Academic Presentations & Publications
Linyan Li, Yuqiang Ning (Co-Author), “From Links to Lanes: A Lane-Level Traffic Assignment Framework.”
Presentation at the 2025 INFORMS Annual Meeting, Atlanta, Georgia, USA, October 2025.
• Presented a lane-level traffic assignment framework addressing the link-based model's aggregation bias.
• Proposed a multi-layer lane-based network structure capturing lane-changing and intersection control interactions.
• Integrated nonlinear BPR functions with KKT-based equilibrium optimization to model realistic congestion.
• Conducted microscopic traffic simulations using Simulation of Urban Mobility on 3 lanes, simulating 1200 vehicles.
• Validated the framework on real-world urban (Sioux Falls) networks, improved accuracy in congestion prediction by 22% and providing practical implications for lane management and transportation policy.
Linyan Li, Lei Liu (Advisor), “A Reinforcement Learning Approach to Dynamic Pricing of Perishable Products.”
Presentation at the 4th Nottingham University Business School Tri-Campus Conference, Nottingham, UK, April 2024.
• Developed RL-based dynamic pricing strategies for perishable products using Deep Q-Learning and Actor-Critic, integrated ML demand forecasting with LSTM and CPLEX optimization to model stochastic demand and shelf life.
• Built an adaptive, real-time pricing AI agent targeting the revenue–waste trade-off, containerized using Docker and served using a webapp (HTML/React) through FastAPI with Redis caching, achieving \( < 100 \) ms inference latency.
• Validated in simulated retail scenarios for 14 days, demonstrating increase of 10% gross margin and 14% inventory turnover rate and reduce 21% daily waste while keeping stockout rate to be less than 5%.