Collective decision making in distributed systems

该项目属于EPSRC的以下研究领域：(1)ICT网络和分布式系统，以及(2)统计和应用概率。集体决策在生物有机体中很普遍，从细菌到群居昆虫，如蚂蚁和蜜蜂(蚁群共同决定觅食和筑巢地点)，再到人类(从消费者到股票交易者)。在互联网以及电力和运输网络等工程系统中也出现了集体决策的需要，这些系统由大量需要协调路线选择或供需决策的自治实体组成。较新的应用包括集群机器人和区块链技术。在这个项目中，我们将探索一些简化的集体决策数学模型，并研究当它们被大量相互作用的自治代理使用时出现的现象。例如，假设代理人收到关于自然状态的独立私人信号，并具有推断自然真实状态的共同目标。最有效的解决方案是汇集所有信息并使用它们进行推理，但系统可能太大，无法实现这一点。代理能否仅使用与邻居的有限通信来合作，以实现与完全信息共享所能实现的推断相媲美的推断？我们将结合严格的数学分析和模拟来解决这些问题。我们将为目标适定的集体决策问题提出和分析各种算法。

The project falls with the following EPSRC research areas: (1) ICT networks and distributed systems, and (2) Statistics and applied probability. Collective decision-making is widespread amongst biological organisms ranging from bacteria to social insects such as ants and bees (where colonies make collective decisions about foraging and nest sites) to human beings (from consumers to stock market traders). The need for collective decision-making also arises in engineered systems such as the Internet, and power and transport networks, which are composed of large numbers of autonomous entities that need to coordinate route choice or supply and demand decisions. Newer applications include swarm robotics and blockchain technologies. In this project, we will explore a number of simplified mathematical models of collective decision-making and study the emergent phenomena that arise when they are employed by large numbers of interacting autonomous agents. As an example, suppose agents receive independent private signals about a state of nature, and have a shared goal of inferring the true state of nature. The most efficient solution is to pool all the information and use it to make the inference, but the system may be too large for this to be feasible. Can agents cooperate using only limited communication with neighbours to achieve inference that is comparable to what could be achieved with full information sharing? We will address such questions using a combination of rigorous mathematical analysis and simulations. We will propose and analyse a variety of algorithms for collective decision-making problems with well-posed objectives.