Greg Valiant : 2013 Plenary Session

One of the factors compounding the challenges of big data is that the objects and distributions that underlie the data are often extremely large and complex---from social networks and gene interactions, to distributions over enormous domains. Thus, while we might have large datasets at our disposal, the data may only represent a tiny fraction of the actual distribution we hope to understand. The question then becomes how (and to what extent) one can infer properties about the unseen portion of a distribution---the portion that is not represented in the data. We consider several concrete instantiations of this question, and show that one can often make surprisingly accurate inferences.

Gregory Valiant is thrilled to be joining Stanford's CS faculty in Fall 2013. He received his BA in Math from Harvard, and his PhD in Computer Science from Berkeley. His main research interests are in algorithms, learning, applied probability, statistics, and evolution, with his recent work focusing on gaining an algorithmic understanding of fundamental statistical tasks.