Past Events

Wireless networks are inherently limited by their own interference. Therefore, a lot of research focuses on interference reduction techniques, such as mutiuser MIMO, interference alignment, interference coordination or multi-cell processing. Although these techniques might lead to considerable performance gains, it is unlikely that they will be able to meet the demand for wireless data traffic in the future. Therefore, a significant network densification, i.e., increasing the number of antennas per unit area, is inevitable.

With the exponential increase in high rate traffic driven by a new generation of wireless devices, data is expected to overwhelm cellular network capacity in the near future. Femtocell networks have been recently proposed as an efficient and cost-effective approach to provide unprecedented levels of network capacity and coverage.

Peer-to-peer networks are networks without a massive central server. Instead, each peer in the network transmits the data it receives to a subset of other peers in the network, thus propagating information in the network.  We will illustrate the application of random graphs to the problem of designing simple distributed algorithms for peer-to-peer streaming applications (such as live video) that can achieve high throughput and low delay, while maintaining a very small neighbor set for each peer. The talk will be based on joint work with Joohwan Kim.

The problem of recovering a sparse signal from an underdetermined set of linear equations is paramount in many applications such as compressed sensing, genomics, and machine learning. While significant advances have been made in this area, providing useful insights and intuitions, many important questions are still open including the fundamental performance limits of the recovery algorithms.

I will discuss deep connections between Statistical Learning, Online Learning and Optimization.  I will show that there is a tight correspondence between the sample size required for learning and the number of local oracle accesses required for optimization, and thesame measures of "complexity" (e.g. the fat-shattering dimension or Rademacher complexity) control both of them.

Climate data presents unique challenges for machine learning due to its spatiotemporal nature and high-dimensionality. In this talk, I will discuss two applications of high-dimensional modeling for climate data analysis. The first application is on abrupt change detection, with emphasis on detecting significant droughts in the past century. The problem is formalized as a graph-structured linear program (GSLP), and solved using KL-ADM, a novel parallel inexact alternating directions method with Bethe entropy based augmentation. KL-ADM is provably guaranteed to solve GSLPs, and is

This talk will deal with the notions of adaptive and non-adaptive information, in the context of statistical learning and inference.  Suppose that we have a collection of models (e.g., signals, systems, representations, etc.) denoted by X and a collection of measurement actions (e.g., samples, probes, queries, experiments, etc.) denoted by Y. A particular model x in X best describes the problem at hand and is measured as follows.  Each measurement action, y in Y, generates an observation y(x) that is a function of the unknown model.

This talk will deal with the notions of adaptive and non-adaptive information, in the context of statistical learning and inference.  Suppose that we have a collection of models (e.g., signals, systems, representations, etc.) denoted by X and a collection of measurement actions (e.g., samples, probes, queries, experiments, etc.) denoted by Y. A particular model x in X best describes the problem at hand and is measured as follows.  Each measurement action, y in Y, generates an observation y(x) that is a function of the unknown model.

Distributed storage systems (DSS) are instrumental for providing reliable storage solutions to satisfy ever growing data demands. DSS stores the content over a network of nodes, and achieves resilience against node failures by maintaining  data redundancy via various coding techniques. When some DSS nodes fail, this coding allows for retrieving the original data from the remaining nodes, as long as there is sufficient number of nodes left.

no results