Collaborative Research: Distributed Mechanism Design with Learning Guarantees: Resource Allocation Among Networked Strategic Agents

在过去的几十年里，人们对通信、交通、能源、经济、社会网络中的战略主体如何在其他主体偏好不确定的情况下做出决策进行了大量研究。这项研究的动机是在我们不断联系的社会和经济中的大量应用，以及认识到在高度分散的网络中，完全知情或完全合规的代理等假设是站不住脚的。在战略主体存在的情况下，资源配置问题的激励设计属于机制设计的研究范畴。MD目前的现状存在两个问题。首先，它假设存在一个接受投标的中央实体，并能够与每个代理进行通信。其次，它没有充分解决主体如何收敛于设计均衡的问题。智力优势：我们在这个提案中的首要目标是创建一个新的研究子领域，在一个统一的框架中解决这些问题。我们利用这个框架来设计分布式机制，并为足够广泛的代理行为提供学习（即收敛）保证。为了实现我们的目标，我们计划沿着一条将基础研究与目标应用相结合的道路前进。特别是，我们首先研究了分布式机制的设计。我们考虑了两个说明性应用，即单播/多播多速率网络中的速率分配和能源社区的需求管理。然后，我们研究了包含学习保证的分布式机制，通过这种机制，社区达到了平衡。最后，我们在具有“无遗憾”型有限理性的非贝叶斯智能体的背景下研究了上述两个问题。更广泛的影响：除了上述两项示范应用外，本计划所进行的研究，将惠及广泛的实务领域，并对社会产生重大影响。例子包括智能能源和基础设施系统、通信系统、网络物理和人类系统、社会和经济系统等等。研究成果将用于课程开发、本科生和研究生的培训，以及在pi所在机构的各种推广活动。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

During the last decades there has been significant research in understanding how strategic agents in communication, transportation, energy, economic, societal networks make decisions in the presence of uncertainty about other agents' preferences. This research is motivated by a multitude of applications in our ever-connected society and economy, and the realization that assumptions such as fully-informed or fully-compliant agents are untenable in vastly decentralized networks. The design of incentives for resource allocation problems in the presence of strategic agents falls under the research area of mechanism design (MD). There are two issues with the current state of the art in MD. First, it assumes the existence of a central entity accepting bids and being capable of communicating with each agent. Second, it does not adequately address the question of how agents converge to the designed equilibria. Intellectual Merit: Our overarching objective in this proposal is to create a new subfield of research that addresses these issues in a unified framework. We utilize this framework to design mechanisms that are distributed and have learning (i.e., convergence) guarantees for a sufficiently broad range of agents' behaviors. To achieve our objective, we plan to proceed along a path that blends in fundamental research with targeted applications. In particular, we first investigate the design of distributed mechanisms. We consider two illustrative applications, namely, rate allocation in unicast/multicast-multirate networks and demand management of energy communities. We then investigate distributed mechanisms that incorporate learning guarantees, by which the community reaches an equilibrium. Finally, we study the above two problems in the context of non-Bayesian agents with ``no-regret'' type limited rationality.Broader Impacts: Besides the two illustrative applications, research to be carried in this project will benefit a broad range of practical fields of large societal impact. Examples include smart energy and infrastructure systems, communication systems, cyber physical and human systems, social and economical systems, to name a few. The research outcomes will be utilized for curriculum development, for training undergraduate and graduate students, and for various outreach activities at the PIs’ institutions.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.