Road traffic crashes are a major global cause of fatalities and serious injuries. Identifying safety-critical links during transportation planning can help mitigate these risks proactively. While conventional safety approaches rank links based on crash frequency, this study introduces a network vulnerability perspective to assess link importance. This new perspective shifts the focus from evaluating the local impact of crashes on individual links to assessing each link’s contribution to overall network safety. Unlike previous research that evaluates link safety contribution in isolation, this study adopts a cooperative game theory framework to account for cooperative interactions among links. The Shapley value, a solution concept from cooperative game theory, is employed to quantify the importance of each link, treating links as players. To account for the flow-dependent nature of road safety, we utilize a safety evaluation metric that calculates the utility of a cooperative game within a stochastic user equilibrium (SUE) model. This approach calculates the average marginal contribution of links to network safety across all possible link coalitions, accounting for traffic interactions rather than considering only their marginal contribution (MC) to the grand coalition. Numerical experiments highlight the advantages of the proposed approach by comparing Shapley value-based rankings with those derived from MC alone. Results show that the Shapley value more comprehensively captures link-level safety contributions, offering planners a useful framework for identifying critical links and prioritizing interventions to enhance network safety during the planning stage.

Bus electrification is a key trend in the global evolution of public transportation systems. However, replacing diesel buses (DBs) with Battery electric buses (BEBs) is a long-term process, where the limited driving range and prolonged charging times might necessitate a larger BEB fleet to maintain trip services compared to the replaced DB fleet. To quantify this fleet expansion across variable replacement decisions, we introduce the fleet increment curve (FIC), a novel conceptual idea that guides BEB procurement decisions during the transition to the zero-emission bus (ZEB) system. First, the FIC is derived from solving a series of mixed-integer linear programming (MILP) (namely MILP-FIC model) by varying the replaced DB fleet as inputs, where each MILP is developed by means of linearization techniques, while formulating the mixed-fleet operation under limited charging accessibility. To solve the MILP-FIC, Lagrangian relaxation (LR) is applied to relax charging accessibility constraints, decomposing the problem into route-specific subproblems. Subsequently, representing FIC by-products as piecewise linear functions enables extended models developed for addressing long-term fleet replacement scheduling and charging resource allocation. A general fleet replacement scheduling is presented, which accommodates multiple BEB types (varying battery capacities and charging power) by deriving type-specific fleet procurement curves. We use real-world bus route data from Hong Kong to explore the FIC, revealing how route characteristics—such as trip frequency, trip duration, and energy consumption—interact with charging site characteristics (e.g., siting and sizing) to shape the FIC. The curves typically follow a non-decreasing trend, while an S-shaped trend occurs across certain routes. Additionally, the results demonstrate the effective incorporation of FIC into the planning for ZEB transition, providing valuable insights for bus operators.

In response to the need for sustainable transportation, this study addresses the challenges of providing the charging service to fluctuating and geographically dispersed electric vehicles (EVs) charging demands. This work proposed a mixed EV charging system that incorporates mobile charging services to complement fixed charging stations. The overall coverage rate of charging services is enhanced under this charging system. To cope with the complexities of estimating the EV charging demands due to evolving adoption of EVs and uncertain external factors (e.g.,driver’s car use behaviors, fluctuations in electricity rates), a two-stage distributionally robust optimization (DRO) model is developed to determine the placement and capacity of fixed charging stations, the location of mobile depots, and the number of mobile trucks to deploy. Acase study on HongKong’s North District is conducted to demonstrate the effectiveness of the proposed model by solving it using a decomposition algorithm, through which the computational complexity is significantly reduced by iteratively solving the first andsecond stage problems. This study demonstrates that the mixed charging strategy significantly improves coverage, even with limited data, and highlights how investment levels affect performance. The mixed system improved coverage by 20% compared to fixed only infrastructure under the budget of 90million HKD. Additionally, the case study shows an increase of average coverage rate as budget level increases, which offers insights into the investment levels required to achieve a specific coverage rate. The contributions include a novel mixed charging strategy and a two-stage DRO model, providing a robust and scalable framework to support EV charging infrastructure planning under uncertainty.

