Ziyuan Gu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Ziyuan Gu has a total of 117 co-authors. Here are the top co-authors by number of shared publications:
- Meead Saberi
University of New South Wales
16 shared publications - Zhiyuan Liu
Jiangsu Key Laboratory of Urban ITS, Jiangsu Province Collaborative Innovation Center of Modern Urban Traffic Technologies, School of Transportation, Southeast University, Nanjing, China
15 shared publications Zhiyuan Liu
12 shared publicationsYunyang Shi
Zhejiang Communications Investment Group Co., Ltd
5 shared publications- Qixiu Cheng
University of Bristol Business School, University of Bristol, Bristol, UK
4 shared publications Majid Sarvi
3 shared publications
Latest Publications
- 23 February 2026
- Published by Elsevier in Transportation Research Part C: Emerging Technologies
- Vol. 185
HMP-RL: Hierarchical multi-Policy reinforcement learning for truck platooning
- 23 February 2026
- Published by Elsevier in Transportation Research Part C: Emerging Technologies
- Vol. 185
HMP-RL: Hierarchical multi-Policy reinforcement learning for truck platooning
- Open Access
Let Experts Feel Uncertainty: A Multi-Expert Label Distribution Approach to Probabilistic Time Series Forecasting
Time series forecasting in real-world applications requires both high predictive accuracy and interpretable uncertainty quantification. Traditional point prediction methods often fail to capture the inherent uncertainty in time series data, while existing probabilistic approaches struggle to balance computational efficiency with interpretability. We propose a novel Multi-Expert Learning Distributional Labels (LDL) framework that addresses these challenges through mixture-of-experts architectures with distributional learning capabilities. Our approach introduces two complementary methods: (1) Multi-Expert LDL, which employs multiple experts with different learned parameters to capture diverse temporal patterns, and (2) Pattern-Aware LDL-MoE, which explicitly decomposes time series into interpretable components (trend, seasonality, changepoints, volatility) through specialized sub-experts. Both frameworks extend traditional point prediction to distributional learning, enabling rich uncertainty quantification through Maximum Mean Discrepancy (MMD). We evaluate our methods on aggregated sales data derived from the M5 dataset, demonstrating superior performance compared to baseline approaches. The continuous Multi-Expert LDL achieves the best overall performance, while the Pattern-Aware LDL-MoE provides enhanced interpretability through component-wise analysis. Our frameworks successfully balance predictive accuracy with interpretability, making them suitable for real-world forecasting applications where both performance and actionable insights are crucial.
Open AccessLet Experts Feel Uncertainty: A Multi-Expert Label Distribution Approach to Probabilistic Time Series Forecasting
Time series forecasting in real-world applications requires both high predictive accuracy and interpretable uncertainty quantification. Traditional point prediction methods often fail to capture the inherent uncertainty in time series data, while existing probabilistic approaches struggle to balance computational efficiency with interpretability. We propose a novel Multi-Expert Learning Distributional Labels (LDL) framework that addresses these challenges through mixture-of-experts architectures with distributional learning capabilities. Our approach introduces two complementary methods: (1) Multi-Expert LDL, which employs multiple experts with different learned parameters to capture diverse temporal patterns, and (2) Pattern-Aware LDL-MoE, which explicitly decomposes time series into interpretable components (trend, seasonality, changepoints, volatility) through specialized sub-experts. Both frameworks extend traditional point prediction to distributional learning, enabling rich uncertainty quantification through Maximum Mean Discrepancy (MMD). We evaluate our methods on aggregated sales data derived from the M5 dataset, demonstrating superior performance compared to baseline approaches. The continuous Multi-Expert LDL achieves the best overall performance, while the Pattern-Aware LDL-MoE provides enhanced interpretability through component-wise analysis. Our frameworks successfully balance predictive accuracy with interpretability, making them suitable for real-world forecasting applications where both performance and actionable insights are crucial.
