This paper introduces Localized Bipartite Match Graph Attention Q-Learning (BMG-Q), a novel Multi-Agent Reinforcement Learning (MARL) algorithm framework tailored for ride-pooling order dispatch. BMG-Q advances ride-pooling decision-making process with the localized bipartite match graph underlying the Markov Decision Process, enabling the development of novel Graph Attention Double Deep Q Network (GATDDQN) as the MARL backbone to capture the dynamic interactions among ride-pooling vehicles in fleet. Our approach enriches the state information for each agent with GATDDQN by leveraging a localized bipartite interdependence graph and enables a centralized global coordinator to optimize order matching and agent behavior using Integer Linear Programming (ILP). Enhanced by gradient clipping and localized graph sampling, our GATDDQN improves scalability and robustness. Furthermore, the inclusion of a posterior score function in the ILP captures the online exploration-exploitation trade-off and reduces the potential overestimation bias of agents, thereby elevating the quality of the derived solutions. Through extensive experiments and validation, BMG-Q has demonstrated superior performance in both training and operations for thousands of vehicle agents, outperforming benchmark reinforcement learning frameworks by around 10% in accumulative rewards and showing a significant reduction in overestimation bias by over 50%. Additionally, it maintains robustness amidst task variations and fleet size changes, establishing BMG-Q as an effective, scalable, and robust framework for advancing ride-pooling order dispatch operations.

The morphological diversity, referring to the variations in traffic network topologies defined in this paper, often emerges and brings difficulties in successfully transferring a pre-trained prediction model from one traffic network to another. Moreover, most existing research primarily assumes that traffic data in source and target networks follow independent and identically distributed (i.i.d.) patterns, which is usually not consistent with real-world situations, particularly when considering morphological diversity. For this inconsistency, many efforts have been made, but they mainly concentrate on temporal aspects, which significantly differ from traffic prediction due to spatial and temporal correlations among road segments, influenced by variations in road topology and traffic behavior. This paper introduces a causality-based spatiotemporal out-of-distribution (OOD) generalization method, which is adaptable to most GNNs for diverse, large-scale, dynamic traffic systems with zero-shot. Furthermore, to enhance the generalization and adaptability of the proposed method, we introduce graph matching and equal-sized graph partitioning to alleviate spatial shift between the source and target traffic networks, reduce and align the scale of the networks. Experiments carried out on traffic flow datasets demonstrate that our method significantly improves the performance of various GNN-based traffic predictors in the situation of morphological diversity, achieving a maximum reduction in MAE of 33.08%. Compared to other OOD-driven baselines, our approach also shows a notable improvement, with up to a 40.58% decrease in MAE.

The on-demand ride-pooling service, defined as two or more passengers sharing the same vehicle en-route along a proportion of their travel trajectories, offers many benefits, such as discounted trip fares for customers, higher income for drivers, increased profit for ride-sourcing companies, and reduced fuel consumption for environmental protection. Motivated by the potential of ride-pooling, many ride-sourcing companies launch pooling services based on the non-pooling market. By providing pooling and non-pooling services simultaneously, they compete with public transit for passengers. Of particular interest to service providers is the upfront pricing problem for the pooling service. It allows pooling riders to be informed of service prices even before the trip starts, and consequently makes the pooling service more attractive to passengers. However, it remains a challenging issue to obtain the optimal spatial-temporal upfront pricing strategy for the pooling service, considering the heterogeneity, dynamics, imbalance of demand /supply, and differentiation between pooling and non-pooling services. To address this problem, two reinforcement learning frameworks (i.e., single-agent Markov Decision Process (MDP) and multi-agent Markov Decision Process (MMDP)) are implemented to gain the pricing policy with the maximum daily profit of the platform, where the pooling price, as the action, not only directly affects the profit of each pooling request, but also has an influence on the mode splitting among pooling, non-pooling, and public transit service. Two tailored reinforcement learning methods are developed and adopted to solve the MDPs. Through extensive empirical experiments with a well-designed simulator, we show that the proposed multi-agent framework is able to remarkably improve the system performance.

Multi-step traffic speed prediction is a challenging issue due to the multiple spatial-temporal dependencies among roads. Some spatial dependencies, especially those formed by different traffic modes, are not fully exploited, and how to simultaneously consider spatial and temporal dependencies and effectively integrate them within a single prediction framework needs further exploration. To tackle the above issues, we propose a multi-view spatial-temporal graph convolutional framework MVSTG, which adequately exploits the multi-view spatial-temporal dependencies and their interactions to improve the accuracy of traffic prediction. Multi-view temporal learning captures the multiple temporal trends by temporal convolution from multi-granularity historical data, and multi-view spatial learning handles the multiple spatial correlations by graph convolution from multiple graphs. In addition, view-wise attention-based fusion is proposed to adaptively identify the importance of each upstream view, fuse the multi-view information, and generate integrated results for downstream views. The experiments on two real-world urban traffic datasets demonstrate that the multi-view data and the proposed model framework enhance performance on the accuracy of speed prediction, especially in mid-term and long-term prediction.

In recent years, deep learning-based dehazing models have presented a momentum of dramatic growth. Unfortunately, most deep learning-based approaches heavily rely on synthetically hazed images for model training, which makes these methods brittle to restore hazy images taken from real-world scenes, due to the sample distribution discrepancy between synthetic and realistic images. Although some attempts have been made to overcome this difficulty by augmenting image spatial features with spectral features, the power of the spectral features still remains underutilized. In this paper, we propose the Spectral Dual-Channel Encoding (SDCE) framework for high-quality image dehazing, by unleashing the power of spectral feature encoding. We argue that hazes impose more adverse impacts on high-frequency image features (e.g., outlines and textures) than low-frequency features (e.g., colors), with theoretical and empirical justifications. To better restore hazed high- and low-frequency features, we decompose the hazed images into high- and low-frequency feature components with spectral dual-channel encoding and respectively design effective neural network architectures to recover hazed images on the two feature components. To be specific, we recover the low-frequency feature components with an encoder-decoder, while we specially design a high-frequency aggregation component (HFAC) to recover hazed images on high-frequency feature components, by referring to neighboring feature distributions. We conduct extensive experiments on four real-world image dehazing benchmarks. The experimental results show that our proposed SDCE framework outperforms the state-of-the-art baselines significantly, with an average 4.4% improvement in PSNR and an average 7.7% gain in SSIM.

Ride-hailing service has witnessed a dramatic growth over the past decade but meanwhile raised various challenging issues, one of which is how to provide a timely and accurate short-term prediction of supply and demand. While the predictions for zone-based demand have been extensively studied, much less efforts have been paid to the predictions for origin-destination (OD) based demand (namely, demand originating from one zone to another). However, OD-based demand prediction is even more important and worth further explorations, since it provides more elaborate trip information in the near future as reference for fine-grained operations, such as the routing and matching of shared ride-hailing services that pick up and drop off two or more passengers in each ride. Simultaneous prediction of both zone-based and OD-based demand can be an interesting and practical problem for the ride-hailing platforms. To address the issue, we propose a multi-task matrix factorized graph neural network (MT-MF-GCN), which consists of two major components: (1) a GCN (graph convolutional network) basic module that captures the spatial correlations among zones via mixture-model graph convolutional (MGC) network, and (2) a matrix factorization module for multi-task predictions of zone-based and OD-based demand. By evaluations on the real-world on-demand data in Manhattan and Haikou, we show that the proposed model outperforms the state-of-the-art baseline methods in both zone- and OD-based predictions.