Northwestern University

May
19
Fri 2:00 PM

## EECS Seminar Speaker: Prof. Achilleas Anastasopoulos, Associate Professor of EECS, University of Michigan, "Incentives for Network Allocation with Strategic Agents: Informational Constraints & Learning Guarantees"

When: Friday, May 19, 2017
2:00 PM - 3:00 PM

Where: Technological Institute, Room L440, 2145 Sheridan Road, Evanston, IL 60208 map it

Audience: Faculty/Staff - Student - Public - Post Docs/Docs - Graduate Students

Contact: Lana Kiperman   847.467.0028

Category: Lectures & Meetings

### Description:

The EECS Department welcomes Prof. Achilleas Anastasopoulos, Associate Professor of EECS, University of Michigan.

Anastasopoulos will present a talk entitled "Incentives for Network Allocation with Strategic Agents: Informational Constraints & Learning Guarantees", on Friday, May 19 at 2:00 PM in the Tech Room L440.

Abstract: Allocation of scarce resources arises in several problems in the general area of networks, including rate allocation in unicast/multicast services on the Internet, transmission-power allocation in wireless networks, spectrum allocation in cellular networks, as well as power/energy production and distribution in the power grid. Incentive designs (i.e., Mechanisms) that implement such allocations in the presence of strategic agents have been studied extensively but they invariably suffer from either or all of the following problems:

a) they are usually presented in a case-by-case fashion, thus obscuring fundamental understanding of these problems;
b) they assume agents’ messages (e.g., bids) are broadcasted throughout the entire network;
c) they focus on the final equilibrium without investigating the process that leads to such equilibrium.

In the first part of this talk we propose a unified semi-systematic methodology for creating indirect mechanisms that fully implement, in Nash equilibria (NE), social utility maximizing functions arising in various contexts with convex constraints. Two additional design goals are the focus of this design: 1) the size of the message space scaling linearly with the number of agents (even if agents' types are entire valuation functions), and 2) allocation being feasible on and also off equilibrium, i.e., during the learning phase of the game.

In the second part of this talk we present mechanisms under the assumption that agents can communicate only through a given network and thus the designed mechanism obeys the agents' informational constraints. This means that each agent's outcome through the mechanism can be determined by only the messages of his/her neighbors. Finally, it is guaranteed that agents can learn the NE induced by the mechanism through repeated play when each agent selects a learning strategy from within the adaptive best-response'' dynamics class. This is a class of adaptive learning strategies that includes well-known dynamics such as Cournot best-response, k-period best-response and fictitious play, among others.

Bio: Prof. Achilleas Anastasopoulos is an Associate Professor of EECS at University of Michigan. His research focuses on Control theory with emphasis in Stochastic sequential teams and their applications in Communication and Information theoretic problems. He also researches dynamic games with asymmetric information and Mechanism design, as well as, Information theory with emphasis in Fundamental QoS limits in multiuser environments. Other examines communication theory with emphasis in design of capacity-achieving transmission schemes for noisy channels and code design and analysis for wireless channels.

Hosted by: Prof. Randy Berry