- 23 January 2026
- Published by Elsevier in Computers & Industrial Engineering
- Vol. 214
Data-driven departure time and traffic assignment model for intercity multimodal transport system
- 23 January 2026
- Published by Elsevier in Computers & Industrial Engineering
- Vol. 214
Data-driven departure time and traffic assignment model for intercity multimodal transport system
- Open Access
- 30 December 2025
- Published by Elsevier in The Innovation
- Vol. 7 (4)
Large-scale transportation governance: A tensorization-parallelization co-empowered framework
- 30 December 2025
- Published by Elsevier in The Innovation
- Vol. 7 (4)
Open AccessLarge-scale transportation governance: A tensorization-parallelization co-empowered framework
- 20 December 2025
- Published by Elsevier in Transportation Research Part C: Emerging Technologies
- Vol. 183
A hybrid ranking and selection procedure to solve simulation-based discrete network design problems
- 20 December 2025
- Published by Elsevier in Transportation Research Part C: Emerging Technologies
- Vol. 183
A hybrid ranking and selection procedure to solve simulation-based discrete network design problems
- 1 November 2025
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Transportation Science
- Vol. 59 (6) , 1353-1374
A Heteroscedastic Robust Bayesian Optimization Method for Solving Simulation-Based Transportation Problems
This study focuses on simulation-based optimization (SBO) in transportation systems considering the pervasive and influential heteroscedastic noise. Existing studies rarely consider the effects of such heteroscedasticity on the solution robustness, giving rise to suboptimal solutions that could compromise the reliability and resilience of the system in real-world applications. To address this concern, a simulation-based robust optimization problem is investigated in this study, which focuses on minimizing the expectation of simulation outputs while maintaining the stochasticity of transportation systems within predefined limits. To solve the problem and identify a robust solution under varying levels of stochasticity, a heteroscedastic robust Bayesian optimization (HRBO) method is proposed by fusing key SBO concepts and techniques with the widely used Bayesian optimization (BO) algorithm. The formulation of surrogate models, strategies for sampling new points, and evaluation issues of samples are systematically designed. Specifically, surrogate models for the stochastic objective and constraint functions are separately formulated using the Gaussian process (GP) model. To accommodate simulation noise, Bayesian posterior inference is employed to estimate objective function values and constraint function values, which are incorporated into the GP models. To locate promising feasible solutions, a constrained expected improvement (EI) function is constructed and optimized using a tailored two-stage method, which can effectively tackle the inherent issue of “flat” areas of EI functions. Considering the usually high computational cost of simulators, an adaptive simulation resource allocation scheme is designed by incorporating ranking and selection techniques into the BO framework to efficiently allocate computational resources. The proposed methods are validated on a test function and two representative simulation-based transportation problems: a variant of the M/M/1 queueing problem and a continuous network design problem. Experimental results demonstrate the superior performance of HRBO in addressing heteroscedastic noise and identifying robust solutions. Funding: This work was supported by the National Natural Science Foundation of China [Grants 52131203 and 72471057], the Jiangsu Provincial Scientific Research Center of Applied Mathematics [Grant BK20233002], and the Natural Science Foundation of Jiangsu Province [Grant BK20232019]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/trsc.2024.0840 .
- 1 November 2025
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Transportation Science
- Vol. 59 (6) , 1353-1374
A Heteroscedastic Robust Bayesian Optimization Method for Solving Simulation-Based Transportation Problems
This study focuses on simulation-based optimization (SBO) in transportation systems considering the pervasive and influential heteroscedastic noise. Existing studies rarely consider the effects of such heteroscedasticity on the solution robustness, giving rise to suboptimal solutions that could compromise the reliability and resilience of the system in real-world applications. To address this concern, a simulation-based robust optimization problem is investigated in this study, which focuses on minimizing the expectation of simulation outputs while maintaining the stochasticity of transportation systems within predefined limits. To solve the problem and identify a robust solution under varying levels of stochasticity, a heteroscedastic robust Bayesian optimization (HRBO) method is proposed by fusing key SBO concepts and techniques with the widely used Bayesian optimization (BO) algorithm. The formulation of surrogate models, strategies for sampling new points, and evaluation issues of samples are systematically designed. Specifically, surrogate models for the stochastic objective and constraint functions are separately formulated using the Gaussian process (GP) model. To accommodate simulation noise, Bayesian posterior inference is employed to estimate objective function values and constraint function values, which are incorporated into the GP models. To locate promising feasible solutions, a constrained expected improvement (EI) function is constructed and optimized using a tailored two-stage method, which can effectively tackle the inherent issue of “flat” areas of EI functions. Considering the usually high computational cost of simulators, an adaptive simulation resource allocation scheme is designed by incorporating ranking and selection techniques into the BO framework to efficiently allocate computational resources. The proposed methods are validated on a test function and two representative simulation-based transportation problems: a variant of the M/M/1 queueing problem and a continuous network design problem. Experimental results demonstrate the superior performance of HRBO in addressing heteroscedastic noise and identifying robust solutions. Funding: This work was supported by the National Natural Science Foundation of China [Grants 52131203 and 72471057], the Jiangsu Provincial Scientific Research Center of Applied Mathematics [Grant BK20233002], and the Natural Science Foundation of Jiangsu Province [Grant BK20232019]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/trsc.2024.0840 .
- 19 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 36 (11) , 19560-19573
Leveraging Semi-Supervised Learning and Meta-Learning for Re-Identification in Few-Shot Spatiotemporal Anomaly Detection
Detecting spatiotemporal anomalies is imperative for addressing critical societal and engineering challenges, including public safety assurance, environmental hazard identification, epidemic surveillance, and transportation system optimization. Existing methodologies, however, face persistent limitations due to sparse labeled datasets and the inherent complexity of dynamic spatiotemporal systems. In order to bridge this gap, we present unsupervised-semi-supervised stacking (USemiS), a novel framework that synergizes semi-supervised learning with ensemble meta-learning. USemiS introduces three core innovations: 1) unsupervised component learners that extract low-level representations of heterogeneous anomalies, 2) a consensus-based tuning mechanism that dynamically weights robust learners via stability metrics, and 3) spatiotemporal MixUp (ST-MixUp), a tailored augmentation strategy that interpolates anomalies across spatial and temporal dimensions to enhance decision boundaries. By integrating these components, USemiS effectively disentangles latent anomaly patterns while mitigating label scarcity. Evaluated on large-scale traffic anomaly and crowd fall detection datasets, USemiS achieves state-of-the-art performance, outperforming existing methods by 1.3% and 2.1% in AUC under extreme low-label regimes (0.4% and 0.8% labeled data, respectively). These results underscore USemiS’s capacity to generalize across diverse spatiotemporal contexts, offering a scalable and robust solution for real-world applications where labeled anomalies are scarce yet critical.
