuit
CONTACT: JavaScript must be enabled to display this email address.
ABSTRACT:
Machine learning is complex and varied, which works against its widespread adoption. Offering machine learning techniques as a service means reducing the complexity for the consumer while, generally, reducing the flexibility of the service offered. The last three years have seen an explosion in the number of established and startup companies offering machine learning services via the web. Although many of these products are individually quite good, they also have low flexibility. The Protocols and Structures for Inference project imagines an alternative way of providing machine learning as a service that is free, flexible and federated. This seminar will examine the major barriers to supporting a diverse range of machine learning approaches with a simple, consistent, RESTful (web) API, and will demonstrate the protocol we have developed using both cURL (to show itas simple) and a JavaScript client (to show it can support end-user applications).
