Feeds:
Posts
Comments
=================================================

Sunday January 17,  10:15-11:45 Feldman building, Middle floor, Hall 130

When should  an expert make a prediction?

Speaker: Amir Ban, The Blavatnik School of Computer Science, Tel Aviv University

==================================================

Abstract:
We consider a setting where in a future known time, a certain continuous variable will be realized. There is a public prediction that converges to its value, and an expert has access to a more accurate prediction. Our goal is to study when should the expert reveal his information, assuming that his reward is based on a logarithmic market scoring rule (i.e.,  his reward is proportional to the gain in log likelihood of the realized value).
Our contributions are: (1) we show that the optimal expert policy is threshold based. (2) we analyze the expert’s asymptotic optimal reward and show a tight connection to the law of the iterated logarithm, (3) we give an efficient dynamic programming algorithm to compute the optimal policy.

 Joint work with Yossi Azar, Yishay Mansour.

Web site: https://compeconseminar.wordpress.com/
Mail list: http://listserver.cc.huji.ac.il/mailman/listinfo/comp-econ
Calendar: http://tinyurl.com/qj8pwha

 

Advertisements
==================================================
Sunday January 10, 10:15-11:45,  Feldman building, Top floor, (Inner) seminar room
Generalized Third-price Auctions
Speaker: Biligbaatar (Bilig) Tumendemberel, The Hebrew University
==================================================
We study an auction mechanism – Generalized Third-price (GTP) Auction – that could be used by search engines to sell online advertising. The properties of GTP are investigated in this paper in comparison to practically used auction mechanisms, GSP and VCG.
Joint work with Yair Tauman
Web site: https://compeconseminar.wordpress.com/
Mail list: http://listserver.cc.huji.ac.il/mailman/listinfo/comp-econ
Calendar: http://tinyurl.com/qj8pwha

=============================================
Sunday January 3, 10:15-11:45 Feldman building, Middle floor, Room 115

Blockchain Protocols and Social Choice

Speaker: Yonatan Sompolinsky, The Hebrew University ==================================================

Blockchain Technology is a disruptive invention in which an open, decentralized, and anonymous system authorizes data using a public ledger – called the Blockchain. Bitcoin is the first and most notable use-case of this concept, using a blockchain to record bitcoin transactions. The prime challenge of such systems is to maintain consistency within the ledger. In Bitcoin, this challenge is met by maintaining a single chain of transaction-batches, called blocks, which do not conflict with one another. This restriction on the ledger introduces severe scalability limitations.
In this work we propose an approach which is both scalable and secure. First, we incorporate in the ledger the entire history of blocks, forming the data in a DAG structure. Second, we observe that any linear ordering of the DAG induces a natural conflict resolution rule. We then discuss what makes an ordering rule secure, and demonstrate how social choice can be used to formalize this challenge. Finally, we describe one specific implementation of this concept, resulting in a provably-secure blockchain protocol. Our new protocol outperforms the Bitcoin protocol by achieving faster confirmation rates and higher transaction throughput.

Joint work with Yoad Lewenberg and Aviv Zohar.

Web site: https://compeconseminar.wordpress.com
Mail list: http://listserver.cc.huji.ac.il/mailman/listinfo/comp-econ
Calendar: http://tinyurl.com/qj8pwha

================================================

Sunday December 27, 10:15-11:45 Feldman building, Middle floor, Hall 130

The Speed of Social Learning

Speaker: Omer Tamuz, Caltech ==================================================

We consider Bayesian agents who learn from exogenously provided private signals, as well as the actions of the others. We show that learning from actions is slower than learning from signals, that increased interaction between agents can lower the speed of learning, and that very large groups do not learn very quickly.

Joint with Matan Harel, Elchanan Mossel and Philipp Strack

Web site: https://compeconseminar.wordpress.com/
Mail list: http://listserver.cc.huji.ac.il/mailman/listinfo/comp-econ
Calendar: http://tinyurl.com/qj8pwha

================================================

Sunday December 20, 10:15-11:45 Feldman building, Middle floor, Room 115

Analyzing games with ambiguous players types using the MINthenMAX decision rule

Speaker: Ilan Nehama, The Hebrew University

==================================================

In many common interactive scenarios, participants lack information on other participants, and specifically on the preferences of other participants. This paper models an extreme case of this information absence which we term games with type ambiguity – scenarios in which the participant lacks even the information enabling him to form a belief on the preferences of others. Under type ambiguity, one cannot analyze scenarios using the Bayesian framework, which is the common practice for scenarios involving partial-information. Hence, one needs to model the participants as following a different decision model under ambiguity.
In this work, we present the MINthenMAX decision rule. This rule is a refinement of Wald’s MiniMax principle, which we show to be too coarse for games with type ambiguity. Moreover, we claim MINthenMAX is the finest refinement of the MiniMax principle that satisfies two necessary properties for games with type ambiguity. The prior-less analysis we use also follows the way mechanisms are analyzed in Computer Science – according to the worst-case scenario.
Last, we define the corresponding equilibrium concept MINthenMAX-NE, and demonstrate it by applying this concept on two common economic scenarios: Coordination games and Bilateral trade. We show that in both cases, pure MINthenMAX-NE always exist and analyze these equilibria.

Web site: https://compeconseminar.wordpress.com/
Mail list: http://listserver.cc.huji.ac.il/mailman/listinfo/comp-econ
Calendar: http://tinyurl.com/qj8pwha

=================================================
Sunday December 6, 10:15-11:45 Feldman building, Middle floor, Hall 130

The Deterministic Communication Complexity of Approximate Nash Equilibrium

Speaker: Omri Weinstein, Courant Institute, NYU
==================================================

A shorter more high-level version of this talk will be presented in
Computer Science colloquium
Mon, Dec 7, 14:00
Rothberg B220

Abstract:
We study the communication complexity of finding an *approximate* Nash equilibrium in a 2-player game where each player initially knows only his own payoff matrix. We prove the first (unbounded-round) communication lower bound for this problem, namely, that the deterministic communication complexity of finding a O(1/log^2 N)-approximate Nash equilibrium is at least N^{Omega(1)}. In contrast, the non-deterministic communication of this problem is only poly-logarithmic. Our result implies that any (deterministic) market dynamic that converges even to an approximately stable market state, requires polynomial communication. The heart of our result is an essentially tight communication lower bound on the geometric problem of finding an approximate fixed-point of a composition of two Lipschitz functions.

Joint work with Tim Roughgarden.

Web site: https://compeconseminar.wordpress.com/
Mail list: http://listserver.cc.huji.ac.il/mailman/listinfo/comp-econ
Calendar: http://tinyurl.com/qj8pwha

================================================
Sunday November 22, 10:15-11:45 Feldman building, Middle floor, Hall 130

Crowd learning without herding;  A mechanism design approach
 
Speaker: Motty Perry, The Hebrew University & University of Warwick
==================================================
 
Crowd funding, internet websites and health care are only a few examples of markets where agents make decisions not only on the basis of their own investigations and knowledge, but also on the basis of information about other agents’ actions, provided to them by some “central planner”. While such a reciprocal learning can be welfare improving, it may also lessen agents’ incentives to conduct their own investigation and may lead to harmful cascades. We study the planner’s optimal policy, regarding when and how much information to provide and show that it involves a delicate balance of mixing between hiding and revealing information.
 
With J Glazer and Ilan Kremer
 

Mail list: http://listserver.cc.huji.ac.il/mailman/listinfo/comp-econ
Calendar: http://tinyurl.com/qj8pwha

%d bloggers like this: