City-scale traffic signal control (TSC) involves thousands of heterogeneous intersections with varying topologies, making cooperative decision-making across intersections particularly challenging. Given the prohibitive computational cost of learning individual policies for each intersection, some researchers explore learning a universal policy to control each intersection in a decentralized manner, where the key challenge is to construct a universal representation method for heterogeneous intersections. However, existing methods are limited to universally representing information of heterogeneous ego intersections, neglecting the essential representation of influence from their heterogeneous neighbors. Universally incorporating neighborhood information is nontrivial due to the intrinsic complexity of traffic flow interactions, as well as the challenge of modeling collective influences from neighbor intersections. To address these challenges, we propose CityLight, which learns a universal policy based on representations obtained with two major modules: a Neighbor Influence Encoder to explicitly model neighbor's influence with specified traffic flow relation and connectivity to the ego intersection; a Neighbor Influence Aggregator to attentively aggregate the influence of neighbors based on their mutual competitive relations. Extensive experiments on five city-scale datasets, ranging from 97 to 13,952 intersections, confirm the efficacy of CityLight, with an average throughput improvement of 11.68% and a lift of 22.59% for generalization. Our codes and datasets are released: https://github.com/tsinghua-fib-lab/CityLight.

In large-scale metropolis, it is critical to efficiently allocate various resources such as electricity, medical care, and transportation to meet the living demands of citizens, according to the spatio-temporal distributions of resources and demands. Previous researchers have done plentiful work on such problems by leveraging Multi-Agent Reinforcement Learning (MARL) methods, where multiple agents cooperatively regulate and allocate the resources to meet the demands. However, facing the great number of agents in large cities, existing MARL methods lack efficient parameter sharing strategies among agents to reduce computational complexity. There remain two primary challenges in efficient parameter sharing: (1) during the RL training process, the behavior of agents changes significantly, limiting the performance of group parameter sharing based on fixed role division decided before training; (2) the behavior of agents forms complicated action trajectories, where their role characteristics are implicit, adding difficulty to dynamically adjusting agent role divisions during the training process. In this paper, we propose Dynamic Parameter Sharing (DyPS) to solve the above challenges. We design self-supervised learning tasks to extract the implicit behavioral characteristics from the action trajectories of agents. Based on the obtained behavioral characteristics, we propose a hierarchical MARL framework capable of dynamically revising the agent role divisions during the training process and thus shares parameters among agents with the same role, reducing computational complexity. In addition, our framework can be combined with various typical MARL algorithms, including IPPO, MAPPO, etc. We conduct 7 experiments in 4 representative resource allocation scenarios, where extensive results demonstrate our method's superior performance, outperforming the state-of-the-art baseline methods by up to 31%. Our source codes are available at https://github.com/tsinghua-fib-lab/DyPS.

Understanding the impact of COVID-19 on urban life rhythms is crucial for accelerating the return-to-normal progress and envisioning more resilient and inclusive cities. While previous studies either depended on small-scale surveys or focused on the response to initial lockdowns, this paper uses large-scale location service data to systematically analyze the urban mobility behavior changes across three distinct phases of the pandemic, i.e., pre-pandemic, lockdown, and reopen. Our analyses reveal two typical patterns that govern the mobility behavior changes in most urban venues: daily life-centered urban venues go through smaller mobility drops during the lockdown and more rapid recovery after reopening, while work-centered urban venues suffer from more significant mobility drops that are likely to persist even after reopening. Such mobility behavior changes exert deeper impacts on the underlying social fabric, where the level of mobility reduction is positively correlated with the experienced segregation at that urban venue. Therefore, urban venues undergoing more mobility reduction are also more filled with people from homogeneous socio-demographic backgrounds. Moreover, mobility behavior changes display significant heterogeneity across geographical regions, which can be largely explained by the partisan inclination at the state level. Our study shows the vast potential of location service data in deriving a timely and comprehensive understanding of the social dynamic in urban space, which is valuable for informing the gradual transition back to the normal lifestyle in a “post-pandemic era”.

Recent advancements in reinforcement learning have witnessed remarkable achievements by intelligent agents ranging from game-playing to industrial applications. Of particular interest is the area of multi-agent reinforcement learning (MARL), which holds significant potential for real-world scenarios. However, typical MARL methods are limited in their ability to handle tens of agents, leaving scenarios with up to hundreds or even thousands of agents almost unexplored. The scaling up of the number of agents presents two primary challenges: (1) agent-agent interactions are crucial in multi-agent systems while the number of interactions grows quadratically with the number of agents, resulting in substantial computational complexity and difficulty in strategies-learning; (2) the strengths of interactions among agents exhibit variations both across agents and over time, making it difficult to precisely model such interactions. In this paper, we propose a novel approach named Graph Attention Mean Field (GAT-MF). By converting agent-agent interactions into interactions between each agent and a weighted mean field, we achieve a substantial reduction in computational complexity. The proposed method offers a precise modeling of interaction dynamics with mathematical proofs of its correctness. Additionally, we design a graph attention mechanism to automatically capture the diverse and time-varying strengths of interactions, ensuring an accurate representation of agent interactions. Through extensive experimentation conducted in both manual and real-world scenarios involving over 3000 agents, we validate the efficacy of our method. The results demonstrate that our method outperforms the best baseline method with a remarkable improvement of 42.7%. Furthermore, our method saves 86.4% training time and 19.2% GPU memory compared to the best baseline method. For reproducibility, our source codes and data are available at https://github.com/tsinghua-fib-lab/Large-Scale-MARL-GATMF.

With the advent of the COVID-19 pandemic, the shortage in medical resources became increasingly more evident. Therefore, efficient strategies for medical resource allocation are urgently needed. However, conventional rule-based methods employed by public health experts have limited capability in dealing with the complex and dynamic pandemic-spreading situation. In addition, model-based optimization methods such as dynamic programming (DP) fail to work since we cannot obtain a precise model in real-world situations most of the time. Model-free reinforcement learning (RL) is a powerful tool for decision-making; however, three key challenges exist in solving this problem via RL: (1) complex situations and countless choices for decision-making in the real world; (2) imperfect information due to the latency of pandemic spreading; and (3) limitations on conducting experiments in the real world since we cannot set up pandemic outbreaks arbitrarily. In this article, we propose a hierarchical RL framework with several specially designed components. We design a decomposed action space with a corresponding training algorithm to deal with the countless choices, ensuring efficient and real-time strategies. We design a recurrent neural network–based framework to utilize the imperfect information obtained from the environment. We also design a multi-agent voting method, which modifies the decision-making process considering the randomness during model training and, thus, improves the performance. We build a pandemic-spreading simulator based on real-world data, serving as the experimental platform. We then conduct extensive experiments. The results show that our method outperforms all baselines, which reduces infections and deaths by 14.25% on average without the multi-agent voting method and up to 15.44% with it.

In the fight against the COVID-19 pandemic, vaccines are the most critical resource but are still in short supply around the world. Therefore, efficient vaccine allocation strategies are urgently called for, especially in large-scale metropolis where uneven health risk is manifested in nearby neighborhoods. However, there exist several key challenges in solving this problem: (1) great complexity in the large scale scenario adds to the difficulty in experts' vaccine allocation decision making; (2) heterogeneous information from all aspects in the metropolis' contact network makes information utilization difficult in decision making; (3) when utilizing the strong decision-making ability of reinforcement learning (RL) to solve the problem, poor explainability limits the credibility of the RL strategies. In this paper, we propose a reinforcement learning enhanced experts method. We deal with the great complexity via a specially designed algorithm aggregating blocks in the metropolis into communities and we hierarchically integrate RL among the communities and experts solution within each community. We design a self-supervised contact network representation algorithm to fuse the heterogeneous information for efficient vaccine allocation decision making. We conduct extensive experiments in three metropolis with real-world data and prove that our method outperforms the best baseline, reducing 9.01% infections and 12.27% deaths.We further demonstrate the explainability of the RL model, adding to its credibility and also enlightening the experts in turn.

Facing the outbreak of COVID-19, shortage in medical resources becomes increasingly outstanding. Therefore, efficient strategies for medical resource allocation are urgently called for. Reinforcement learning (RL) is powerful for decision making, but three key challenges exist in solving this problem via RL: (1) complex situation and countless choices for decision making in the real world; (2) only imperfect information are available due to the latency of pandemic spreading; (3) limitations on conducting experiments in real world since we cannot set pandemic outbreaks arbitrarily. In this paper, we propose a hierarchical reinforcement learning method with a corresponding training algorithm. We design a decomposed action space to deal with the countless choices to ensure efficient and real time strategies. We also design a recurrent neural network based framework to utilize the imperfect information obtained from the environment. We build a pandemic spreading simulator based on real world data, serving as the experimental platform. We conduct extensive experiments and the results show that our method outperforms all the baselines, which reduces infections and deaths by 14.25% on average.

Facing the worldwide rapid spreading of COVID-19 pandemic, we need to understand its diffusion in the urban environments with heterogeneous population distribution and mobility. However, challenges exist in the choice of proper spatial resolution, integration of mobility data into epidemic modelling, as well as incorporation of unique characteristics of COVID-19.