Ihaka 2019: Deep learning: why is it deep, and what is it learning?

[Music] tonight we are fortunate and grateful to have Tomas speaking to us the original speaker JJ had to cancel unfortunately so a pretty short notice Thomas kindly stepped into the breach so tonight I speaker Thomas something he's been a professor of biostatistics here at Auckland for eight and a half years before that he was at University of Washington in Seattle didn't tell me what he was doing there but I'm sure he's keeping himself busy and he his work has been recognized by being elect as a fellow of the American Statistical Association and of the museum War Society he has very wide range of interests and expertise we used to have a minister Steven Joyce who they called the minister of everything because he was into so much stuff and I reckon we should call Thomas the professor of everything because he's he knows so much about so many things and tonight he's going to talk about and tell us about neural networks he's welcome Chris Thomas Longley [Applause] you're a Tartar assalamu alaikum today i'm going to talk about how deep your networks work and don't work in primarily an image recognition a lot of statisticians have missed out on the big changes in neural nets in the last decade and it really is the vast decade this is new Google Pagerank didn't use neural networks the Amazon recommender from the early Amazon Hong Kong didn't use neural networks the Netflix prize candidates didn't use neural networks it's a they only started working very recently and a lot of statisticians haven't caught up on the big changes so this is a system called visual dialogue you can go to that website upload a picture and have a conversation with the visual chat bot about the picture so this example is from Janelle Shane who is a photonics research of it also has a large sideline in making your networks do stupid things so the duck rabbit picture and you upload it and the visual chat bot says a close-up of a black and white bird that's really impressive if you think about it...