We had a large contingent at the Economics and Computation conference (formerly Electronic Commerce, sigecom.org/ec14/) held at Stanford University. This is the premier conference for We had a large contingent at the Economics and Computation conference (formerly Electronic Commerce, sigecom.org/ec14/) held at Stanford University. This is the premier conference for applied game theory and internet economics and covers topics that lie at the intersection of computer science and economics like auctions and mechanism design. We gave 4 presentations during the workshops and had two papers accepted to the main session. The EC conference traditionally consists of a mix of theory, AI, and empirical/experimental papers, with theory papers making up the majority. This year was no exception: 75% of accepted papers were labeled theory, about 25% were labeled AI, and about 25% were labeled experimental/empirical (sigecom.org/exchanges/volume_13/1/CONITZER.pdf). However, one trend this year seemed to be an increased push on solving real-world economic problems over stylized models. Kevin Leyton Brown’s keynote presented work on “pragmatic algorithmic game theory”, which has two aims beyond standard algorithmic game theory: (1) achieve good performance on problems of interest, rather than in general settings, and (2) focus on statistical over analytical methods. The most notable example of pragmatic algorithmic game theory was the numerous talks this year on FCC incentive auctions (fcc.gov/incentiveauctions). These are two-sided auctions that will be run by the US government in 2016 to reallocate wireless spectrum from TV broadcasters to wireless carriers. In the inner-loop of the auction is a computationally hard problem reallocating spectrum of non-selling TV broadcasters so that the contiguous blocks of available spectrum is maximized while not violating quality guarantees for the non-selling broadcasters. Various papers across EC, NBER, and the Ad Auctions workshop focused on different aspects of the optimization and pricing problems. **Papers by Facebook Employees:** *Recency, Records and Recaps: Learning and Non-Equilibrium Behavior in a Simple Decision Problem*, Authors: Drew Fudenberg, Alexander Peysakhovich delivery.acm.org/10.1145/2610000/2602872/p971-fudenberg.pdf Abstract: People make mistakes. An important question is whether repeated experience with a decision will lead individuals to behave optimally. We argue the answer is: sometimes. We focus on renency bias, or the tendency to discount past information as a culprit for the persistence of mistakes. We show experimentally how this bias can lead to persistent sub-optimal decision-making. We also show how simple informational interventions or recaps can be used to mitigate the worst impacts of the bias. Our results highlight the importance of going beyond static equilibrium analysis in both descriptive and engineering applications of game theory. *Long-run Learning in Games of Cooperation*, Authors: Winter Mason, Siddharth Suri and Duncan Watts Abstract: Cooperation in repeated games has been widely studied in experimental settings; however, the duration over which players participate in such experiments is typically confined to at most hours, and often to a single game. Given that in real world settings people may have years of experience, it is natural to ask how behavior in cooperative games evolves over the long run. Here we analyze behavioral data from three distinct games involving 571 individual experiments conducted over a two-year interval. First, in the case of a standard linear public goods game we show that as players gain experience, they become less generous both on average and in particular towards the end of each game. Second, we analyze a multiplayer prisoner’s dilemma where players are also allowed to make and break ties with their neighbors, finding that experienced players show an increase in cooperativeness early on in the game, but exhibit sharper “endgame” effects. Third, and finally, we analyze a collaborative search game in which players can choose to act selfishly or cooperatively, finding again that experienced players exhibit more cooperative behavior as well as sharper endgame effects. Together these results show consistent evidence of long-run learning, but also highlight directions for future theoretical work that may account for the observed direction and magnitude of the effects. **Workshop Presentations by Facebook Employees:** Winter Mason served on the organizing committee of the Second Annual Workshop on Crowdsourcing and Online Behavioral Experiments (decisionresearchlab/cobe/). Summary: The World Wide Web has resulted in new and unanticipated avenues for conducting large-scale behavioral experiments. Crowdsourcing sites like Amazon Mechanical Turk, oDesk, and Taskcn, among