LIN Shu 林 舒

LIN Shu 林 舒

Associate Professor

Publications

Conference Papers

RailSched: A Hybrid Decision Framework for Railway Scheduling Optimization

Published in International Conference on Distributed Artificial Intelligence, 2025

Railway transportation systems worldwide face increasing pressure to maximize capacity while maintaining safety and operational efficiency. Complex scheduling conflicts frequently arise due to the inherent strict constraints in railway networks, including track sharing, station capacity limitations, and various train operational requirements. Effective conflict resolution methods are crucial for maintaining reliable railway services and optimizing infrastructure utilization. To address these challenges, we propose the RailSched, a novel intelligent decision framework for railway scheduling optimization. Our research by providing an intelligent and multi-strategy approach that reduces dispatcher workload, minimizes conflicts, and improves overall system reliability. The key contributions of this paper are fourfold: 1) we propose a railway schedule environment model for simulation; 2) develop a diverse strategy library targeting conflicts; 3) present an intelligent algorithmic framework for policy decision making and 4) explore the integration of large language models(LLMs) for railway schedule, an optional AI-enhanced decision module. Our evaluation using real railway data validates RailSched’s effectiveness, achieving a 85.05% improvement over expert-designed solutions in the most complex scenario, highlighting its practical value for modern railway dispatching.

Recommended citation: Wentian Fan, Yongcheng Zeng, Siyu Xia, Junyan Shi, Shu Lin, Mengyao Zhang, Yiwei Guo, Xin Zhang and Haifeng Zhang, "RailSched: A Hybrid Decision Framework for Railway Scheduling Optimization," presented at International Conference on Distributed Artificial Intelligence, London, UK, 2025.
Download Paper | Download Bibtex

Adaptive Command: Real-Time Policy Adjustment via Language Models in StarCraft II

Published in International Conference on Distributed Artificial Intelligence, 2024

We present Adaptive Command, a novel framework integrating large language models (LLMs) with behavior trees for real-time strategic decision-making in StarCraft II. Our system focuses on enhancing human-AI collaboration in complex, dynamic environments through natural language interactions. The framework comprises: (1) an LLM-based strategic advisor, (2) a behavior tree for action execution, and (3) a natural language interface with speech capabilities. User studies demonstrate significant improvements in player decision-making and strategic adaptability, particularly benefiting novice players and those with disabilities. This work contributes to the field of real-time human-AI collaborative decision-making, offering insights applicable beyond RTS games to various complex decision-making scenarios.

Recommended citation: Weiyu Ma, Dongyu Xu, Shu Lin, Haifeng Zhang, and Jun Wang, "Adaptive Command: Real-Time Policy Adjustment via Language Models in StarCraft II," presented at International Conference on Distributed Artificial Intelligence, Singapore, Singapore, 2024.
Download Paper | Download Bibtex

AI-Olympics: Exploring the Generalization of Agents through Open Competitions

Published in IJCAI Demonstrations Track, 2024

Between 2021 and 2023, AI-Olympics–a series of online AI competitions, was hosted by the online evaluation platform Jidi in collaboration with the IJCAI committee. In these competitions, an agent is required to accomplish diverse sports tasks in a two-dimensional continuous world, while competing against an opponent. This paper provides a brief overview of the competition series and highlights notable findings. We aim to contribute insights to the field of multi-agent decision-making and explore the generalization of agents through engineering efforts.