Video generation models have made significant progress in simulating future states, showcasing their potential as world simulators in embodied scenarios. However, existing models often lack robust understanding, limiting their ability to perform multi-step predictions or handle Out-of-Distribution (OOD) scenarios. To address this challenge, we propose the Reflection of Generation (RoG), a set of intermediate reasoning strategies designed to enhance video prediction. It leverages the complementary strengths of pre-trained vision-language and video generation models, enabling them to function as a world model in embodied scenarios. To support RoG, we introduce Embodied Video Anticipation Benchmark(EVA-Bench), a comprehensive benchmark that evaluates embodied world models across diverse tasks and scenarios, utilizing both in-domain and OOD datasets. Building on this foundation, we devise a world model, Embodied Video Anticipator (EVA), that follows a multistage training paradigm to generate high-fidelity video frames and apply an autoregressive strategy to enable adaptive generalization for longer video sequences. Extensive experiments demonstrate the efficacy of EVA in various downstream tasks like video generation and robotics, thereby paving the way for large-scale pre-trained models in real-world video prediction applications. The video demos are available at https://sites.google.com/view/icml-eva.

Observed traffic data are widely recognized as an essential source of information for monitoring, evaluating, and planning transportation systems. The traffic sensor location problem is aimed at determining the optimal locations for collecting the most informative partial observations. This study focuses on the screen-line traffic counting location problem (SLTCLP). Screen lines are commonly used to validate traffic assignment results because of the ease of interpreting their positions. Therefore, addressing this problem is valuable for effective transportation management. Conventional solutions to this problem are based on path enumeration, which is computationally expensive and difficult to implement for large-scale transportation networks. Thus, we establish an exact and efficient solution method for the SLTCLP, using the concept of cut in graph theory and formulating the problem as the “cut optimization problem.” The proposed method is applied to different types of network instances, including a large network, and its performance and effectiveness are evaluated.

Urban Rail Transit (URT) accidents not only seriously affect the safety and reliability of its operations, but also reduce service level to passengers. Based on historical URT accident data, this study develops a hybrid ensemble learning model based on a Convolutional Neural Network (CNN) and Bidirectional Encoder Representations from Transformers (BERT) for predicting accident consequences in URT. The CNN is employed to capture spatial patterns from the diverse accident data, while the BERT is applied to learn complex relations in accident text descriptions. The results of the two models are combined for classifying accident consequences. The proposed hybrid ensemble learning model was applied to predict accident consequences in Chongqing’s URT using historical accident records. It achieved a prediction accuracy of 0.805 on testing data set, which is at least 20% higher than that of commonly used machine learning models, including multilayer perceptrons, support vector machines, and Bayesian networks. Furthermore, the reapplication of the proposed model to historical accident records of the URT in Chengdu demonstrates the generalizability and reusability of the model. This study forecasts the consequences of URT accidents with high accuracy using limited historical data, which supports operators in identifying high-frequency and high-impact accidents. Consequently, targeted maintenance and timely emergency response strategies can be developed to decrease accident rates and mitigate the impacts.

The Weibit choice model has gained increasing attention in transportation studies. Compared with the commonly used Logit model, the Weibit model inherently captures the heterogeneous travel perceptions by allowing non-identical variances for different alternatives. Nevertheless, how travelers’ heterogeneous perception errors may influence day-to-day (DTD) network dynamics, in which route choice decisions are made on each day, remains underexplored. In this study, we present several deterministic discrete DTD dynamic traffic models with Weibit-based network loading function, termed Weibit-based DTD dynamic models. We provide the asymptotic stability conditions of the Weibit stochastic equilibrium states based on the Jacobian matrices of the dynamical systems. We demonstrate how the features of non-identical perception variances and asymmetric response curve of the Weibit model can influence the evolution of network states and eventually the equilibrium points of the dynamical systems compared to the Logit case. Under fair comparison, the equilibrium states of Weibit DTD models are shown to have larger stable regions of adjustment rates than those of Logit DTD models. This research contributes to understanding the significance of considering travelers’ non-identical perception errors in DTD dynamics.

