PoET: design and implementation of collaborative machine learning

Machine Learning provides productivity improvements in a numerous fields, such as Healthcare or Insurance. Productivity could be further improved within industries, if individual firms could see if their machine learning models work not only on their data, but on data across the industry.A lack of means to collaborate in a trustless, incentive-compatible manner - information asymmetry - prevents this, leading to a centralisation of data and a more undemocratic machine learning process.

In this paper we propose an interdisciplinary solution to facilitate collaborative machine learning, the "PoET" protocol, utilising concepts from economics (mechanism design), and computer science (blockchain).

Chris Chia, Department of Mathematics