8:00–8:45am |
Light Breakfast |
8:45–9:00am |
Opening Remarks |
9:00–9:45am |
 Reaching a Rational Expectations Equilibrium Among a Group of AgentsPravin Varaiya |
| Abstract: Each agent i = 1,··· ,N seeks to estimate the expected value of a random variable X. Agent i’s estimate at time t is denoted Xi(t). This estimate is based on i’s initial private observation zi and and messages mj(τ),1 ≤ τ ≤ t −1 broadcast by agent j to the entire group. Thus Xi(t) = E{X | zi N, τ ≤ t −1}. A random variable X for all i. The existence and uniqueness of a REE depends on the structure of the broadcast messages. Borkar-Varaiya studied the situation wherein mi(τ) = Xi(τ) and when all agents have the same prior. Teneketzis-Varaiya examined the same setting but permitted different priors. In this discussion we consider other signal structures. | |
9:45–10:30am |
 Mobile Networks Evolution from LTE to 5G: Algorithms and ArchitectureRajeev Agrawal |
| Abstract: In this talk we will provide an overview of the recent developments in the mobile networks industry and also the upcoming opportunities and challenges with the move towards 5G. We will focus on the radio access part of the network and highlight the algorithmic and architectural solutions central to some of the key problems areas covering spectrum, site densification, and spectral efficiency improvements. Through the talk we will illustrate issues of centralization versus decentralization, (stochastic) optimization and control, learning versus optimization - key themes of Prof. Teneketzis' research. | |
10:30–11:00am |
Break |
11:00–11:45am |
 The Prices of Packets and Watts: Optimal Control of Decentralized SystemsPR Kumar Professor and College of Engineering Chair in Computer Engineering Dept of Electrical and Computer Engineering, Texas A&M University |
| Abstract: We examine two problems of contemporary interest, both involving the optimal operation of distributed stochastic dynamic systems, one in power systems and another in communication networks.
In power systems we address the problem faced by an Independent System Operator which has to optimally utilize a mix of fossil fuel based and renewable generators, controllable and uncontrollable loads, prosumers and storage services, so as to maximize social welfare, without knowing details of any of the entities involved. We revisit general equilibrium theory, addressing the complexity raised by dynamic stochastic agents with privacy concerns. We show how optimality can be attained through stochastic prices discovered by iterative tatonnement mechanisms. In communication networks we address the problem of how to optimally schedule multiple flows with hard end-to-end deadlines over a network of unreliable links with average power constraints so as to maximize their throughput. We exhibit an exactly optimal policy for this distributed system that employs easily determined prices to distributedly schedule link transmissions throughput the entire network. [Joint work with Rahul Singh and Le Xie] |
|
11:45–12:30pm |
 Interaction of Controllers in Decentralized ControlJan van Schuppen |
| Abstract: The control theoretic problem of decentralized control is to synthesize two or more controllers which together achieve control objectives of performance and of stability. Effective examples of control engineering and of communication engineering provide motivation for the research on decentralized control. Demosthenis Teneketzis and his students have substantially contributed to the theory of decentralized control partly based on the fundamental framework developed by H.S. Witsenhausen. However, the theory of decentralized control is far from complete.
In this lecture the focus is on the interactions of the controllers via the control system, for which concepts and theory are needed. Research issues include: What information should a controller communicate to which other controllers via the control system? How is this communication best carried out? How should a controller use information received from other controllers to attain the control objectives? Concepts of a system theory of decentralized control systems leading to system decompositions and based on conditional independence and on information theory, will be formulated. The theory of decentralized control can be advanced using these concepts. The research is partly based on cooperation with Cesar Uribe. |
|
12:30–2:00pm |
Lunch (provided) |
2:00–2:45pm |
 Backoff Algorithms for Random Access ProtocolsNah-Oak Song |
| Abstract: Random access protocols are widely used in wireless and wired networks where centralized resource scheduling is not feasible. Random access schemes use backoff algorithms for contention resolution. In this talk, analytical results for exponential backoff (EB) algorithm are presented. EB has been extensively studied, but most available studies on EB focus on the stability and little attention has been paid to the performance of EB. We present analytical saturation throughput and medium access delay of a packet when the number of nodes N is given. The analysis considers the general case of EB with backoff factor r, and the optimal backoff factor r opt is derived. Binary exponential backoff (BEB) algorithm is a special case with backoff factor r=2. Also analyzed are truncated EB, which is a practical version of EB. The analysis technique is used to evaluate the performance of IEEE 802.11 DCF, a well known random access scheme for WLAN which uses BEB.