With the advances of Virtual Reality (VR) / Augmented Reality (AR), there arises a compelling need for transmission of point clouds over lossy channels (e.g., a 5G millimeter wave (mmWave) link that tends to be easily blocked). In this paper, we revisit the traditional Multiple Description Coding (MDC) concept and propose a simple point cloud MDC scheme that takes advantage of voxelization and is built upon a typical geometric point cloud compression codec. Our simulation study demonstrates the efficacy of the proposed scheme, as well as the tradeoff between compression efficiency and point cloud quality gain offered by MDC.

Uncertainty estimation is crucial for deep learning models to detect out-of-distribution (OOD) inputs. However, the naive deep learning classifiers produce uncalibrated uncertainty for OOD data. Improving the uncertainty estimation typically requires external data for OOD-aware training or considerable costs to build an ensemble. In this work, we improve on uncertainty estimation without extra OOD data or additional inference costs using an alternative Split-Ensemble method. Specifically, we propose a novel subtask-splitting ensemble training objective where a task is split into several complementary subtasks based on feature similarity. Each subtask considers part of the data as in distribution while all the rest as OOD data. Diverse submodels can therefore be trained on each subtask with OOD-aware objectives, learning generalizable uncertainty estimation. To avoid overheads, we enable low-level feature sharing among submodels, building a tree-like Split-Ensemble architecture via iterative splitting and pruning. Empirical study shows Split-Ensemble, without additional computational cost, improves accuracy over a single model by 0.8%, 1.8%, and 25.5% on CIFAR-10, CIFAR-100, and Tiny-ImageNet, respectively. OOD detection for the same backbone and in-distribution datasets surpasses a single model baseline by 2.2%, 8.1%, and 29.6% in mean AUROC, respectively.

We extend congestion games to the setting where players need to make multiple joint choices with interactions in a hierarchical manner (termed joint congestion game). At each choice dimension, players are involved in a typical congestion game. This game has a feature that the output of one choice dimension serves as an input of another one, and the costs paid by players in different choice dimensions are interdependent. Focusing on the joint congestion game with destination and route choices (i.e., select which destination and which route to complete a trip), we show the existence and uniqueness of the Nash equilibrium under mild assumptions in a nonatomic game setting. Then we investigate the property of the general quantal response equilibrium (QRE) for the joint congestion game in which players have perception errors of their costs (characterized by a probabilistic distribution). The QRE condition for the joint congestion game is further extended to the case where the analyst has only incomplete information about players’ perceived costs. A specific cross moment QRE model using the mean and covariance information is accordingly developed to account for both the analyst’s and players’ imperfect information/perception. We present an equivalent convex program that promises a unique solution for the cross moment QRE model, and provide a polynomial algorithm to solve it. Numerical results illustrate the features of the developed model for the joint congestion game and demonstrate the efficiency of the solution algorithm on two realistic transportation networks.

Luyu Gao, Zhuyun Dai, Panupong Pasupat, Anthony Chen, Arun Tejasvi Chaganty, Yicheng Fan, Vincent Zhao, Ni Lao, Hongrae Lee, Da-Cheng Juan, Kelvin Guu. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Masked Autoencoders learn strong visual representations and achieve state-of-the-art results in several independent modalities, yet very few works have addressed their capabilities in multi-modality settings. In this work, we focus on point cloud and RGB image data, two modalities that are often presented together in the real world, and explore their meaningful interactions. To improve upon the cross-modal synergy in existing works, we propose Pi-MAE, a self-supervised pre-training framework that promotes 3D and 2D interaction through three aspects. Specifically, we first notice the importance of masking strategies between the two sources and utilize a projection module to complementarily align the mask and visible tokens of the two modalities. Then, we utilize a well-crafted two-branch MAE pipeline with a novel shared decoder to promote cross-modality interaction in the mask tokens. Finally, we design a unique cross-modal reconstruction module to enhance representation learning for both modalities. Through extensive experiments performed on large-scale RGB-D scene understanding benchmarks (SUN RGB-D and ScannetV2), we discover it is nontrivial to interactively learn point-image features, where we greatly improve multiple 3D detectors, 2D detectors, and few-shot classifiers by 2.9%, 6.7%, and 2.4%, respectively. Code is available at https://github.com/BLVLab/PiMAE.

