Can we design machines that learn the way humans learn, namely, without supervision and in a way that does not quickly plateau, but is able to scale to the complexities of the real world? Discussion and study of intelligence dates back centuries, but it is only within the last century that advances in computing hardware have allowed us, for the first time, to attempt to duplicate aspects of human intelligence. Current researchers are scrambling to answer several questions, particularly, just what can we do with this new found abundance of computational power, and how can we apply it to the exploding volume of data now available to learning algorithms? This talk will outline several recent research projects aimed at chipping away at these questions, starting with robots that learn how to walk, robots that accomplish tricky tasks by delegating high-level components to humans, and finally, crowdsourced design of 3D objects. Common themes of the projects include evolutionary computation and machine learning combined with limited human interaction.
Bio: Jason Yosinski is a PhD student at the Cornell Creative Machines Lab, where he investigates artificial intelligence, tries to get robots to learn on their own, and hopes someday to have his robots finish his thesis so he doesn't have to. Since starting grad school, he has helped create the first robotic guest to be interviewed on NPR, and his work in AI has been featured in New Scientist, Slashdot, MIT’s Technology Review, Kurzweil AI, and the BBC. Before coming to Cornell, Mr. Yosinski graduated from Caltech, worked at a statistics startup, and then spent a year developing a program in Pasadena that tricks middle school students into learning math while they play with robots.