This article first appeared in the 2019 Physics World Focus on Computing under the headline “Learning from incomplete data”
November 18, 2019
Condensed-matter theorist Gareth Conduit developed an algorithm that can “learn” from incomplete data. His next challenge: turning it into a business.
How did you get the idea for your company?
I was chatting to a materials science PhD student in a pub a few years ago, and he started telling me about some mathematical problems his group was facing. They were trying to use neural networks to predict the properties of new materials as a function of their composition, and I showed them how to use a tool called a covariance matrix to calculate the overall probability that a new material will satisfy various requirements – strength, cost, density and so on – at once. By doing that, we were able to design several new metal alloys, which are now being tested by Rolls-Royce.