80 likes | 173 Views
Some overarching themes on electronic marketplaces. Tuomas Sandholm Computer Science Department Carnegie Mellon University. Need to simultaneously handle:. Strategyness “CS issues” Computation complexity Communication complexity Privacy (amount of preference info revealed, and to whom).
E N D
Some overarching themes on electronic marketplaces Tuomas Sandholm Computer Science Department Carnegie Mellon University
Need to simultaneously handle: • Strategyness • “CS issues” • Computation complexity • Communication complexity • Privacy (amount of preference info revealed, and to whom)
Some high-level goals • Design electronic market mechanisms that lead to economically efficient outcomes via efficient processes • Design software agents that act optimally (subject to their computational limitations) on behalf of the users that they represent • => move strategic & computational complexity from human to machine
Perspectives • Leverage increasing computing (and communication) power to increase economic efficiency • E.g. running more complex mechanisms • E.g. automatically designing the mechanism • Capitalize on the piles of information that the parties have about each other - unlike traditional market mechanisms • Tackle the new issues that ecommerce has brought about, such as anonymity & cheap pseudonyms (lack of personal relationships and legal enforcement between transaction parties) • E.g. safe exchange mechanisms • E.g. reputation servers • E.g. false-name proof combinatorial markets
Interplay of computing & incentives Computational complexity of • Executing a mechanism (rules of the game) • Determining the winners in • Combinatorial auctions [Sandholm IJCAI-99, AIJ-02, AIJ-03, …] • Voting [Bartholdi, Tovey, Trick 89, …] • Executing a strategy • How should computationally bounded agents play strategically? • Limits on memory [Rubinstein, Papadimitriou, Kalai, Gilboa, …] • Costly or limited computing [Sandholm ICMAS-96, IJEC-00; Larson&Sandholm AIJ-01, AAMAS-02, TARK-01, AGENTS-01 workshop, …] • Manipulating a mechanism (by determining a beneficial insincere revelation) [Conitzer&Sandholm AAAI-02, IJCAI-03, TARK-03, …] • Determining which agents’ preferences should be elicited [Conitzer&Sandholm AAAI-02] • Determining how to play (finding the game’s equilibrium) [Gilboa&Zemel GEB-89, Conitzer&Sandholm IJCAI-03, …] • Finding a payoff division (e.g., according to the core) [Conitzer&Sandholm IJCAI-03, …] • Designing a mechanism automatically [Conitzer&Sandholm UAI-02, ACM-EC-03, …] • Perhaps without knowledge of the prior (or at least not complete knowledge)
Expressiveness => economic efficiency computational complexity of clearing no need for lookahead • Clearing markets with expressive bidding • Bidding with supply/demand curves in multi-unit market • Package bidding • Side constraints • Multiple attributes • How to make combinatorial exchanges fast ? • Approximate clearing while maintaining incentive properties?
Other promising new directions • Online clearing • Preference elicitation (& other multi-stage mechanisms) • Mechanisms with insincere equilibrium play
Randomization can help • To keep the adversary at bay • Universal revelation reducers • Online clearing • To keep agents at bay (= yield better mechanisms) • Automated mechanism design • To reduce computational complexity when desired • Designing a randomized mechanism is in P • To increase computational complexity when desired • Randomized cup voting protocol is harder to manipulate than the cup • In voting protocol tweaks paper, if preround pairing is randomly selected after votes are collected, then manipulation is #P-hard instead of NP-hard • Correlated uncertainty about other voters serves the same role as weighted coalitional voters => can get hardness even for a constant number of candidates