In the second half of the talk, we present exponential increase exponential decrease (EIED) backoff algorithm. The analytical performance of EIED is compared with BEB in the context of WLAN. The analytical and simulation results show that EIED significantly improves the performance of WLAN. |
|
2:45–3:30pm |
 Intersecting Random GraphsArmand Makowski |
| Abstract: Given two graphs $G_1 = (V,E_1)$ and $G_2 = (V,E_2)$ with same vertex set $V$, we define their intersection as the graph $(V,E)$ with vertex set $V$ and edge set $E = E_1 \cap E_2$. In this talk we consider the situation where each of the component graphs is itself a random graph, e.g.,a Bernoulli random graph, a geometric random graph or a random key graph. Random graphs obtained in this manner occur naturally as simple models for a number of issues in wireless networking.
We are interested in understanding how the structural properties of the component random graphs shape those of the random graph obtained by intersection. Particular attention will be given to the existence of zero-one laws for graph connectivity and for the property that there are no isolated nodes, and the form of the associated critical scalings. |
|
3:30–4:00pm |
Break |
4:00–4:45pm |
 Scheduling Problems in NFVManjari Asawa |
| Abstract: Network Functions Virtualization (NFV) is an evolving network architecture to bring advantages of virtualization to telecommunications operations. NFV will allow telecommunication operators to replace special purpose networking equipment with commodity servers running virtualized network functions (VNFs). Shifting dedicated, hardware-based network function processing to software running on commoditized hardware has the potential to enable flexible and cost-effective provisioning of network functions. This does come with additional complexity and presents new operational and management challenges. This talk will provide an overview of ETSI NFV architecture and the effective placement and scheduling issues in the context of virtual IP Multimedia Service (vIMS) use case. We will also discuss some of the work being done by open-source organizations, industry and academia as well as the complexity and need of an end-to-end framework for efficient scheduling of VNFs onto physical infrastructure while meeting end-to-end use experience requirements. | |
4:45–5:30pm |
 Dynamic Stochastic Network Interdiction GamesDavid Castañón |
| Abstract: Network interdiction problems consist of zero-sum games between an attacker and an intelligent network, where the attacker seeks to degrade network operations while the network adapts its operations to counteract the effects of the attacker. This problem has received significant attention in recent years due to its relevance to network security in different contexts. When the attacker’s actions achieve uncertain effects, the resulting problems become stochastic network interdiction problems. In this talk, we discuss recent results in network interdiction problems when attacks occur over a sequence of times. In particular, we discuss network interdiction games when there is an asymmetry of information concerning the network state between the attacker and the defender, and present algorithms for the solution of the resulting dynamic stochastic games. | |
6:30pm |
Dinner at the Gandy Dancer (401 Depot Street, Ann Arbor)
Directions
(Google Maps) (Map and Written Directions) |
8:00–9:00am |
Light Breakfast |
9:00–9:45am |
 Service-State Based SchedulingMark Andersland |
| Abstract: The scheduling of competing task flows' unit-sized requests for service
from a shared, slotted-time, constant capacity server is a standard abstraction of many resource allocation problems. In deterministic settings, the typical aim is to schedule service to maintain per-flow, worst-case guarantees. In this talk we outline how, when the guarantees are viewed as states that evolve as tasks arrive and are served, the scheduling problem can be reduced to that of scheduling service so as to maintain, slot by slot, the collective service state in an acceptable set. Additionally we characterize, slot by slot, the schedules capable of maintaining the collective service state in this set. This perspective exposes, in a setting rich enough to include both H-FSC and time-varying SCED scheduling, tradeoffs among all policies capable of maintaining a particular set of guarantees.
(This is based on joint work with Yike Xu) |
|
9:45–10:30am |
 Strategic Information TransmissionTamer Başar |
| Abstract: Strategic information transmission refers to a variation (and a substantial one) of the standard paradigm of information transmission in communication (design of an encoder and a decoder in unison to minimize some distortion measure), where now the encoder and the decoder have (intentionally) misaligned objectives. This leads to a non- cooperative game with a dynamic (non-classical) information structure, where one can adopt as a solution concept either the Nash or the Stackelberg equilibrium. This talk will introduce this class of problems, which have been of interest to multiple communities, including economics, information theory, communication, signal processing, and control, since the early 1980s, having picked up considerable steam very recently. As an overview of the topic, both old and new results will be presented, with one of the highlights (and perhaps a surprising element) being that there appears to be a major difference between the structures of the encoders and the decoders under Nash and Stackelberg equilibria, even when the channel is Gaussian and the (misaligned) distortion measures are quadratic.