Recommended citation: Chen Wang, Yan Song, Shuai Wu, Sa Wu, Ruizhi Zhang, Shu Lin, and Haifeng Zhang, "AI-Olympics: Exploring the Generalization of Agents through Open Competitions," presented at IJCAI Demonstrations Track, Jeju, South Korea, 2024.
Download Paper | Download Bibtex

Generating Efficient Solvers from Constraint Models

Published in ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering, 2021

Combinatorial problems (CPs) arise in many areas, and people use constraint solvers to automatically solve these problems. However, the state-of-the-art constraint solvers (e.g., Gecode and Chuffed) have overly complicated software architectures; they compute solutions inefficiently. This paper presents a novel and model-driven approach–SoGen–to synthesize efficient problem-specific solvers from constraint models. Namely, when users model a CP with our domain-specific language PDL (short for Problem Description Language), SoGen automatically analyzes various properties of the problem (e.g., search space, value boundaries, function monotonicity, and overlapping subproblems), synthesizes an efficient solver algorithm based on those properties, and generates a C program as the problem solver. PDL is unique because it can create solvers that resolve constraints via dynamic programming (DP) search. For evaluation, we compared the solvers generated by SoGen with two state-of-the-art constraint solvers: Gecode and Chuffed. PDL’s solvers resolved constraints more efficiently; they achieved up to 6,058x speedup over Gecode and up to 31,300x speedup over Chuffed. Additionally, we experimented with both SoGen and the state-of-the-art solver generator—Dominion. We found SoGen to generate solvers faster and the produced solvers are more efficient.

Recommended citation: Shu Lin, Na Meng, and Wenxin Li, "Generating Efficient Solvers from Constraint Models," presented at ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering, Athens, Greece, 2021.
Download Paper | Download Bibtex

PDL: Scaffolding Problem Solving in Programming Courses

Published in ACM Conference on Innovation and Technology in Computer Science Education, 2021

Programming tasks provide an opportunity for students to improve their problem-solving skills (PSS). However, when programming tasks are challenging, students could become demotivated and lose the opportunity to improve PSS in the process. To scaffold the difficulty of programming tasks and better motivate students to enhance PSS via coding, this paper introduces PDL (Problem Description Language). Given the natural-language description of a combinatorial optimization problem (COP), PDL requires students to describe (i) inputs, (ii) constraints, (iii) the optimization objective, and (iv) outputs, based on their problem comprehension. PDL then validates each problem description by (1) compiling a solution program from the description and (2) executing the generated program with predefined test cases. Based on the compiling and testing results, PDL provides feedback to students, and assists students to adjust their problem comprehension and improve problem descriptions. To evaluate PDL’s effectiveness in motivating students to fulfill challenging programming tasks, we conducted a user study with 185 undergraduates and asked the students to solve COPs with or without PDL. We found that the students with PDL were less likely to give up than students without PDL. By using PDL, students solved more COPs and spent less time on each problem; they became more confident and motivated in handling COPs after using PDL.

Recommended citation: Shu Lin, Na Meng, Dennis Kafura, and Wenxin Li, "PDL: Scaffolding Problem Solving in Programming Courses," presented at ACM Conference on Innovation and Technology in Computer Science Education, Paderborn, Germany, 2021.
Download Paper | Download Bibtex

Optimizing Constraint Solving via Dynamic Programming

Published in International Joint Conference on Artificial Intelligence, 2019

Combinatorial problems (CPs) arise in many areas, and people use constraint solvers to automatically solve these problems. However, the state-of-the-art constraint solvers (e.g., Gecode and Chuffed) have overly complicated software architectures; they compute solutions inefficiently. This paper presents a novel and model-driven approach–SoGen–to synthesize efficient problem-specific solvers from constraint models. Namely, when users model a CP with our domain-specific language PDL (short for Problem Description Language), SoGen automatically analyzes various properties of the problem (e.g., search space, value boundaries, function monotonicity, and overlapping subproblems), synthesizes an efficient solver algorithm based on those properties, and generates a C program as the problem solver. SoGen is unique because it can create solvers that resolve constraints via dynamic programming (DP) search. For evaluation, we compared the solvers generated by SoGen with two state-of-the-art constraint solvers: Gecode and Chuffed. SoGen’s solvers resolved constraints more efficiently; they achieved up to 6,058x speedup over Gecode and up to 31,300x speedup over Chuffed. Additionally, we experimented with both SoGen and the state-of-the-art solver generator–Dominion. We found SoGen to generate solvers faster and the produced solvers are more efficient.

