Ryan Adams, a Gates scholar who has just completed his PhD in physics at Cambridge, gave a presentation at one of the world’s leading ideas conferences last week.
Ryan spoke at the ideaCity conference, held in Toronto on 15-17 July. He is profiled in the presenters’ section, alongside other speakers such as Robert F Kennedy Junior (chairman of Waterkeeper Alliance,) architect Will Alsop, Philip Rosedale (founder of Second Life) and Randi Zuckerberg (director of market development at Facebook and sister of its founder Mark).
Ryan, who specialises in statistical machine learning, spoke about machine vision. He was recently awarded a Fellowship at the prestigious Canadian Institute for Advanced Research to work on research into how machines can be created which can see in the same way humans do. He says human vision works in a very different way to machines like cameras. Humans focus on a particular spot when they are looking and this is fairly small. The rest of their vision is peripheral and fairly blurred. However, it is highly mobile and the brain stitches together the small spots of high definition vision. “The brain fools people into thinking that they have a complete view of what they are looking at and know where everything is,” says Ryan. The same occurs when people look at faces. They mainly focus on the eyes, nose and mouth and this information allows them to read emotions better. “We know how to do this from evolution and learning. Computers do not have that skill,” says Ryan, who will be based at the University of Toronto from August. The CIFAR is a nationwide agency which funds research in different venues across the country. Ryan, who was based at St John’s College, did his undergraduate degree at MIT before coming to Cambridge in 2004.
Ryan has also had two papers accepted to appear in the proceedings of the 2009 International Conference on Machine Learning (ICML).
The first paper, “Archipelago: Nonparametric Bayesian Semi-Supervised Learning“, prepared with Professor Zoubin Ghahramani from the Engineering Department, was given an honorable mention for best paper. It looks at how machines can determine and classify images. The second paper, “Tractable Nonparametric Bayesian Inference in Poisson Processes with Gaussian Process Intensities“, prepared with Iain Murray (University of Toronto) and Professor David MacKay from the Cavendish Laboratory, was given an honorable mention for best student paper. It deals with the idea that many data in nature are well described by point processes: bus arrivals, trees in a forest, earthquakes, etc. The researchers seek to estimate the rates of these kinds of processes in a way that makes very few assumptions. The paper provides one of the first solutions to this problem that does not require a simplifying approximation.