We introduce a new model of stochastic bandits with adversarial corruptions which aims to capture settings where most the input follows pattern but some fraction it can be adversarially changed trick algorithm, e.g., click fraud, fake reviews and email spam. The goal this is encourage design bandit algorithms that (i) work well in mixed models, (ii) whose performance deteriorates gracefully as we move from fully models.

We consider the problem faced by a firm that receives highly differentiated products in an online fashion. The needs to price these sell them its customer base. Products are described vectors of features and market value each product is linear values features. does not initially know different features, but can learn based on whether were sold at posted prices past. This model motivated applications such as marketplaces, flash sales, loan pricing. first multidimensional version binary search...

The Generalized Second Price Auction has been the main mechanism used by search companies to auction positions for advertisements on pages. In this paper we study social welfare of Nash equilibria game in various models. full information setting, socially optimal are known exist (i.e., Stability is 1). This first prove bounds price anarchy, and give any Bayesian setting. Our result show that anarchy small assuming all bidders play un-dominated strategies. setting a bound 1.618 pure...

The Generalized Second Price (GSP) auction is the primary used for selling sponsored search advertisements. In this paper we consider revenue of at equilibrium. We prove that if agent values are drawn from identical regular distributions, then GSP paired with an appropriate reserve price generates a constant fraction (1/6th) optimal revenue. full-information game, show any Nash equilibrium obtains least half VCG mechanism excluding payment single participant. This bound holds also price, and tight.

The buying and selling of information is taking place at a scale unprecedented in the history commerce, thanks to formation online marketplaces for user data. Data providing agencies sell advertisers allow them match ads viewers more effectively. In this paper we study design optimal mechanisms monopolistic data provider buyer, model where both parties have (possibly correlated) private signals about state world, buyer uses learned from seller, along with his own signal, choose an action...

We propose ego-splitting, a new framework for detecting clusters in complex networks which leverage the local structures known as ego-nets (i.e. subgraph induced by neighborhood of each node) to de-couple overlapping clusters. Ego-splitting is highly scalable and flexible framework, with provable theoretical guarantees, that reduces clustering problem simpler more amenable non-overlapping (partitioning) problem. can scale community detection graphs tens billions edges outperform previous...

In many settings agents participate in multiple different auctions that are not necessarily implemented simultaneously. Future opportunities affect strategic considerations of the players each auction, introducing externalities. Motivated by this consideration, we study a setting market buyers and sellers, where seller holds one item, bidders have combinatorial valuations sellers hold item sequentially.Our results qualitatively from those simultaneous auctions, proving simultaneity is...

The Generalized Second Price (GSP) auction is the primary method by which sponsered search advertisements are sold. We study performance of this in Bayesian setting for players with correlated types. Correlation arises very naturally context sponsored auctions, especially as a result uncertainty inherent behaviour underlying ad allocation algorithm. demonstrate that Anarchy GSP bounded $4$, even when agents have arbitrarily Our proof highlights connection between mechanism and concept...

We consider the problem faced by a firm that receives highly differentiated products in an online fashion and needs to price them order sell its customer base. Products are described vectors of features market value each product is linear values features. The does not initially know different features, but it can learn based on whether were sold at posted prices past. This model motivated question advertising, where impressions arrive over time be first multi-dimensional version binary...

In many settings agents participate in multiple different auctions that are not necessarily implemented simultaneously. Future opportunities affect strategic considerations of the players each auction, introducing externalities. Motivated by this consideration, we study a setting market buyers and sellers, where seller holds one item, bidders have combinatorial valuations sellers hold item sequentially.Our results qualitatively from those simultaneous auctions, proving simultaneity is...

A central issue in applying auction theory practice is the problem of dealing with budget-constrained agents. desirable goal to design incentive compatible, individually rational, and Pareto optimal auctions while respecting budget constraints. Achieving this particularly challenging presence nontrivial combinatorial constraints over set feasible allocations. Toward motivated by AdWords auctions, we present an for polymatroidal environments satisfying above properties. Our employs a novel...

Typical models of strategic interactions in computer science use simultaneous move games. However, applications simultaneity is often hard or impossible to achieve. In this paper, we study the robustness Nash Equilibrium when assumption dropped. particular propose studying sequential price anarchy: quality outcomes versions games whose counterparts are prototypical algorithmic game theory. We different classes with high anarchy, and show that subgame perfect equilibrium their version a much...

Central results in economics guarantee the existence of efficient equilibria for various classes markets. An underlying assumption early work is that agents are price-takers, i.e., honestly report their true demand response to prices. A line research economics, initiated by Hurwicz (1972), devoted understanding how such markets perform when strategic about demands. This captured Walrasian Mechanism proceeds collecting reported demands, finding clearing prices market via an ascending price...

We study the problem of minimizing swap regret in structured normal-form games. Players have a very large (potentially infinite) number pure actions, but each action has an embedding into $d$-dimensional space and payoffs are given by bilinear functions these embeddings. provide efficient learning algorithm for this setting that incurs at most $\tilde{O}(T^{(d+1)/(d+3)})$ after $T$ rounds. To achieve this, we introduce new online call \emph{full minimization}. In problem, learner repeatedly...

A central issue in applying auction theory practice is the problem of dealing with budget-constrained agents. desirable goal to design incentive compatible, individually rational, and Pareto optimal auctions while respecting budget constraints. Achieving this particularly challenging presence nontrivial combinatorial constraints over set feasible allocations. Toward motivated by AdWords auctions, we present an for polymatroidal environments satisfying these properties. Our employs a novel...

We study the question of setting and testing reserve prices in single item auctions when bidders are not identical. At a high level, there two generalizations standard second price auction: lazy version we first determine winner, then apply prices; eager discard meeting their reserves, winner among rest. show that versions have dramatically different properties: reserves easy to optimize, A/B test production, whereas always lead higher welfare, but optimization is NP-complete, naive will...

We consider a multidimensional search problem that is motivated by questions in contextual decision making, such as dynamic pricing and personalized medicine. Nature selects state from d-dimensional unit ball then generates sequence of directions. are given access to the directions but not state. After receiving direction, we have guess value dot product between direction. Our goal minimize number times when our more than ϵ away true answer. construct polynomial time algorithm call projected...

We show that the class of preferences satisfying gross substitutes condition Kelso and Crawford (1982) is strictly larger than endowed assignment valuations Hatfield Milgrom (2005), thus resolving open question posed by latter paper.In particular, our result implies not every substitutable valuation function can be "decomposed" into a combination unitdemand valuations.