Recommended citation: Shu Lin, Na Meng, and Wenxin Li, "Optimizing Constraint Solving via Dynamic Programming," presented at International Joint Conference on Artificial Intelligence, Macao, China, 2019.
Download Paper | Download Bibtex

Object Representation in Angry Birds Game

Published in IJCAI Symposium on AI in Angry Birds, 2013

In the Angry Birds game, there are dynamic objects and static objects. Dynamic objects are usually convex polygons, while static objects can be concave. To represent dynamic objects, it is more accurate to use bounding convex polygons (BCPs) than Axis-aligned minimum bounding boxes (AABBs). BCPs can be roughly detected from AABBs, and some of them can be fitted into rectangles. We take another approach to detect concave objects, such as mountains. We apply edge detection and Hough Transform to build polygons to represent mountains. In this paper, we mainly introduce our work on detecting BCPs and representing concave objects.

Recommended citation: Shu Lin, Qinjian Zhang and Haifeng Zhang, "Object Representation in Angry Birds Game," presented at IJCAI Symposium on AI in Angry Birds, Beijing, China, 2013.
Download Paper | Download Bibtex

A Programmer Self-Training System with Programming Skill Evaluation and Personalized Task Recommendation

Published in International Conference on E-Learning, E-Business, Enterprise Information Systems, and E-Government, 2013

Thanks to the widespread use of computers, the basic programming ability is becoming an essential skill for almost all the college students. Although many current online judge systems provide stable and efficient services in automated programs testing, it seems that there is limited help to the learners since few study suggestions is given to them. In this case, learners may feel disoriented during their studies. This paper aims to solve this problem by implementing a system based on one existing platform – Bailian, which will evaluate the learner’s programming skills by combining three scores from different perspectives separately, and then recommend some suitable tasks to them according to the grades of both learners and problems. The system also includes an automatic contest generator to generate contests for self-testing. The quality of these contests is guaranteed since the grades as well as the categories of problems are considered during generating.

Recommended citation: Shu Lin, Qinjian Zhang and Wenxin Li, "A Programmer Self-Training System with Programming Skill Evaluation and Personalized Task Recommendation," presented at the International Conference on E-Learning, E-Business, Enterprise Information Systems, and E-Government, Las Vegas, USA, 2013.
Download Paper | Download Bibtex

Journal Articles

TrafSched: Integrating Bayesian Adaptation with LLMs for Traffic Scheduling Optimization

Published in Electronics, 2026

Railway timetabling requires resolving complex scheduling conflicts arising from shared tracks, station capacity limits, and strict safety intervals. Existing optimization or learning-based approaches often struggle to scale or generalize across diverse operational scenarios. We present TrafSched, a novel hybrid decision framework that combines a curated strategy library, multi-dimensional conflict prioritization, Bayesian strategy adaptation and an optional Large Language Models (LLMs) integration module. TrafSched iteratively detects and resolves conflicts through adaptive strategy selection and backtracking, enabling robust exploration of feasible timetables without costly model retraining. Experiments on real-world-scale datasets involving 50–120 trains show that TrafSched consistently outperforms heuristic and reinforcement learning baselines, achieving up to 85.05% conflict-resolution success in the most challenging cases. These results demonstrate TrafSched’s effectiveness and scalability for modern railway scheduling operations.