(Joint work with Emrah Akyol and Cedric Langbort) |
|
10:30–11:00am |
Break |
11:00–11:45am |
 Multi-Agent Collaborative Decision Making: Constrained Event Algebras, New Logics and Their Game Theoretic SemanticsJohn Baras |
| Abstract: We briefly review various instances of non-commutative probability models from quantum mechanics, directed information, nonlinear control systems, Web-based inference, multi-agent systems, collective human cognition and decision making from psychology. We then investigate how the probability models in these areas could be constructed from observational data following a "Kolmogorov-like" construction. We develop an axiomatic method for such a construction that reveals some new foundational principles including: multiple non-collocated observers, existence of incompatible events, notions of information sharing, directed information, probabilities on constrained event algebras. We next describe independence-friendly logic (IF Logic) and its multi-agent game theory semantics. We close with our recent work on an approach for unifying these various non-commutative probability models and their utilization. | |
11:45–12:30pm |
 A Relative Entropy Characterization of the Growth Rate of Reward in Risk-Sensitive ControlVenkat Anantharam |
We study the infinite horizon risk-sensitive control problem for discrete time Markov decision processes with compact metric state and action spaces. We derive a variational formula for the optimal growth rate of risksensitive reward. This parallels the usual variational formulation of the long term average reward in the absence of risksensitivity given by the ergodic control viewpoint, with an additional relative entropy penalty on occupation measures. It can also be viewed as an extension of the characterization of Donsker and Varadhan for the PerronFrobenius eigenvalue of a positive operator. The problem of determining this optimal growth rate of risksensitive reward is thereby presented as a problem of maximizing a concave function over a convex set. (Joint work with Vivek Borkar, IIT Bombay) |
|
12:30–2:00pm |
Lunch (provided) |
2:00–2:45pm |
 Optimal Server Allocation in Tandem SystemsDimitris Pandelis |
| Abstract: We consider two-stage tandem systems (both clearing systems and with Poisson arrivals) with one dedicated server in each station and one flexible server that can serve both stations. We assume exponential service times and linear holding costs for jobs present in the system. We study the optimal dynamic assignment of servers to jobs in the class of preemptive and non-preemptive policies.
Preemptive policies. We assume that two servers can collaborate to work on the same job with a total service rate that is less than the sum of their individual service rates (partial collaboration). For larger holding costs in the first station we show that (i) non-idling policies are optimal and (ii) identify conditions under which the optimal allocation strategy for the flexible server has a threshold-type structure and a cμ-type rule is optimal. For larger holding costs in the second station we characterize the optimal policy for a large number of jobs downstream. Finally, we show by examples that the optimal policy may have counterintuitive properties, which is not the case when full collaboration is assumed. Non-preemptive policies. We assume that collaboration is not allowed and that the flexible server is slower than each of the dedicated servers (a two-stage version of the slow server problem). We show that the dedicated server in station 2 should always work and the same is true for the dedicated server upstream for larger holding costs there. As for the optimal allocation of the slow server we conjecture that it is characterized by two thresholds. When the slow server is constrained to work in one station, numerical experiments suggest that the optimal policy is closely related to that of the classical slow server problem. |
|
2:45–3:30pm |
 When we Should Trust Self-Driving CarsManju Hegde |
| Abstract: The automotive industry, one of the workhorse segments of the world’s economy, is undergoing a sea change as the automotive industry embraces Advanced Driver Assisted Systems (ADAS) and Self-Driving Cars. Safety and robustness are the hallmarks of automotive engineering and in order to meet this criteria for ADAS and self-driving car, multiple sensor systems are being utilized in cars gathering data about the car’s environment at a prodigious rate. This data from the sensors must be fused at multiple levels in order to prevent confounding of the decisions and to be conformant with the requirements of automobile OEMs. In addition, once the fused data is understood, inferences must be made about the actions of the other “drivers” and concomitant actions in order to ensure safe driving.
In this talk we talk about these trends in the automotive industry and pose problems that need to be resolved for this self-driving paradigm to emerge and be accepted as a mainstream "driving" experience. |
|
3:30–4:15pm |
 101 Definitions of Diagnosability or Lack ThereofStéphane Lafortune |
| Abstract: We will review the work of Sampath et al. on diagnosis of discrete event systems, focusing on the definition of diagnosability and of its many variants that have appeared in the literature over the last 20 years. Then we will describe the property of opacity, which is receiving considerable attention lately in security and privacy research. Opacity is almost, but not quite, the negation of diagnosability. | |
4:15–4:30pm |
Closing |