Decision-focused learning (DFL) offers an end-to-end approach to the predict-then-optimize (PO) framework by training predictive models directly on decision loss (DL), enhancing decision-making performance within PO contexts. However, the implementation of DFL poses distinct challenges. Primarily, DL can result in deviation from the physical significance of the predictions under limited data. Additionally, some predictive models are non-differentiable or black-box, which cannot be adjusted using gradient-based methods. To tackle the above challenges, we propose a novel framework, Decision-Focused Fine-tuning (DFF), which embeds the DFL module into the PO pipeline via a novel bias correction module. DFF is formulated as a constrained optimization problem that maintains the proximity of the DL-enhanced model to the original predictive model within a defined trust region. We theoretically prove that DFF strictly confines prediction bias within a predetermined upper bound, even with limited datasets, thereby substantially reducing prediction shifts caused by DL under limited data. Furthermore, the bias correction module can be integrated into diverse predictive models, enhancing adaptability to a broad range of PO tasks. Extensive evaluations on synthetic and real-world datasets, including network flow, portfolio optimization, and resource allocation problems with different predictive models, demonstrate that DFF not only improves decision performance but also adheres to fine-tuning constraints, showcasing robust adaptability across various scenarios.

In this paper, we identify and analyze a recurring training loss pattern, which we term the Epochal Sawtooth Phenomenon (ESP), commonly observed during training with adaptive gradient-based optimizers, particularly Adam optimizer. This pattern is characterized by a sharp drop in loss at the beginning of each epoch, followed by a gradual increase, resulting in a sawtooth-shaped loss curve. Through empirical observations, we demonstrate that while this effect is most pronounced with Adam, it persists, although less severely, with other optimizers such as RMSProp. We empirically analyze the mechanisms underlying ESP, focusing on key factors such as Adam’s $$\beta $$ parameters, batch size, data shuffling, and sample replacement. Our analysis shows that ESP arises from adaptive learning rate adjustments controlled by the second moment estimate. Additionally, we identify the “immediate re-exposure to samples” effect during data shuffling, which causes the model to learn or memorize more at the beginning of each epoch. We also find that smaller values of $$\beta _2$$ exacerbate ESP but can act as a form of regularization. While ESP is not necessarily indicative of overfitting, higher model capacity can amplify the phenomenon. To further support our analysis, we replicate ESP through a high-dimensional quadratic minimization task. We demonstrate that ESP can emerge even in simple optimization scenarios, reinforcing the generality of this pattern. The code for reproducing our experiments is available at https://github.com/qiliuchn/training-loss-pattern .

Accurate and timely detection of incident-induced congestion (IIC) is essential for mitigating its negative impact on traffic efficiency. Existing studies on IIC detection mainly focus on traffic flow on freeways and face challenges on urban roads due to the impacts of signal lights at intersections and diverse road networks. Additionally, the low penetration rate of probe vehicle trajectories poses another challenge. This study proposes a probe-vehicle-trajectory-based algorithm for IIC detection on urban roads at the movement and cycle levels. Two critical features (i.e., the average speed and the entrance time into the road segment) are defined to capture the characteristics of trajectory segments. A Vehicle Trajectory Polar Coordinate Transformation (VTPCT) method is proposed to differentiate anomalous trajectory segments (ATS) affected by IIC from normal ones, considering the periodicity of fixed signal timing at the intersections. Anomaly rates calculated from the identified ATS within a spatiotemporal window are introduced to reflect the movement-cycle-level traffic states. A two-stage algorithm framework is designed to enhance the algorithm’s adaptability to spatially sparse trajectories and diverse road networks. Experimental studies show that the proposed algorithm is applicable to trajectory data with low penetration rates and outperforms benchmarks of typical statistical and AI-based algorithms.

