Abstract
The talk will cover a variety of probabilistic hashing methods for compact data representations and their applications in search and learning with massive, high-dimensional, and possibly dynamic streaming data. For example, fitting logistic regression (or SVM) with billion or billion square variables will be challenging and highly useful in the context of search; and the recent development of b-bit minwise hashing and one permutation hashing will be highly effective for this type of applications. Another exciting example is sparse recovery (compressed sensing). This talk will also present new hashing algorithms for accomplishing efficient signal recovery.