Past Events

In the matrix completion problem, one sees a few entries of an (approximately) low-rank matrix and hopes to (approximately) recover the entire matrix.  This problem has gotten a lot of attention recently, partly due to its applicability to recommender systems.  It is known that, by solving a convex program, the original matrix can be recovered with a number of observations that is linear in n.  However, for current analyses of faster algorithms (with runtime linear in n), the number of samples additionally depends at least quadratically on the *condition number* of the matrix.&n

Personalized models often revolve around per-user parameters quantifying, say, an individual's interest in a certain product category or susceptibility to a certain type of advertisement, even after known features of the product and the person have been taken into account.  Social networks offer an appealing way to make inferences about such parameters, the intuition being that one's parameter is "close'' to that of one's friends.  We look at this basic scenario from two angles.First, we consider a Bayesian model that incorporates the social ne

The 12th annual Texas Wireless Summit continues the tradition of providing a forum for industry leaders and academics to discuss emerging technologies and business models that will shape the industry over the upcoming two to three years. Co-hosted by the Austin Technology Incubator and The University of Texas at Austin’s Wireless Networking and Communications Group (WNCG), the Summit has direct access to cutting edge research and innovations from industry leaders, investors, academics, and startups.

In high speed network measurement, such as in core Internet routers, there may only be a few nanoseconds available per packet for measurement purposes.

This work studies the problem of sequentially recovering a sparse vector x_t and a vector from a low-dimensional subspace l_t from knowledge of their sum m_t=x_t+l_t. If the primary goal is to recover the low-dimensional subspace where the l_t's lie, then the problem is one of online or recursive robust principal components analysis (PCA). An example of where such a problem might arise is in separating a sparse foreground and a slowly changing dense background in a surveillance video.

Topic to be determined.

Several general-purpose deterministic global optimization algorithms have been developed for mixed-integer nonlinear optimization problems over the past two decades. Central to the efficiency of such methods is their ability to construct sharp convex relaxations. Current global solvers rely on factorable programming techniques to iteratively decompose nonconvex factorable functions, until each intermediate expression can be outer-approximated by a convex feasible set.

Traditionally, noise in communication systems has been modeled as an additive, white Gaussian noise process with independent, identically distributed samples. Although this model accurately reflects thermal noise present in communication system electronics, it fails to capture the statistics of interference and other sources of noise, e.g. in unlicensed communication bands.

Online social networks like Facebook, Twitter, Quora, etc have greatly influenced our social lives. This effect extends to economic and political realms, where social networks have become the best media for targeted campaigns of products and political ideas. From these perspectives as well as from the perspective of social sciences understanding the evolution of opinions and its effect on societies is of great importance. In the first part of this proposal we study opinion evolution in a social network.

A basic problem in the design of privacy-preserving algorithms is the private maximization problem: the goal is to pick an item from a universe that (approximately) maximizes a data-dependent function, all under the constraint of differential privacy. This problem is central to many privacy-preserving algorithms for statistics and machine learning.Previous algorithms for this problem are either range-dependent---i.e., their utility diminishes with the size of the universe---or only apply to very restricted function classes.  Prof.