- 19 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 36 (11) , 19560-19573
Leveraging Semi-Supervised Learning and Meta-Learning for Re-Identification in Few-Shot Spatiotemporal Anomaly Detection
Detecting spatiotemporal anomalies is imperative for addressing critical societal and engineering challenges, including public safety assurance, environmental hazard identification, epidemic surveillance, and transportation system optimization. Existing methodologies, however, face persistent limitations due to sparse labeled datasets and the inherent complexity of dynamic spatiotemporal systems. In order to bridge this gap, we present unsupervised-semi-supervised stacking (USemiS), a novel framework that synergizes semi-supervised learning with ensemble meta-learning. USemiS introduces three core innovations: 1) unsupervised component learners that extract low-level representations of heterogeneous anomalies, 2) a consensus-based tuning mechanism that dynamically weights robust learners via stability metrics, and 3) spatiotemporal MixUp (ST-MixUp), a tailored augmentation strategy that interpolates anomalies across spatial and temporal dimensions to enhance decision boundaries. By integrating these components, USemiS effectively disentangles latent anomaly patterns while mitigating label scarcity. Evaluated on large-scale traffic anomaly and crowd fall detection datasets, USemiS achieves state-of-the-art performance, outperforming existing methods by 1.3% and 2.1% in AUC under extreme low-label regimes (0.4% and 0.8% labeled data, respectively). These results underscore USemiS’s capacity to generalize across diverse spatiotemporal contexts, offering a scalable and robust solution for real-world applications where labeled anomalies are scarce yet critical.
- 18 July 2025
- Published by Elsevier in Transportation Research Part C: Emerging Technologies
- Vol. 179
TRIP: Transport reasoning with intelligence progression — A foundation framework
- 18 July 2025
- Published by Elsevier in Transportation Research Part C: Emerging Technologies
- Vol. 179
TRIP: Transport reasoning with intelligence progression — A foundation framework
- 16 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Knowledge and Data Engineering
- Vol. 37 (9) , 5666-5678
Topological Information Utilization in Label Enhancement and Label Distribution Learning Based on Optimal Transport Theory
Label Distribution Learning (LDL) offers a promising solution to label ambiguity by employing Label Distributions (LDs) instead of traditional logical labels. However, acquiring LDs for real-world data is both expensive and challenging. To address this issue, Label Enhancement (LE) techniques have been proposed to derive LDs from readily available logical labels. While much of the prior work has focused on enhancing LE for better recovery performance, the ultimate objective remains improving LDL’s overall effectiveness. In this paper, we introduce a novel LE method, Topological Label Enhancement via Optimal Transport (TLEOT), which integrates Optimal Transport (OT) theory with topological space analysis. This method goes beyond improving LE, targeting the enhancement of LDL performance by aligning the feature and label distributions within a unified topological framework. Additionally, we present two innovative topological techniques designed to further improve LDL. Extensive experimental evaluations on real-world datasets demonstrate that TLEOT consistently outperforms nine state-of-the-art methods in predictive tasks. Furthermore, the proposed topological techniques significantly enhance LDL’s performance, validating their practical utility in real-world applications.
- 16 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Knowledge and Data Engineering
- Vol. 37 (9) , 5666-5678
Topological Information Utilization in Label Enhancement and Label Distribution Learning Based on Optimal Transport Theory
Label Distribution Learning (LDL) offers a promising solution to label ambiguity by employing Label Distributions (LDs) instead of traditional logical labels. However, acquiring LDs for real-world data is both expensive and challenging. To address this issue, Label Enhancement (LE) techniques have been proposed to derive LDs from readily available logical labels. While much of the prior work has focused on enhancing LE for better recovery performance, the ultimate objective remains improving LDL’s overall effectiveness. In this paper, we introduce a novel LE method, Topological Label Enhancement via Optimal Transport (TLEOT), which integrates Optimal Transport (OT) theory with topological space analysis. This method goes beyond improving LE, targeting the enhancement of LDL performance by aligning the feature and label distributions within a unified topological framework. Additionally, we present two innovative topological techniques designed to further improve LDL. Extensive experimental evaluations on real-world datasets demonstrate that TLEOT consistently outperforms nine state-of-the-art methods in predictive tasks. Furthermore, the proposed topological techniques significantly enhance LDL’s performance, validating their practical utility in real-world applications.