Click-through rate (CTR) prediction has been one of the most central problems in computational advertising. Lately, embedding techniques that produce low-dimensional representations ad IDs drastically improve CTR accuracies. However, such learning are data demanding and work poorly on new ads with little logging data, which is known as cold-start problem.

Bike sharing provides an environment-friendly way for traveling and is booming all over the world. Yet, due to high similarity of user travel patterns, bike imbalance problem constantly occurs, especially dockless systems, causing significant impact on service quality company revenue. Thus, it has become a critical task operators resolve such efficiently. In this paper, we propose novel deep reinforcement learning framework incentivizing users rebalance systems. We model as Markov decision...

We study the problem of allocating impressions to sellers in e-commerce websites, such as Amazon, eBay or Taobao, aiming maximize total revenue generated by platform. employ a general framework reinforcement mechanism design, which uses deep learning design efficient algorithms, taking strategic behaviour into account. Specifically, we model impression allocation Markov decision process, where states encode history impressions, prices, transactions and actions are possible allocations each...

We put forward a modeling and algorithmic framework to design optimize mechanisms in dynamic industrial environments where designer can make use of the data generated process automatically improve future design. Our solution, coined reinforcement mechanism design, is rooted game theory but incorporates recent AI techniques get rid nonrealistic assumptions automated optimization feasible. instantiate our on key application scenarios Baidu Taobao, two largest mobile app companies China. For...

Abstract Cooperation in transboundary river basins can make water resources systems more efficient and benefit riparian stakeholders. However, a basin with upstream downstream stakeholders that have different interests, noncooperative outcomes often been observed. These be described by one‐shot prisoners' dilemma game where noncooperation (defection) is dominant equilibrium strategy. cooperative also observed several settings, such as the Lancang‐Mekong River Basin Asia. Such cooperation...

In many social systems in which individuals and organizations interact with each other, there can be no easy laws to govern the rules of environment, agents' payoffs are often influenced by other actions. We examine such a system setting sponsored search auctions tackle engine's dynamic pricing problem combining tools from both mechanism design AI domain. this setting, environment not only changes over time, but also behaves strategically. Over repeated interactions bidders, engine...

Decision making is a challenging task in online recommender systems. The decision maker often needs to choose contextual item at each step from set of candidates. Contextual bandit algorithms have been successfully deployed such applications, for the trade-off between exploration and exploitation state-of-art performance on minimizing costs. However, applicability existing methods limited by over-simplified assumptions problem, as assuming simple form reward function or static environment...

We consider revenue-optimal mechanism design for the case with one buyer and two items. The buyer's valuations towards items are independent additive. In this setting, optimal is unknown general valuation distributions. obtain categories of structural results that shed light on mechanisms. These can be summarized into conclusion: under certain conditions, mechanisms have simple menus.

In large e-commerce websites, sellers have been observed to engage in fraudulent behaviour, faking historical transactions order receive favourable treatment from the platforms, specifically through allocation of additional buyer impressions which results higher revenue for them, but not system as a whole. This emergent phenomenon has attracted considerable attention, with previous approaches focusing on trying detect illicit practices and punish miscreants. this paper, we employ principles...

Click-through rate (CTR) prediction has been one of the most central problems in computational advertising. Lately, embedding techniques that produce low-dimensional representations ad IDs drastically improve CTR accuracies. However, such learning are data demanding and work poorly on new ads with little logging data, which is known as cold-start problem. In this paper, we aim to predictions during both phase warm-up when a added candidate pool. We propose Meta-Embedding, meta-learning-based...

Designing revenue-optimal auctions for various settings is perhaps the most important, yet sometimes elusive, problem in mechanism design. Spiteful bidders have been intensely studied recently, especially because spite occurs many applications multiagent system and electronic commerce. We derive optimal auction such (as well as that are altruistic). It a generalization of Myerson’s (1981) auction. chooses an allocation maximizes agents’ virtual valuations, but generalized definition...

Despite their better revenue and welfare guarantees for repeated auctions, dynamic mechanisms have not been widely adopted in practice. This is partly due to the complexity of implementation as well unrealistic use forecasting future periods. We address these shortcomings present a new family that are simple require no distribution knowledge

Lately, the problem of designing multi-stage dynamic mechanisms has been shown to be both theoretically challenging and practically important. In this paper, we consider revenue optimal mechanism for a setting where an auctioneer sells set items buyer in multiple stages. At each stage, there could sale but item can only appear one stage. The type at stage is thus multi-dimensional vector characterizing buyer's valuations that assumed stage-wise independent. particular, propose novel class...

It is often observed that the probabilistic predictions given by a machine learning model can disagree with averaged actual outcomes on specific subsets of data, which also known as issue miscalibration. responsible for unreliability practical systems. For example, in online advertising, an ad receive click-through rate prediction 0.1 over some population users where its click 0.15. In such cases, have to be fixed before system deployed.

Modern machine learning models (such as deep neural networks and boosting decision tree models) have become increasingly popular in financial market prediction, due to their superior capacity extract complex non-linear patterns. However, since datasets very low signal-to-noise ratio are non-stationary, often prone overfitting suffer from instability issues. Moreover, various data mining tools more widely used quantitative trading, many trading firms been producing an increasing number of...

We revisit the pairwise kidney exchange problem established by Roth Sonmez and Unver [23]. Our goal, explained in terms of graph theory, is to find a maximum fractional matching on an undirected graph, that Lorenz-dominates any other matching. The Lorenz-dominant matching, which can be implemented as lottery integral matchings, some sense fairest allocation also enjoys property being incentive compatible. original algorithm et al. runs time exponential size input. In this paper, we target at...

Over the past decades, various theories and algorithms have been developed under framework of Stackelberg games part these innovations fielded scenarios national security defenses wildlife protections. However, one remaining difficulties in literature is that most theoretical works assume full information payoff matrices, while applications, leader often has no prior knowledge about follower’s matrix, but may gain utility function through repeated interactions. In this paper, we study...

We introduce a new family of dynamic mechanisms that restricts sellers from using future distributional knowledge. Since the allocation and pricing each auction period do not depend on type distributions periods, we call this non‐clairvoyant. develop framework (bank account mechanisms) for characterizing, designing, proving lower bounds (clairvoyant or non‐clairvoyant). use same methods to compare revenue extraction power clairvoyant non‐clairvoyant mechanisms.

Time-inconsistency refers to a paradox in decision making where agents exhibit inconsistent behaviors over time. Examples are procrastination tend postpone easy tasks, and abandonments start plan quit the middle. To capture such quantify inefficiency caused by behaviors, Kleinberg Oren (2014) propose graph model with certain cost structure initiate study of several interesting computation problems: 1) ratio: worst ratio between actual agent optimal cost, all instances; 2) motivating...