Talk Abstracts On the Proactive Resource Allocation Paradigm: Prediction, Servicing, and Shaping, Atilla Eryilmaz Many real-world systems, such as wireless communication networks, power grids, etc., include components with statistically predictable characteristics. For example, the future mobility and demands of wireless users exhibit highly predictable and usually repetitive behavior. In contrast to the opportunistic and reactive resource allocation approaches that dominate the research efforts, in this work we explore the "proactive" resource allocation paradigm that incorporates the statistical prediction of future system resources and user demands along with the service costs. We will present the models, challenges, and solutions that our recent works have produced in this direction. Moreover, we will share our findings on the means of utilizing users' elasticity to recommendations and/or prices in order to shape their demand, thereby increasing the proactive service gains while benefiting the service provider and the users simultaneously. Joint Work With: John Tadrous, and Hesham El Gamal A Mean Field Game Approach to Promoting Collaboration in Device-to-Device Networks, Srinivas Shakkottai Hand held wireless devices often have diverse communication interfaces that can be used simultaneously, such as expensive (energy and dollars) long-range cellular data interfaces (3G/4G) and inexpensive short-range interfaces (WiFi/Bluetooth). In order to reduce cellular data costs, we design a content streaming system in which the cellular interface is used for communication with a server, while the short-range interface is used for device-to-device (D2D) communication. We show how D2D communication can significantly reduce the 3G/4G costs for the system while all users can still achieve a good quality of service. Using the idea of mean field equilibrium (MFE), we then design an incentive framework wherein selfish users are motivated to share content with each other. Joint work with Jian Li, Rajarshi Bhattacharyya, Suman Paul and Vijay Subramanian. Rethinking Video Transport: Quality of Experience meets Multi-user Rate Adaptation, Gustavo de Veciana User perceived video quality depends on a variety of only partially understood factors, e.g., the application domain, content, compression, transport mechanism, and most importantly psycho-visual systems determining the ultimate Quality of Experience (QoE) of users. This talk centers on two key observations in addressing the problem of joint rate adaptation for video streams sharing a congested resource. First, we note that a user viewing a given video will experience temporal variations in the dependence of perceived video quality to the compression rate. Intuitively this is due to the possibly changing nature of the content, e.g., from an action to a slower scene. Thus, in allocating rates to users sharing a congested resource, in particular a wireless system where additional temporal variability in users' capacity may be high, content dependent tradeoffs can be realized to deliver a better overall average perceived video quality. Second, we note that such adaptation of users' rates, may result in temporal variations in video quality which combined with perceptual hysteresis effects will degrade users' QoE. We develop an asymptotically optimal online algorithm, requiring minimal statistical information, for optimizing users' QoE by realizing tradeoffs across mean, variance and fairness. Simulations show that our approach achieves significant gains in viewers' QoE. QoE-Optimal Scheduling for Wireless Networks, I-Hong Hou It has been shown that the quality of experience (QoE) of video streaming is most significantly influenced by the duration of video stops/pauses during the playback. We study the achievable durations of video stops/pauses for a wireless system with multiple video streams. We show that a weighted sum of stop/pause durations of all streams is stochastically lower-bounded by a random variable. We then propose a QoE-optimal scheduling policy that achieves this lower-bound. Both theoretical analysis and experiments show that our policy has much better performance than other policies. Transitory Queues, Rahul Jain We introduce a framework and develop a theory of 'transitory' queueing models. These are models that are not only non-stationary and time-varying but also have other features such as the queueing system operates over finite time, or only a finite population arrives. Such models are relevant in many real-world settings, from queues at post-offices, DMV, concert halls and stadia to out-patient departments at hospitals. We develop fluid and diffusion limits for a large class of 'transitory' queueing models. We then introduce three specific models that fit within this framework, namely, the (i)/GI/1 model, the conditioned G/GI/1 model, and an arrival model of scheduled trac with epoch uncertainty. We show that asymptotically these models are distributionally equivalent, i.e., they have the same fluid and diffusion limits. We note that our framework provides the first ever way of analyzing the standard G/GI/1 model when we condition on the number of arrivals. In obtaining these results, we provide generalizations and extensions of the Glivenko-Cantelli and Donsker's Theorems to triangular arrays. Our analysis uses a technique we call population acceleration, which we discuss in some detail. Accelerated Second Order Methods for Deterministic and Stochastic Network Optimization, Ali Jadbabaie In this talk I will present a summary of our recent work on second order methods for convex network flow and network rate optimization problems. The proposed approach uses distributed algorithms for computation of exact and approximate hessian inverses to achieve faster convergence rates than existing distributed methods. The approximate Newton directions are obtained through matrix splitting techniques and Taylor approximations of the inverse Hessian. We couple this descent direction with a distributed line search algorithm which requires the same information as our descent direction. We show that, similarly to conventional Newton methods, the proposed algorithm exhibits super-linear convergence within a neighborhood of the optimal value. Numerical experiments corroborate that convergence times are between one to two orders of magnitude faster than existing distributed optimization methods. Next, we extend these methods to develop an Accelerated Back Pressure (ABP) algorithm for joint routing and scheduling in packet networks. Joint work with Michael Zargham(Penn), Alejandro Ribeiro (Penn), Ermin Wei(MIT), and Asu Ozdaglar(MIT) Optimal Control of Wireless Networks: From Theory to Practice, Eytan Modiano This talk reviews recent advances on network control for wireless networks with stochastic traffic and time-varying channel conditions. We start with a review of the seminal work of Tassiulas and Ephremides on optimal scheduling and routing, i.e., the now famous backpressure algorithm. Despite its theoretical promise, the optimal control strategy has not taken hold in practice; due, in part, to some of the modeling assumptions that fail to take into account practical considerations. Thus, we will discuss recent efforts to develop variants of backpressure that take into account practical considerations. These include efficient distributed scheduling algorithms, as well as new algorithms that take into account practical hardware and protocol limitations. Locating the Source of Cascaded Information in Networks, Lei Ying Who leaked classified information or started a rumor on online social networks? Who uploaded contraband materials to the Internet? Where are the sources of epidemics? These questions are of great importance to the worlda?#8482;s safety and security, but are difficult to answer. In this talk, I will introduce a sample path based approach, which is to identify the most likely sample path and view the source of the optimal sample path as the information source; and present some recent results based on the sample path based approach. Observational Learning with Errors, Randall Berry Observational learning is one of the leading frameworks for studying how agents aggregate information in an underlying social network. In such models, Bayesian agents make decisions based on their own private signals and the observed actions of a subset of the other agents. In this talk we consider the impact of two types of errors on such models, action errors, where agents sometimes choose incorrect actions, and observation errors, where agents sometimes observe the action of other agents incorrectly. This is joint work with Tho Le and Vijay Subramanian. Topological Interference Management, Syed Jafar While much progress has been made in our understanding of the capacity limits of interference networks through degrees of freedom studies, the insights are often limited by the overly optimistic assumption of abundant channel knowledge and the overly pessimistic assumption of a fully connected network where all interfering signals are comparable in strength to the desired signal. Topological interference management is an alternative problem formulation where the channel knowledge is limited to the network topology, which refers to a coarse knowledge of relative channel strengths. When the topology knowledge is only based on the distinction of strong versus weak channels, the topological interference management problem is shown to be essentially the index coding problem.