A lot of research and business activity in the Cambridge/Boston area is engaged in economic and computational questions in regard to understanding and developing the economics of Internet activity. Examples of topics of interest include theoretical, modeling, algorithmic, and empirical work on electronic commerce, networked behavior, and social networks.
One of the main purposes of CAEC is to encourage collaboration between local researchers. Significant emphasis will be placed on a poster session and short talks. The overall structure of the day will involve three longer talks, by Avrim Blum, Cynthia Dwork, Mitsuru Igami and Elie Tamer, with a short talks session and a poster session over lunch, along with brief poster announcements.
CAEC17 will take place at Microsoft Research, New England, 1 Memorial Drive, Cambridge on Friday, December 1st 2017 and is supported by Microsoft Research.
Abstract. In this talk I will discuss the problem of trying to learn the requirements and preferences of economic agents by observing the outcomes of an allocation mechanism whose rules you also don’t initially know. As an example, consider observing web pages where the agents are advertisers and the winners are those whose ads show up on the given page. We know these ads are placed based on bids and other constraints given to some auction mechanism, but we do not get to see these bids and constraints. What we would like to do is from repeated observations of this type to learn what the requirements and preferences of the agents are. Or consider observing the input-output behavior of a cloud computing service, where the input consists of a set of agents requesting service, and the output tells us which actually received service and which did not. In this case, we assume the agents who did not receive service were not served due to overlap of their resource needs with higher-priority requests. From such input-output behavior, we would like to learn the underlying structure. Our goal will be from observing a series of such interactions to try to learn both the needs and preferences of the agents and perhaps also the rules of the allocation mechanism. This talk is based on work joint with Yishay Mansour and Jamie Morgenstern, as well as work joint with undergraduate student Michael Liang.
Abstract. “Unfairness” of algorithms – for tasks ranging from advertising to recidivism prediction – has attracted considerable attention in the popular press. Algorithmic techniques for achieving fairness now routinely appear in dedicated workshops and symposia as well as in established research conferences. This talk will focus on the (relatively) new study of a mathematically rigorous theory of fairness: definitions, methods, and provable limits and tradeoffs, providing a lens for hot-button policy issues such as “interpretability” and raising new questions for future research.
Abstract. Artificial intelligence (AI) has achieved superhuman performance in a growing number of tasks, including the classical games of chess, shogi, and Go, but understanding and explaining AI remain challenging. This paper studies the machine-learning algorithms for developing the game AIs, and provides their structural interpretations. Specifically, chess-playing Deep Blue is a calibrated value function, whereas shogi-playing Bonanza represents an estimated value function via Rust's (1987) nested fixed-point method. AlphaGo's "supervised-learning policy network" is a deep neural network (DNN) version of Hotz and Miller's (1993) conditional choice probability estimates; its "reinforcement-learning value network" is equivalent to Hotz, Miller, Sanders, and Smith's (1994) simulation method for estimating the value function. Their performances suggest DNNs are a useful functional form when the state space is large and data are sparse. Explicitly incorporating strategic interactions and unobserved heterogeneity in the data-generating process would further improve AIs' explicability.
Abstract. Given a sample of bids from independent auctions, this paper examines the question of inference on auction fundamentals (e.g. valuation distributions, welfare measures) under weak assumptions on information structure. The question is important as it allows us to learn about the valuation distribution in a robust way, i.e., without assuming that a particular information structure holds across observations. We leverage recent contributions in the robust mechanism design literature that exploit the link between Bayesian Correlated Equilibria and Bayesian Nash Equilibria in incomplete information games to construct an econometrics framework for learning about auction fundamentals using observed data on bids. We showcase our construction of identified sets in private value and common value auctions. Our approach for constructing these sets inherits the computational simplicity of solving for correlated equilibria: checking whether a particular valuation distribution belongs to the identified set is as simple as determining whether a linear program is feasible. A similar linear program can be used to construct the identified set on various welfare measures and counterfactual objects. For inference and to summarize statistical uncertainty, we propose novel finite sample methods using tail inequalities that are used to construct confidence regions on sets. We also highlight methods based on Bayesian bootstrap and subsampling. A set of Monte Carlo experiments show adequate finite sample properties of our inference procedures. We also illustrate our methods using data from OCS auctions.
Location: CAEC17 will take place at Microsoft, 1 Memorial Drive, 1st Floor, Horace Mann Theater on Fri Dec 01.
