My PhD research applies hierarchical reinforcement learning to real-time policy optimization — routing and logistics problems where demand is stochastic, the fleet is heterogeneous, and a decision has to be made now rather than optimally.
The recurring question across this work is what to do when the exact method is too slow to be useful. Classical operations research gives provably optimal answers to problems that have usually stopped being the problem by the time the solver finishes. Learned policies decide in milliseconds but come with no guarantees. Most of my work lives in the space between: using learning to guide search, or to produce a policy that respects the structure the exact methods made explicit.
PhD candidate, Geomatics Engineering, University of Calgary — defending 9 October 2026, supervised by Xin Wang. GPA 4.0/4.0.
Publications
Journal articles
HELP-RL: Real-Time Policy Optimization with Hierarchical Decoder for Heterogeneous and Stochastic Disaster Response Logistics (2026) Aghazadeh, H., and Wang, X. — Computers & Operations Research
Dray-Q: Demand-Dependent Trailer Repositioning Using Deep Reinforcement Learning (2023) Aghazadeh, H., Wang, Y., Sun, B., and Wang, X. — Transportation Research Part C: Emerging Technologies
Conference and workshop papers
Multi-Task Vehicle Routing with Hierarchical Option Learning (2026) Aghazadeh, H., and Wang, X. — 34th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems
GAGE-Q: Reinforced Genetic Algorithm Using Spatial Neighborhood Graph Embedding for Green Intermodal Transportation (2025) Aghazadeh, H., Safarzadeh, R., and Wang, X. — Advances in Cartography and GIScience of the ICA
Hierarchical Reinforcement Learning for Real-Time Policy Optimization in Complex Logistics Networks (2025) Aghazadeh, H., and Wang, X. — ACM KDD 2025, Supply Chain workshop, Toronto
Reinforcement Learning for Intermodal Transportation Planning with Time Windows and Limited Cargo Capacity (2023) Aghazadeh, H., and Wang, X. — ACM SIGSPATIAL International Workshop on Computational Transportation
Book
Applied Reinforcement Learning: Business optimization and LLM fine-tuning (2026) Aghazadeh, H. — Manning Publications. Details · Publisher
For citations and the current list, see Google Scholar.
Applied research
Some of the research happens at work rather than in a journal. At Enverus I prototype large-scale mixed-integer linear programming for power-flow and security-constrained unit commitment, and investigate reinforcement-learning-guided cutting-plane selection — using a learned policy to decide which cuts are worth adding, so branch-and-bound converges fast enough to be useful at market speed.
It’s the same question as the routing work in a different domain: exact methods know the structure, learned methods are fast, and the interesting result is usually a combination rather than a winner.
Service
Peer reviewer for machine learning, transportation, optimization and spatial-computing journals. Vice President and Student Representative of the Graduate Geomatics Group at the University of Calgary (2023–2024).