Organizers: Adam Klivans (University of Texas at Austin; chair), Guy Bresler (Massachusetts Institute of Technology), Anindya De (Northwestern University), Philippe Rigollet (MIT), Sébastien Bubeck (Microsoft Research)
Many learning and testing problems naturally occur in a high-dimensional setting, where it is important to obtain results that are dimension-free (or with only mild dimension-dependence). As a representative example, one can consider the problems of testing and learning juntas: functions of many variables that depend (either precisely or approximately) only on a small subset of the variables. The problem of testing juntas is relatively well understood, while for the problem of learning juntas, there is a large gap between information-theoretic lower bounds and algorithmic results. More recent work introduced some variants of junta testing of a more geometric flavor, where many basic questions remain open. The techniques developed for these problems so far have involved a mixture of algorithmic methods and tools from high-dimensional probability. The purpose of this workshop is to make progress on these problems by bringing learning theorists together with geometers and probabilists who have expertise in high-dimensional phenomena.
All events take place in the Calvin Lab auditorium.
Further details about this workshop will be posted in due course. Enquiries may be sent to the organizers workshop-hd3 [at] lists.simons.berkeley.edu (at this address).