Abstract. “Unfairness” of algorithms – for tasks ranging from advertising to recidivism prediction – has attracted considerable attention in the popular press. Algorithmic techniques for achieving fairness now routinely appear in dedicated workshops and symposia as well as in established research conferences. This talk will focus on the (relatively) new study of a mathematically rigorous theory of fairness: definitions, methods, and provable limits and tradeoffs, providing a lens for hot-button policy issues such as “interpretability” and raising new questions for future research.
Annie Liang. Predicting and Understanding Initial Play (joint with Drew Fudenberg)
Abstract. We take a machine learning approach to the problem of predicting initial play in strategic-form games. We predict game
play data from previous laboratory experiments, and also a new data set of 200 games with randomly distributed payoffs that were played on Mechanical Turk.
We consider two approaches, with the goals of uncovering new regularities in play and improving the predictions of existing theories. First, we use
machine learning algorithms to train prediction rules based on a large set of game features. Examination of the games where our algorithm predicts play correctly
but the existing models do not leads us to introduce a risk aversion parameter, which we find significantly improves predictive accuracy. Second, we augment existing
empirical models by using play in a set of training games to predict how the models' parameters vary across new games. This modified approach generates better out-of-sample predictions,
and provides insight into how and why the parameters vary. These methodologies are not special to the problem of predicting play in games, and may be useful in other contexts.
Hongyao Ma. Spatio-Temporal Pricing for Ridesharing Platforms
Abstract. Ridesharing platforms match drivers and riders and use dynamic pricing to balance supply and demand. A challenge is to set prices in a way they are appropriately smooth in time and space, in the sense that drivers will choose to accept their dispatches rather drive to another area, or wait for higher prices or a better trip. We develop a discrete time, multi-period and multi-location model for which there always exists anonymous, origin-destination competitive equilibrium prices, and give an efficient planning algorithm. The Spatio-Temporal Pricing (STP) mechanism computes a driver-pessimal competitive equilibrium plan at the beginning of time and after any driver deviation. The plan (dispatches of drivers to pick up drivers, driver re-positioning, and payments) is welfare-maximizing and envy free. The STP mechanism is incentive aligned: it is a subgame perfect equilibrium for drivers to always accept their dispatches at all times. We also give an impossibility result, that there can be no dominant strategy, welfare-optimal, envy-free and budget balanced mechanism. Numerical simulations show that in comparison to a baseline, myopic pricing mechanism, where drivers incur a high regret, the STP mechanism achieves significantly higher social welfare and no regret.
Georgios Zervas. Consumer Reviews and Regulation: Evidence from NYC Restaurants (joint with Chiara Farronato)
Abstract. We investigate complementarities between two signals of restaurant quality:
health inspections and online reviews. To protect consumers, health inspectors
periodically evaluate restaurant hygiene, and assign health grades to
restaurants. Consumers are also able to rate restaurant quality online,
through review platforms like Yelp. We investigate whether consumer reviews
can predict hygiene violations. To do so, we implement we use machine learning
to predict individual restaurant violations from the text of Yelp reviews. We
find that consumer reviews are good predictors of food handling violations,
but are poor predictors of facilities and maintenance violations. We then
investigate how the hygienic information contained in online reviews affects
demand and supply. On the demand side, we find that conditional on hygiene
related information contained in online reviews, consumers still take health grades
into account when choosing restaurants. On the supply side, we find that
review platforms create reputational incentives for restaurants to reduce
hygiene violations that are predictable by Yelp reviews. Specifically, we find
that relative to restaurants not on Yelp, restaurants on Yelp score better on
hygiene dimensions detectable by consumers than on dimensions not detectable
by consumers. Our results have implications for the design of government
regulation in a world where consumers rate their service experiences online.
Abstract. Given a sample of bids from independent auctions, this paper examines the question of inference on auction fundamentals (e.g. valuation distributions, welfare measures) under weak assumptions on information structure. The question is important as it allows us to learn about the valuation distribution in a robust way, i.e., without assuming that a particular information structure holds across observations. We leverage recent contributions in the robust mechanism design literature that exploit the link between Bayesian Correlated Equilibria and Bayesian Nash Equilibria in incomplete information games to construct an econometrics framework for learning about auction fundamentals using observed data on bids. We showcase our construction of identified sets in private value and common value auctions. Our approach for constructing these sets inherits the computational simplicity of solving for correlated equilibria: checking whether a particular valuation distribution belongs to the identified set is as simple as determining whether a linear program is feasible. A similar linear program can be used to construct the identified set on various welfare measures and counterfactual objects. For inference and to summarize statistical uncertainty, we propose novel finite sample methods using tail inequalities that are used to construct confidence regions on sets. We also highlight methods based on Bayesian bootstrap and subsampling. A set of Monte Carlo experiments show adequate finite sample properties of our inference procedures. We also illustrate our methods using data from OCS auctions.