Queue length is one of the most commonly used indicators to evaluate traffic operation at signalized intersections. Many studies aim to estimate queue length using trajectory data, but this remains a challenge with spatially sparse trajectories. This paper introduces a probabilistic-based method for cycle-by-cycle estimation of queue length distribution and the point estimate in closed form using connected vehicle (CV) trajectories. The method is applicable to isolated signalized intersections with under-saturated traffic. It works well in a low CV penetration rate environment by exploiting the trajectories of both queued and non-queued CVs. Vehicle arrival rates and CV penetration rates are first estimated to capture the vehicle arrival pattern within a time of day (TOD) using the maximum likelihood estimation approach. The closed forms of the probability distribution of queue length are derived for each cycle, which are attractive for applications such as adaptive signal timing considering traffic uncertainty. The queue length with the highest probability is taken as the point estimate. Numerical and empirical studies demonstrate that the proposed method outperforms the benchmark method in estimating CV penetration rates, particularly at low penetration rates. In term of queue length estimation, the proposed method is also superior to existing methods in most scenarios and is especially effective for cycles with observed trajectories. Sensitivity analysis reveals that the method is robust to different demand levels, signal timings, and variability in arrival patterns with under-saturated traffic. Additionally, the suggested requirements for the number of collected trajectories are investigated to provide practical guidance.

Previous studies in predicting crash risks primarily associated the number or likelihood of crashes on a road segment with traffic parameters or geometric characteristics, usually neglecting the impact of vehicles’ continuous movement and interactions with nearby vehicles. Recent technology advances, such as Connected and Automated Vehicles (CAVs) and drones, are able to collect high-resolution trajectory data, which enable trajectory-based risk analysis. This study investigates a new vehicle group (VG) based risk analysis method and explores risk evolution mechanisms considering VG features. An impact-based vehicle grouping method is proposed to cluster vehicles into VGs by evaluating their responses to the erratic behaviors of nearby vehicles. The risk of a VG is aggregated based on the risk between each vehicle pair in the VG, measured by inverse Time-to-Collision (iTTC). Logistic Regression and a Graph Neural Network (GNN) are used to predict VG risks based on both aggregated and disaggregated VG information. Both methods achieve excellent performance with AUC values exceeding 0.93. For the GNN model, GNNExplainer with feature perturbation is applied to identify critical individual vehicle features and their directional impact on VG risks. Overall, this research contributes a new perspective for identifying, predicting, and interpreting traffic risks.

How (dis)similar are the learning trajectories of vision–language models and children? Recent modeling work has attempted to understand the gap between models’ and humans’ data efficiency by constructing models trained on less data, especially multimodal naturalistic data. However, such models are often evaluated on adult-level benchmarks, with limited breadth in language abilities tested, and without direct comparison to behavioral data. We introduce DevBench, a multimodal benchmark comprising seven language evaluation tasks spanning the domains of lexical, syntactic, and semantic ability, with behavioral data from both children and adults. We evaluate a set of vision–language models on these tasks, comparing models and humans on their response patterns, not their absolute performance. Across tasks, models exhibit variation in their closeness to human response patterns, and models that perform better on a task also more closely resemble human behavioral responses. We also examine the developmental trajectory of OpenCLIP over training, finding that greater training results in closer approximations to adult response patterns. DevBench thus provides a benchmark for comparing models to human language development. These comparisons highlight ways in which model and human language learning processes diverge, providing insight into entry points for improving language models.

As an emerging data source, the trajectories of probe vehicles can compensate for the deficiencies of high maintenance costs and low coverage ranges of infrastructure-based detectors (e.g., loop detectors). However, existing arterial signal coordination studies typically assume high-penetration-rate trajectories, which are difficult to achieve in reality. Utilizing low-penetration-rate vehicle trajectories for arterial signal timing with cyclic stochastic traffic demand remains a significant challenge. To address this issue, this study developed a nonlinear optimization model for arterial signal coordination that is applicable to low-penetration-rate vehicle trajectories. Offsets and green splits were optimized to minimize the average delay of probe vehicles on both major and minor roads. Probe vehicle trajectories across cycles were aggregated into one cycle to compensate for the low penetration rate of the trajectory data. The concepts and estimation of the sampled arrival pattern, sampled departure pattern, and transition period were proposed to capture the spatiotemporal progression of probe vehicles along the arterial with varying signal timings. A genetic algorithm (GA)-based solution algorithm was designed to solve the proposed model. Simulation studies validated the advantages of the proposed model over the models in Synchro Studio, MULTIBAND, and the simplified model without considering the transition period. The sensitivity analysis showed that: 1) number of sampled trajectories matters instead of the penetration rate; 2) required number of sampled trajectories increases approximately linearly with the number of intersections and the demand factor; and 3) proposed model is robust to the sampling interval that is no longer than 7 s.

