ICES: Small: Risk Aversion in Algorithmic Game Theory and Mechanism Design

算法博弈论和机制设计领域研究多个智能体之间的战略交互。现有数据和计算能力的增长使这些领域能够在理解代理人的激励、改进他们的决策和由此产生的社会结果方面产生真正的影响。一个巨大的挑战是不确定性和风险的存在，这就需要新的博弈论模型和算法技术。举个简单的例子，考虑去机场赶飞机。在交通不确定和截止日期的情况下，风险考虑至关重要，风险较小但较长的路线可能比最大限度地减少预期延误的路线更可取。但根本的问题不仅仅取决于一个人的喜好，而是取决于一个更广泛的社会模式，它涵盖了许多个人及其互动。例如，多个用户的战略性风险感知路由如何影响路由选择和网络中的总体拥塞？多个风险厌恶用户的个人最优(均衡)路线选择与社会最优交通分配有何不同？这项研究计划概述了一项议程，旨在调查不确定性和风险厌恶如何改变算法博弈论和机制设计所采用的传统模型和方法。它将从金融工程等在风险研究方面拥有丰富传统的领域引进模型和工具，并使其适应多用户计算环境。最终，目标是开发新的博弈论模型和分析，以及解决这些模型中的计算问题的新算法。该计划关注现实世界中复杂的多用户系统中出现的问题，并有可能改进涉及不确定性和风险厌恶用户的各种应用，例如，减少交通和电信网络的拥塞，提高互联网上的流媒体应用的质量，增加军事行动、自主和机器人导航等的安全和保障。从研究的角度来看，拟议研究的变革潜力是从根本上转变对不确定多用户环境的思考，从优化简单的预期性能转向系统地处理突发事件和风险。

The fields of Algorithmic Game Theory and Mechanism Design study the strategic interaction of multiple agents. The growth of available data and computational power enable these fields to make a real impact in understanding the agents' incentives and improving their decisions and the resulting social outcomes. A grand challenge is the presence of uncertainty and risk, which calls for new game theoretic models and algorithmic techniques. As a brief example, consider going to the airport to catch a flight. In uncertain traffic and a deadline, risk considerations are critical and a less risky, albeit longer, route might be preferable over one that minimizes expected delay. But the fundamental questions do not depend merely on one individual's preferences, but on a broader social model that captures many individuals and their interactions. For instance, how does strategic risk-aware routing by multiple users affect route selection and the overall congestion in the network? And how different is the individually optimal (equilibrium) route selection of multiple risk-averse users from a socially optimal traffic allocation?This research program outlines an agenda for investigating how uncertainty and risk aversion transform traditional models and methodologies employed by Algorithmic Game Theory and Mechanism Design. It will import models and tools from areas with a rich tradition in studying risk, such as Financial Engineering, and adapt them to multi-user computational environments. Ultimately, the goal is to develop new game theoretic models and analysis, as well as new algorithms for solving the computational questions in these models. The program focuses on problems that arise in complex multi-user systems in the real world and has the potential to improve a variety of applications that involve uncertainty and risk-averse users, for example, reduce congestion in transportation and telecommunication networks, enhance the quality of streaming applications over the Internet, increase the safety and security in military operations, autonomous and robot navigation, etc. From a research standpoint, the transformative potential of the proposed research is to fundamentally shift thinking about uncertain multi-user environments away from optimizing simple expected performance and instead towards a systematic treatment of contingencies and risk.