Recommended citation: Wentian Fan, Li Xu, Yongcheng Zeng, Siyu Xia, Xinyu Cui, Junyan Shi, Shu Lin, Mengyao Zhang, Yiwei Guo, Xin Zhang and Haifeng Zhang, "TrafSched: Integrating Bayesian Adaptation with LLMs for Traffic Scheduling Optimization," Electronics, Volume 15, February, 2026.
Download Paper | Download Bibtex

Bilevel Scheduling in Downstream Oil Supply Chain: Integrating Reinforcement Learning with Mathematical Programming

Published in Computers & Chemical Engineering, 2026

With the growth of global energy demand, optimizing the oil supply chain has become crucial. This paper proposes a hybrid reinforcement learning (RL) and mathematical programming (MP) scheduling approach to optimize downstream oil supply chain operations, including refinery production scheduling, logistics distribution, and inventory management. This approach decomposes the complex problem into multiple sub-problems using a Rolling-Horizon method (RH), enhancing computational efficiency and flexibility. We conduct a comparative analysis to evaluate two RL training algorithms with RH: Proximal Policy Optimization (PPO) and Soft Actor-Critic (SAC) denoted as SAC-RH and PPO-RH respectively. Experimental results from the simulation-based evaluation demonstrate that the SAC version excels in handling complex dynamic environments and continuous action space problems, significantly reducing the number of early warnings and improving overall optimization results. This study demonstrates the applicability of RL in industrial automation and identifies potential avenues for future research.

Recommended citation: Qipeng Yang, Wentian Fan, Nan Ma, Shu Lin, Jiawen Chang, Zhiqiang Zou, Liang Sun and Haifeng Zhang, "Bilevel Scheduling in Downstream Oil Supply Chain: Integrating Reinforcement Learning with Mathematical Programming," Computers & Chemical Engineering, Volume 204, January, 2026.
Download Paper | Download Bibtex

Theses

Research of Key Technologies in Program Synthesis for Combinatorial Optimization Problems

Published:

This thesis selects combinatorial optimization problems as the research object, investigates the research progress and major problems of both combinatorial optimization problem solving and program synthesis, analyzes the challenges and key points of program synthesis for combinatorial optimization problems, and proposes a new approach for solving problem synthesis, which differs from traditional combinatorial optimization problem solvers. This approach tackles the major challenges and derives several software tools, including (1) a combinatorial optimization problem description language COPDL and the solving program synthesis tool COPDL2C; (2) a natural language-COPDL automatic translation tool NL2COPDL; (3) a problem modeling training platform COPDLOpenJudge (has already been applied to teaching practical courses). The products of this thesis are not only software tools for automatically generating solving programs for combinatorial optimization problems, but also aided educating tools for training programmers.

Recommended citation: Shu Lin, "Research of Key Technologies in Program Synthesis for Combinatorial Optimization Problems," (in Chinese), Doctoral Dissertation, School of Electronics Engineering and Computer Science, Peking University, Beijing, China, 2021
Download Paper | Download Bibtex

A Research on Improving the Evaluation System for Reflecting Users’ Programming Ability and User Behavior Tracking

Published:

Thanks to the widespread use of computers, the basic programming ability is becoming an essential skill for almost all the college students. Although many current online judge systems provide stable and efficient services in automated programs testing, it seems that there is limited help to the learners since few study suggestions is given to them. In this case, learners may feel disoriented during their studies. This paper aims to solve this problem by implementing a system based on one existing platform–Bailian, which will evaluate the learner’s programming skills by combining three scores from different perspectives separately, and then recommend some suitable tasks to them according to the grades of both learners and problems. The system also includes an automatic problem set generator to generate contests for self-testing. The quality of these contests is guaranteed since the grades as well as the categories of problems are considered during generating. The evaluation of actual effectiveness of the improved system is also provided in this paper. By collecting and analyzing the feedbacks from learners as well as the records of learner’s behavior, how much has the system changed in learner’s programming learning can be seen clearly.

Recommended citation: Shu Lin, "A Research on Improving the Evaluation System for Reflecting Users' Programming Ability and User Behavior Tracking," (in Chinese), Bachelor's thesis, School of Electronics Engineering and Computer Science, Peking University, Beijing, China, 2013
Download Paper | Download Bibtex