Diving deep into learning
Daniel Rodgers-Pryor uses quantum mechanics to develop neural networks in machines.
Daniel Rodgers-Pryor is something of an old soul. Start up a conversation with the fresh-faced 23-year-old on philosophy or the workings of the brain and you find his knowledge here runs deep.
When it comes to learning, Rodgers-Pryor has a big appetite. Doing a master’s degree in physics at the Centre for Quantum Computation and Communication Technology of the University of Melbourne is just not enough. Along the way he snuck in open online courses on neural networks, cryptography and classes in philosophy. “I refuse to give up anything I’m interested in,” he confesses.
So deep learning describes how Rodgers-Pryor spends much of his time. And he’s also trying to teach computers to do it. “Deep learning” algorithms have given iPhone’s Siri the ability to understand a human voice and to obey instructions. These algorithms rely on neural networks that are modelled on the way the brain learns.
For his masters’ project Rodgers-Pryor is taking these neural networks to the atomic scale. On the weird edge of computing and physics, these circuits use atoms of phosphorus as transistors. They take advantage of the “uncertainty” behaviour of electrons at the quantum level to model how real neurons learn.
These types of atomic “quantum dot” circuits could provide the next generation of computer chips. The techniques lend themselves to building circuits in three-dimensions, packing in more computer power in a smaller space. “There’s lots of room at the bottom”, says Rodgers-Pryor, quoting one of his heroes: physicist Richard Feynman.
What’s next? Joining a new company teaching computers to be seriously smart – and ethical - seems a natural choice.