Recent developments in miniaturization of computing devices, in location‐sensing technology and in ubiquitous short‐range wireless networks enable new types of social behaviour. This paper investigates one novel application of these technologies, ad hoc inner‐urban shared‐ride trip planning: Transportation clients such as pedestrians are seeking ad hoc shared rides from transportation hosts such as private automobiles, buses, taxi cabs or trains. While centralized trip planners are challenged by assigning clients and hosts in an ad hoc manner, in particular for non‐scheduled hosts, we consider the transportation network as a mobile geosensor network of agents that interact locally by short‐range communication and heuristic wayfinding strategies. This approach is not only fully scalable; we can also demonstrate that with short‐range communication, and hence, incomplete transportation network knowledge a system still can deliver near‐to‐optimal trips.

/pdf/ad-hoc-shared-ride-trip-planning-by-mobile-geosensor-112uwbaclk.pdf

Ad Hoc Shared-Ride Trip Planning by Mobile Geosensor Networks

This article proposes a cell-based multi-class dynamic traffic assignment problem that considers the random evolution of traffic states. Travelers are assumed to select routes based on perceived effective travel time, where effective travel time is the sum of mean travel time and safety margin. The proposed problem is formulated as a fixed point problem, which includes a Monte-Carlo-based stochastic cell transmission model to capture the effect of physical queues and the random evolution of traffic states during flow propagation. The fixed point problem is solved by the self-regulated averaging method. The results illustrate the properties of the problem and the effectiveness of the solution method. The key findings include the following: (1) Reducing perception errors on traffic conditions may not be able to reduce the uncertainty of estimating system performance, (2) Using the self-regulated averaging method can give a much faster rate of convergence in most test cases compared with using the method of successive averages, (3) The combination of the values of the step size parameters highly affects the speed of convergence, (4) A higher demand, a better information quality, or a higher degree of the risk aversion of drivers can lead to a higher computation time, (5) More driver classes do not necessarily result in a longer computation time, and (6) Computation time can be significantly reduced by using small sample sizes in the early stage of solution processes.

/pdf/a-cell-based-model-for-multi-class-doubly-stochastic-dynamic-35gxmhgow7.pdf

A Cell-Based Model for Multi-class Doubly Stochastic Dynamic Traffic Assignment

Intelligent transport systems provide various means to improve traffic congestion in road networks. Evaluation of the benefits of these improvements requires consideration of commuters’ response to reliability and/or uncertainty of travel time under various circumstances. Various disruptions cause recurrent or non-recurrent congestion on road networks, which make road travel times intrinsically fluctuating and unpredictable. Confronted with such uncertain traffic conditions, commuters are known to develop some simple decision-making process to adjust their travel choices. This paper represents the decision-making process involved in departure-time and route choices as risk-taking behavior under uncertainty. An expected travel disutility function associated with commuters’ departure-time and route choices is formulated with taking into account the travel delay (due the recurrent congestion), the uncertainty of travel times (due to incident-induced congestion) and the consequent early or late arrival penalty. Commuters are assumed to make decision on the departure-time and route choices on the basis of the minimal expected travel disutility. Thus the network will achieve a simultaneous route and departure-time user equilibrium, in which no commuter can decrease his or her expected disutility by unilaterally changing the route or departure-time. The equilibrium is further formulated as an equivalent nonlinear complementarity problem and is then converted into an unconstrained minimization problem with the use of a gap function suggested recently. Two algorithms based on the Nelder–Mead multidimensional simplex method and the heuristic route/time-swapping approach, are adapted to solve the problem. Finally, numerical example is given to illustrate the application of the proposed model and algorithms.

New technology and the modeling of risk-taking behavior in congested road networks

Travel time variability is increasingly being recognized as a major factor influencing travel decisions and, consequently, as an important performance measure in transportation management. In this research project, we examine a number of questions related to travel time variability: How should travel time variability be quantified at both the section level as well as at the route level?; How do travelers value travel time and its reliability?; How much does the travel time reliability contribute to travelers' route choices?; How much variation is there in travelers' preferences regarding the potential tradeoff between reliability and travel time itself?; How can travel time variability be incorporated into the route choice models for transportation planning purposes?; and, How can the effects of travel time reliability be incorporated in considering risk-taking behavior in route choice models? Answering these questions can help in the design and evaluation of transportation planning and managing strategies.

/pdf/considering-risk-taking-behavior-in-travel-time-reliability-3efsqu6kpj.pdf

Considering Risk-Taking Behavior in Travel Time Reliability

Dynamic Traffic Assignment (DTA) has been studied for more than four decades and numerous reviews of this research area have been conducted. This review focuses on the travel choice principle and the classification of DTA models, and is supplementary to the existing reviews. The implications of the travel choice principle for the existence and uniqueness of DTA solutions are discussed, and the interrelation between the travel choice principle and the traffic flow component is explained using the nonlinear complementarity problem, the variational inequality problem, the mathematical programming problem, and the fixed point problem formulations. This paper also points out that all of the reviewed travel choice principles are extended from those used in static traffic assignment. There are also many classifications of DTA models, in which each classification addresses one aspect of DTA modeling. Finally, some future research directions are identified.