others, have given researchers access to a large participant pool that operates around the clock. As a result, behavioral researchers in academia have turned to crowdsourcing sites in large numbers. Moreover, websites like eBay, Yelp and Reddit have become places where researchers can conduct field experiments. Companies like Microsoft, Facebook, Google and Yahoo! conduct hundreds of randomized experiments on a daily basis. We may be rapidly reaching a point where most behavioral experiments will be done online. *The Cooperative Phenotype*, Authors: Alexander Peysakhovich, Martin Nowak, David Rand Summary: Cooperation is the willingness to pay costs to give benefits to other persons or groups. We ask three questions about commonly made assumptions. Is there a unified underlying trait that we can call cooperativeness or is each cooperation decision completely unique? Can we study this trait using simple stripped down economic games? Our data suggests that the answer to both of these questions is yes. Is propensity to cooperate related to other social behavioral such as punishment or competitiveness? We find no relationships between these behaviors in the data, suggesting they may be three orthogonal individual-level characteristics. Our results demonstrate the power of using simple economic games to study pro-social in a controlled environment. *The Facebook Ads System: Mechanism Design and Machine Learning*, Authors: Chinmay Karande, Joaquin Quinonero Candela Summary: With well over 1 Billion active users and 1 Million advertiser, the Facebook ads ecosystem presents unique challenges in mechanism design and machine learning. In this talk, we describe high-level design decisions and insights gathered from development of the production system in these areas. We describe our two-pronged approach to value maximization. First, we talk about a unified auction that seamlessly blends sponsored content to organic content in a variety of innovative display formats. Second budget optimization system that helps maximize return on investment to the advertisers. Finally, for the auction to be efficient, it is essential that the algorithms to predict clicks and conversions be as accurate and calibrated, and as real-time as possible. Given the unprecedented scale of the problem, the machine learning algorithms we use need to strike a balance between accuracy and speed. We will share some practical lessons we have learnt. *Asymmetric effects of personalized social cues: Evidence from advertising experiments*, Authors: Sean J. Taylor, Eytan Bakshy, Dean Eckles, and Sinan Aral Summary: We aim to characterize which social relationships transmit the most social influence. The setup is simple: if a person has two friends who have previously adopted a product and she sees only one of them has adopted it, which of the friends’ adoptions will be most influential? In this paper we show that previous analyses of experiments where social cues or messages are randomly delivered or suppressed (e.g. Aral and Walker, 2012; Bakshy et al., 2012), while able to show that treatment effects of social cues vary by subpopulation, are generally not capable of identifying the difference in effectiveness between social cues. Without a randomization of adopter peers (which consequently randomizes dyadic characteristics), social cue effects estimates apply to populations of users with different unobservable affinities for products which can explain variation in adoption rates. We measure the bias associated with simpler experimental approaches to measuring peer effects heterogeneity and show that in some cases it changes the substantive policy implications of the findings. Additionally, Jon Kleinberg, a frequent Facebook collaborator, presented an overview of his work with Facebookers Lars Backstrom, Lada Adamic, and Alex Dow. *Computational Problems for Designed Social Systems*, Authors: Jon Kleinberg, Ashton Anderson, Lars Backstrom, Dan Huttenlocher, and Jure Leskovec, Lada Adamic, Alex Dow. Summary: As on-line social media systems become an increasingly central setting for human social interaction, it is important to appreciate the ways in which these are not simply venues for people to come together, but explicitly designed systems whose architectures serve to shape behavior. Developing, analyzing, and refining these designs is an important issue at the interface of computing, economics, and the social sciences. Here we consider several facets of the computational challenges around designed social systems, including the issue of filtering personal information streams -- managing a persons interface to the rest of a large social network -- and the design of complex incentive systems for steering individual and group behavior.
Posted on: Thu, 17 Jul 2014 02:52:30 +0000
Trending Topics
Recently Viewed Topics
© 2015