Model-based traffic state estimation is used to reproduce traffic flow states from available observation data to assist traffic control and management. Owing to temporal and spatial observation limitations, numerous unknown traffic states and parameters exist in a nonlinear traffic flow model. These cannot be accurately estimated using filter methods with unavoidable parametric assumptions. In addition, the difficulty of estimation increases owing to an increased number of diverse traffic flow states and more unfavorable observation conditions at signalized intersections compared with those on the freeway. To overcome these problems, in this study, we developed switching state-space models to approximate the description of dynamic traffic flows at signalized intersections. By setting the traffic flow rate at the upstream and downstream boundaries of the road as observation data, we constructed a Bayesian learning framework in which all unknown variables in the models can be learned from the observed data. Finally, the utility of our method is demonstrated with the synthetic and the real NGSIM data. The result demonstrated that the proposed method could perform reasonably well in estimating the traffic flow dynamic process and states at signalized intersections, and could constitute an efficient scheme that avoids critical dependence on parametric calibrations and assumptions.

Existing unconventional intersection designs mainly apply to large intersections, such as the intersection of two major roads, in which there are numerous lanes on each leg and the allowed queue length is long. This paper presents a new unconventional intersection design to improve the operational efficiency of the intersections with limited traffic lanes and short distance for vehicular queuing, such as the intersections on collector roads. The new design combines the advantage of the tandem control and exit-lanes for left-turn control (CTE). An optimization model is established to maximize the capacity of the intersection under the CTE design in which the allocation of the mixed-usage lane, lane markings, and signal timings are integrated. A case study and extensive numerical analysis are conducted to evaluate the performance of the proposed model. Comparisons are made between the proposed CTE design and the other three designs, namely the tandem design, exit-lanes for left-turn design, and conventional design, under different geometric and traffic demand cases. The results show that CTE design has promising property when the left turn ratio is low to medium (<; 50%), and the queue length is limited (<; 200m).

With the looming promise of wireless recharging technology, electric vehicles (EVs) are going to be able to acquire energy while still in motion. This paper focuses on the optimal deployment of wireless recharging facilities on signalized arterials for EVs. To address this issue, a bi-objective model considering both traffic operation efficiency (i.e., traffic delay saving) and charging infrastructure utilization rate (i.e., electricity gain from charging) has been formulated. A modified cell transmission model (CTM) is used as a base to simulate traffic flow on an arterial with traffic signals. The cells in the CTM also serve as a potential installation site for wireless recharging facilities. The essential goal of this model is to maximize the recharging electricity for EVs traveling on arterials while maintaining low travel delay. Due to the complexity in solving the bi-objective model, heuristic approaches, such as genetic algorithm and particle swarm optimization, are employed. The numerical experiments based on real day-to-day traffic demand are executed. A Pareto set is obtained and a sensitivity analysis regarding recharging rate, investment, and minimum recharging region length is provided.

Arterial traffic signal coordination makes traffic flow more efficient and safer. This paper presents a partition-enabled multi-mode band (PM-BAND) model that is designed to solve the signal coordination problem for arterials with multiple modes, i.e., passenger cars and transit vehicles. The proposed method permits the progression bands to be broken if necessary and optimizes system partition and signal coordination in one unified framework. The impacts of traffic demand of passenger cars and transit vehicles as well as the geometry characteristics of the arterials are taken into account. Signal timings and waiting time of transit vehicles at stations are optimized simultaneously. The PM-BAND model is formulated as a mixed-integer linear program, which can be solved by the standard branch-and-bound technique. Numerical example results have demonstrated that the PM-BAND model can significantly reduce the average number of stops and delay compared with the other models, i.e., MAXBAND and MULTIBAND. Moreover, the progression bands generated by the PM-BAND model have a higher reliability and effectiveness.