The compression of satellite imagery remains an important research area as hundreds of terabytes of images are collected every day, which drives up storage and bandwidth costs. Although progress has been made in increasing the resolution of these satellite images, many downstream tasks are only interested in small regions of any given image. These areas of interest vary by task but, once known, can be used to optimize how information within the image is encoded. Whereas standard image encoding methods, even those optimized for remote sensing, work on the whole image equally, there are emerging methods that can be guided by saliency maps to focus on important areas. In this work we show how imagery preprocessing techniques driven by saliency maps can be used with traditional lossy compression coding standards to create variable rate image compression within a single large satellite image. Specifically, we use variable sized smoothing kernels that map to different quantized saliency levels to process imagery pixels in order to optimize downstream compression and encoding schemes.

We propose using surface and aerial shared autonomous electric vehicles (SAEVs) to improve the resilience of infrastructure and communities, or SAEV-R. In disruptive events, SAEVs can be temporarily deployed to evacuate and rescue at-risk populations, provide essential supplies and services to vulnerable households, and transport repair crews and equipment. We present a modeling framework for feasibility analysis and strategic planning associated with deploying SAEVs for disaster relief. The framework guides our examination of three scenarios: a hurricane-induced power outage, a pandemic-affected vulnerable population, and earthquake-damaged infrastructure. The results demonstrate the flexibility of the proposed framework and showcase the potential and versatility of SAEV-R systems to improve resilience.

The concentric circles (CC) map design is an alternative approach for schematically representing metro systems. Compared with traditional octo-linear maps, CC maps can effectively simplify the perception of a network by visually accenting circular line patterns. This design offers new insights into the schematic drawing of metro systems that can improve map readability and engagement. Automated mapping studies in the literature have mostly applied the traditional octo-linear design using optimization methods, where design criteria are modeled as constraints and/or objective functions in a constrained mixed-integer optimization program, whereas the automated CC map drawing approach has received less attention. In this article, we develop an automatic CC map drawing method by adopting map design criteria as a mixed-integer programming problem. Numerical experiments are conducted using (a) a simple network to illustrate the model procedure in detail, (b) two real-world metro networks in Vienna and Montréal to analyze the effects of the selected map center and parameter settings and (c) the Beijing subway to analyze the applicability of the proposed approach to large-scale metro networks.

In many cities, subways are expanding with new or extended lines being built and put into operations. The prediction of future travel demand in subway with the planned expansion is of significant importance because such information is crucial for new line planning and new network operations. In this study, we identify the determinant features from potential influential factors of passenger travel demand and develop a two-step model for predicting passenger travel demand in expanding subways. The proposed model is tested in an actual subway with a new line being put into operations, and achieves higher prediction accuracy than the benchmark models.

