We study a multi-server model with n flexible servers and rn queues, connected through a fixed bipartite graph, where the level of flexibility is captured by the average degree, d(n), of the queues. Applications in content replication in data centers, skill-based routing in call centers, and flexible supply chains are among our main motivations. We focus on the scaling regime where the system size n tends to infinity, while the overall traffic intensity stays fixed. We show that a large…

Low-rank matrix regularization is an important area of research in statistics and machine learning with a wide range of applications. For example, in the context of Missing data imputation arising in recommender systems (e.g. the Netflix Prize), DNA microarray and audio signal processing, among others, the object of interest is a matrix (X) for which the training data comprises of a few noisy linear measurements. The task is to estimate X, under a low rank constraint and possibly additional affine…

Multi-Armed bandits are an elegant model of learning in an unknown and uncertain environment. Such models are relevant in many scenarios, and of late have received increased attention recently due to various problems of distributed control that have arisen in wireless networks, pricing models on the internet, etc. We consider a non-Bayesian multi-armed bandit setting proposed by Lai and Robbins in mid-80s. There are multiple arms each of which generates an i.i.d. reward from an unknown distribution. There are multiple…

We analyze large random two-sided matching markets with different numbers of men and women. We show that being on the short side of the market confers a large advantage, in a sense that the short side roughly chooses the matching while the long side roughly gets chosen. In addition, the matching is roughly the same under all stable matchings. Consider a market with n men and n+k women, for arbitrary 1 <= k<= n/2. We show that, with high probability,…

"Transitory" queueing systems, namely those that exist only for finite time are all around us. And yet, current queueing theory provides with very few models and even fewer tools to understand them. In this talk, I'll introduce a model that may be more appropriate in "transitory" queueing situations. I'll start by first talking about the ``concert arrival game" that we introduced. This model captures the tradeoff in decision-making when users must decide whether to arrive early and wait in a…

In many modern optimization problems, specifically those arising in machine learning, the amount data is too large to apply standard convex optimization methods. We'll discuss new optimization algorithms that make use of randomization to prune the data produce a correct solution albeit running in time which is smaller than the data representation, i.e. sublinear running time. We'll present such sublinear-time algorithms for linear classification, support vector machine training, semi-definite programming and other optimization problems. These new algorithms are based on…

Multidimensional semimartingale reflecting Brownian motions (SRBMs) arise as the diffusion limits for stochastic networks. I will describe a powerful methodology to obtain the tail asymptotics of the stationary distribution of an SRBM. The methodology uses a moment generating function version of the basic adjoint relationship that characterizes the stationary distribution. The tail asymptotics can be used to predict quality of service in stochastic networks. It can also be used to speed up an algorithm, devised in Dai and Harrison (1992),…

We analyze the problem of partitioning a 0-1 array or bipartite graph into subgroups (also known as co-clustering), under a relatively mild assumption that the data is generated by a general nonparametric process. This problem can be thought of as co-clustering under model misspecification; we show that the additional error due to misspecification can be bounded by O(n^(-1/4)). Our result suggests that under certain sparsity regimes, community detection algorithms may be robust to modeling assumptions, and that their usage is…

We propose a new method named calibrated multivariate regression (CMR) for fitting high dimensional multivariate regression models. Compared to existing methods, CMR calibrates the regularization for each regression task with respect to its noise level so that it is simultaneously tuning insensitive and achieves an improved finite sample performance. We also develop an efficient smoothed proximal gradient algorithm to implement it. Theoretically, it is proved that CMR achieves the optimal rate of convergence in parameter estimation. We illustrate the usefulness…

Random phenomena are ubiquitous throughout science and engineering. Beyond the study of stochastic models in their own right, it is of importance in many applications to understand what information can be extracted on the basis of observed data. Mathematically, such ``data assimilation'' problems motivate the investigation of probabilistic phenomena that arise from conditioning. This topic has connections to several areas of probability, ergodic theory, measure theory, and statistical mechanics, as well as practical implications for the design and analysis of…