/pdf/dynamic-traffic-assignment-model-classifications-and-recent-4xqr697r45.pdf

Dynamic traffic assignment: model classifications and recent advances in travel choice principles

Dynamic traffic assignment (DTA) has been a topic of substantial research during the past decade and the DTA models offer the potential to support effective evaluation and operation Advanced Transportation Management Systems (ATMS) and Advanced Transportation Information Systems (ATIS). The model presented in this paper is a stochastic dynamic user optimal model based on a stochastic dynamic network. The assumption is that route travel times are variable and perceived as such by travelers at each time instant. Route choices are assumed to be made with the objective of minimizing perceived disutilities at each time. These perceived disutilities depend on the distribution of the variable route travel times, the distribution of individual perception errors and the individual traveler's risk taking nature at each time.

Considering travelers' risk-taking behavior in dynamic traffic assignment

Efficient Transformers have been developed for long sequence modeling, due to their subquadratic memory and time complexity. Sparse Transformer is a popular approach to improving the efficiency of Transformers by restricting self-attention to locations specified by the predefined sparse patterns. However, leveraging sparsity may sacrifice expressiveness compared to full-attention, when important token correlations are multiple hops away. To combine advantages of both the efficiency of sparse transformer and the expressiveness of full-attention Transformer, we propose Diffuser, a new state-of-the-art efficient Transformer. Diffuser incorporates all token interactions within one attention layer while maintaining low computation and memory costs. The key idea is to expand the receptive field of sparse attention using Attention Diffusion, which computes multi-hop token correlations based on all paths between corresponding disconnected tokens, besides attention among neighboring tokens. Theoretically, we show the expressiveness of Diffuser as a universal sequence approximator for sequence-to-sequence modeling, and investigate its ability to approximate full-attention by analyzing the graph expander property from the spectral perspective. Experimentally, we investigate the effectiveness of Diffuser with extensive evaluations, including language modeling, image modeling, and Long Range Arena (LRA). Evaluation results show that Diffuser achieves improvements by an average of 0.94% on text classification tasks and 2.30% on LRA, with 1.67x memory savings compared to state-of-the-art benchmarks, which demonstrates superior performance of Diffuser in both expressiveness and efficiency aspects.

Diffuser: Efficient Transformers with Multi-hop Attention Diffusion for Long Sequences

Coreference resolution over semantic graphs like AMRs aims to group the graph nodes that represent the same entity. This is a crucial step for making document-level formal semantic representations. With annotated data on AMR coreference resolution, deep learning approaches have recently shown great potential for this task, yet they are usually data hunger and annotations are costly. We propose a general pretraining method using variational graph autoencoder (VGAE) for AMR coreference resolution, which can leverage any general AMR corpus and even automatically parsed AMR data. Experiments on benchmarks show that the pretraining approach achieves performance gains of up to 6% absolute F1 points. Moreover, our model significantly improves on the previous state-of-the-art model by up to 11% F1.

/pdf/variational-graph-autoencoding-as-cheap-supervision-for-amr-1vemqs86.pdf

Variational Graph Autoencoding as Cheap Supervision for AMR Coreference Resolution

Encoding long sequences in Natural Language Processing (NLP) is a challenging problem. Though recent pretraining language models achieve satisfying performances in many NLP tasks, they are still restricted by a pre-defined maximum length, making them challenging to be extended to longer sequences. So some recent works utilize hierarchies to model long sequences. However, most of them apply sequential models for upper hierarchies, suffering from long dependency issues. In this paper, we alleviate these issues through a graph-based method. We first chunk the sequence with a fixed length to model the sentence-level information. We then leverage graphs to model intra- and cross-sentence correlations with a new attention mechanism. Additionally, due to limited standard benchmarks for long document classification (LDC), we propose a new challenging benchmark, totaling six datasets with up to 53k samples and 4034 average tokens’ length. Evaluation shows our model surpasses competitive baselines by 2.6% in F1 score, and 4.8% on the longest sequence dataset. Our method is shown to outperform hierarchical sequential models with better performance and scalability, especially for longer sequences.

I Y Li

Papers

Considering travelers' risk-taking behavior in dynamic traffic assignment

Considering travelers' risk-taking behavior in dynamic traffic assignment

Diffuser: Efficient Transformers with Multi-hop Attention Diffusion for Long Sequences

Variational Graph Autoencoding as Cheap Supervision for AMR Coreference Resolution

HiPool: Modeling Long Documents Using Graph Neural Networks