Anthony Chen, Pallavi Gudipati, Shayne Longpre, Xiao Ling, Sameer Singh. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Knowledge-dependent tasks typically use two sources of knowledge: parametric, learned at training time, and contextual, given as a passage at inference time.To understand how models use these sources together, we formalize the problem of knowledge conflicts, where the contextual information contradicts the learned information.Analyzing the behaviour of popular models, we measure their over-reliance on memorized information (the cause of hallucinations), and uncover important factors that exacerbate this behaviour.Lastly, we propose a simple method to mitigate over-reliance on parametric knowledge which minimizes hallucination and improves out-of-distribution generalization by 4% -7%.Our findings demonstrate the importance for practitioners to evaluate model tendency to hallucinate rather than read, and show that our mitigation strategy encourages generalization to evolving information (i.e., time-dependent queries).To encourage these practices, we have released our framework for generating knowledge conflicts.1

Posing reading comprehension as a generation problem provides a great deal of flexibility, allowing for open-ended questions with few restrictions on possible answers.However, progress is impeded by existing generation metrics, which rely on token overlap and are agnostic to the nuances of reading comprehension.To address this, we introduce a benchmark for training and evaluating generative reading comprehension metrics: MOdeling Correctness with Human Annotations.MOCHA contains 40K human judgement scores on model outputs from 6 diverse question answering datasets and an additional set of minimal pairs for evaluation.Using MOCHA, we train a Learned Evaluation metric for Reading Comprehension, LERC, to mimic human judgement scores.LERC outperforms baseline metrics by 10 to 36 absolute Pearson points on held-out annotations.When we evaluate robustness on minimal pairs, LERC achieves 80% accuracy, outperforming baselines by 14 to 26 absolute percentage points while leaving significant room for improvement.MOCHA presents a challenging problem for developing accurate and robust generative reading comprehension metrics. 1

In the Internet of Things (IoT) era, with the growing number of data sources, we need to face some challenges such as high cost of the cloud storage caused by large amounts of data. To minimize the communication time and enhance the performance, sending the entire large amount of data is not practical. Thus, it is appropriate to make use of edge computing, or data preprocessing on IoT gateways. In this paper, we propose a data reduction algorithm for the gateway of bridge vibration G-sensors. The data reduction algorithm is based on a pattern system, which is comprised of a pattern library and a pattern classifier. The pattern library is generated by using the K-means clustering method. The results show that the proposed approach is effective in data reduction and outlier detection for bridge vibration data collection on the IoT gateway.

Foveated rendering leverages human visual system to increase video quality under limited computing resources for Virtual Reality (VR). More specifically, it increases the frame rate and the video quality of the foveal vision via lowering the resolution of the peripheral vision. Optimizing foveated rendering systems is, however, not an easy task, because there are numerous parameters that need to be carefully chosen, such as the number of layers, the eccentricity degrees, and the resolution of the peripheral region. Furthermore, there is no standard and efficient way to evaluate the Quality of Experiment (QoE) of foveated rendering systems. In this paper, we propose a framework to compare the performance of different subjective assessment methods on foveated rendering systems. We consider two performance metrics: efficiency and consistency, using the perceptual ratio, which is the probability of the foveated rendering is perceivable by users. A regression model is proposed to model the relationship between the human perceived quality and foveated rendering parameters. Our comprehensive study and analysis reveal several insights: 1) there is no absolute superior subjective assessment method, 2) subjects need to make more observations to confirm the foveated rendering is imperceptible than perceptible, 3) subjects barely notice the foveated rendering with an eccentricity degree of 7.5 degrees+ and peripheral region of a resolution of 540p+, and 4) QoE levels are highly dependent on the individuals and scenes. Our findings are crucial for optimizing the foveated rendering systems for future VR applications.

Most stations and tracks in metro networks are irreplaceable due to daily operations. If any of them were disrupted, it would impact not only the individual metro line but also the whole metro network. Therefore, metro managers need to have a good understanding of alternative routes between each pair of stations in the metro network. In the event of incidents, metro managers can make use of this information to reroute passengers to minimize the impact of disruptions. This paper aims to develop a route diversity index to address two questions: “how many reasonable routes are there for passengers between any two stations in normal operations or in the event of a disruption?” and “which stations are most vulnerable (i.e., the largest impact to the overall metro network when they are disrupted)?” To implement this measure in practice, definitions of routes and route diversity and a solution algorithm based on characteristics of metro networks are described to calculate the route diversity index. To show proof of the concept, a simple network example and a real-world network based on the Beijing Metro network in China are presented to demonstrate the feasibility of the route diversity index and its application to a real-world metro network.

