Invited talk by Dr. Sujit Gujar from École polytechnique fédérale de Lausanne (EPFL).on Mechanism Design for heterogeneous resource allocation with strategic agents

Title of the talk: Mechanism Design for heterogeneous resource allocation with strategic agents
Speaker:Dr. Sujit Gujar
TimeJuly 6th (Mon) at 11:30 a.m
Venue: Room 121
Host Faculty: Dr. Aravind NR

Abstract:
Mechanism design is an important game theory based tool in microeconomics which has found widespread applications in many modern engineering and computer science applications.Mechanism design is concerned with settings where a social planner faces the problem of aggregating the announced preferences of multiple agents into a collective decision when the  agents exhibit strategic behaviour The mechanism designer achieves desired optimality by inducing a game among agents such that, at equilibrium, the strategic agents behave honestly.
In this talk we describe two mechanism design problems and our proposed solutions.
(1) In many real world situations, the mechanism designer's goal is to achieve social efficiency in resource allocation rather than revenue optimization. This calls for designing a redistribution mechanism,a sub-class of celebrated Groves mechanisms. If the objects to be allocated are heterogeneous (rather than homogeneous), the information that agents possess, needs to be represented in multi-dimensional spaces. This makes the allocation problem non-trivial.We design a novel redistribution mechanism that is optimal for allocation of heterogeneous objects with unit demand.

(2) It may be desirable  in a combinatorial exchange to preserve the privacy of the bids as well as the bidding details from the remaining agents in order not to reveal sensitive information regarding business strategies. The state of the art in secure auctions does not address preserving the privacy of the fact of bidding itself. We design a privacy preserving protocol for combinatorial exchanges. We show that how mechanism design theory could be used in conjunction with privacy preserving distributed constrained optimization to build secure combinatorial exchanges.

Brief Biodata Of the Speaker:

  • Dr Sujit Gujar is currently a post doctoral researcher in Artificial Intelligence Laboratory at École polytechnique fédérale de Lausanne (EPFL). His research broadly falls in the intersection of Economics and Computer Science (EconCS). 
  • His main research interests are algorithmic game theory and mechanism design. Prior to joining EPFL, he has worked in Xerox Research Centre India as a research scientist.
  • He has defended his doctoral thesis in Department of Computer Science and Automation at Indian Institute of Science. He was awarded a alumni medal for the best doctoral thesis in the department of Computer Science and Automation for academic year 2011-12.
  • He was recipient of Infosys fellowship.
  • He has 23 publications in national and international conferences and journals. He has filed 7 patents and 9 under submission.
 
Dates: 
Monday, July 6, 2015 - 11:30