Abstract. In this talk I will discuss the problem of trying to learn the requirements and preferences of economic agents by observing the outcomes of an allocation mechanism whose rules you also don’t initially know. As an example, consider observing web pages where the agents are advertisers and the winners are those whose ads show up on the given page. We know these ads are placed based on bids and other constraints given to some auction mechanism, but we do not get to see these bids and constraints. What we would like to do is from repeated observations of this type to learn what the requirements and preferences of the agents are. Or consider observing the input-output behavior of a cloud computing service, where the input consists of a set of agents requesting service, and the output tells us which actually received service and which did not. In this case, we assume the agents who did not receive service were not served due to overlap of their resource needs with higher-priority requests. From such input-output behavior, we would like to learn the underlying structure. Our goal will be from observing a series of such interactions to try to learn both the needs and preferences of the agents and perhaps also the rules of the allocation mechanism. This talk is based on work joint with Yishay Mansour and Jamie Morgenstern, as well as work joint with undergraduate student Michael Liang.
Nicole Immorlica. Approximate Efficiency in Matching Markets
Abstract. We propose a measure of approximate ex-ante Pareto e fficiency in matching markets. Using this notion, we quantify the intuited e fficiency improvement of the so-called Boston mechanism and the recently-proposed choice-augmented deferred acceptance mechanism over the random serial dictatorship mechanism. Furthermore, we provide a formal statement and analysis of a ra ffle-type mechanism, which is conceptually simpler than the Boston mechanism and has a comparable effi ciency guarantee.
Dean Eckles. Randomization inference for spillovers in networks
Abstract.
Social and behavioral scientists are interested in testing of hypotheses about spillovers (i.e. interference, exogenous peer effects) in social networks. However, when there is a single network, this is complicated by lack of independent observations. We explore Fisherian randomization inference as an approach to exact finite-population inference, where the main problem is that the relevant hypotheses are non-sharp null hypotheses. Fisherian randomization inference can be used to test these hypotheses either by (a) making the hypotheses sharp by assuming a model for direct effects or (b) conducting conditional randomization inference such that the hypotheses are sharp. I present both of these approaches, the latter of which is developed in Aronow (2012) and our paper (Athey, Eckles & Imbens, 2017). I illustrate these methods with application to a large voter turnout experiment on Facebook (Jones et al., 2017)
Nir Rosenfeld. Optimal Tagging with Markov Chain Optimization
Abstract. In many information systems, content creators and owners use tags to describe and annotate content. Since tags can have a considerable effect on the volume of traffic that eventually reaches an item, content creators should benefit from a disciplined approach to tag selection.
Here we introduce the task of optimal tagging, where the goal is to choose a subset of tags for a new item, such that the probability of browsing users reaching that item is maximized. We formulate this as a general state-selection problem over Markov chains, show that it is NP-hard, but is monotone and submodular and thus has a greedy approximation. We provide an efficient implementation of the greedy step, and demonstrate how individual content creators can optimize tags by simple trial and error.
Abstract. Artificial intelligence (AI) has achieved superhuman performance in a growing number of tasks, including the classical games of chess, shogi, and Go, but understanding and explaining AI remain challenging. This paper studies the machine-learning algorithms for developing the game AIs, and provides their structural interpretations. Specifically, chess-playing Deep Blue is a calibrated value function, whereas shogi-playing Bonanza represents an estimated value function via Rust's (1987) nested fixed-point method. AlphaGo's "supervised-learning policy network" is a deep neural network (DNN) version of Hotz and Miller's (1993) conditional choice probability estimates; its "reinforcement-learning value network" is equivalent to Hotz, Miller, Sanders, and Smith's (1994) simulation method for estimating the value function. Their performances suggest DNNs are a useful functional form when the state space is large and data are sparse. Explicitly incorporating strategic interactions and unobserved heterogeneity in the data-generating process would further improve AIs' explicability.