Emergency evacuation from airports is an important consideration given the continuing occurrence of both natural and human caused disasters. Unfortunately, the traditional evacuation-drill approach to prepare for emergency situations presents several practical challenges at the scale and magnitude required for airports. In this paper, we present an agent-based model (ABM) called exitus which is capable of determining the extent to which collective behavior and overall evacuation time of passenger groups is affected by changes in the built environment for large, complex structures. The model is unique because it explicitly considers the physical and psychological characteristics of individuals with disabilities. In our first experiment, several bomb simulation scenarios were conducted at an international airport using exitus. Several important findings were revealed including: 1) the importance of stairway and exit configurations; 2) the inherent weaknesses of the pier airport design in affecting timely evacuations; 3) who the most vulnerable groups of people are; 4) the particular risk engendered from crowded or complex building interiors for individuals with disabilities; and 5) the potential problems caused by locating explosive detection system machines near passenger processing areas. The results of a second exploratory experiment also revealed the importance of realistically modeling psychological attributes of individuals with disabilities and their potential impact on collective evacuation performance. Overall the findings demonstrated the model's ability to generate a common operational picture capable of guiding preparedness efforts for both public and private organizations encompassing a wide variety of professional endeavor.

Ongoing, rapid advances in three-dimensional (3D) printing technology are making it inexpensive for lay people to manufacture 3D objects. However, the lack of tools to help nontechnical users design interesting, complex objects represents a significant barrier preventing the public from benefitting from 3D printers. Previous work has shown that an evolutionary algorithm with a generative encoding based on developmental biology-a compositional pattern-producing network (CPPN)-can automate the design of interesting 3D shapes, but users collectively had to start each act of creation from a random object, making it difficult to evolve preconceived target shapes. In this paper, we describe how to modify that algorithm to allow the further evolution of any uploaded shape. The technical insight is to inject the distance to the surface of the object as an input to the CPPN. We show that this seeded-CPPN technique reproduces the original shape to an arbitrary resolution, yet enables morphing the shape in interesting, complex ways. This technology also raises the possibility of two new, important types of science: (1) It could work equally well for CPPN-encoded neural networks, meaning neural wiring diagrams from nature, such as the mouse or human connectome, could be injected into a neural network and further evolved via the CPPN encoding. (2) The technique could be generalized to recreate any CPPN phenotype, but substituting a flat CPPN representation for the rich, originally evolved one. Any evolvability extant in the original CPPN genome can be assessed by comparing the two, a project we take first steps toward in this paper. Overall, this paper introduces a method that will enable non-technical users to modify complex, existing 3D shapes and opens new types of scientific inquiry that can catalyze research on bio-inspired artificial intelligence and the evolvability benefits of generative encodings.

A modified distance-based genetic algorithm is proposed to solve the multi-objective automatic vehicle identification (AVI) reader location problem studied in this paper. The objectives are: (1) minimizing the number of AVI readers, (2) maximizing the coverage of origin-destination (O-D) pairs, and (3) maximizing the number of AVI readings. These three objectives are strategically designed to catch the maximum number of trips covering the maximum number of AVI readers. In order to study the trade-off among the three objectives, non-dominated solutions are retained and analyzed. The results show that there is a trade-off between the quality (measured by objectives 2 and 3) and cost (measured by objective 1) of coverage.

Path finding is a fundamental research topic in transportation due to its wide applications in transportation planning and intelligent transportation system (ITS). In transportation, the path finding problem is usually defined as the shortest path (SP) problem in terms of distance, time, cost, or a combination of criteria under a deterministic environment. However, in real life situations, the environment is often uncertain. In this paper, we develop a simulation-based genetic algorithm to find multi-objective paths in stochastic networks. Numerical experiments are presented to demonstrate the algorithm feasibility.