Operations Research · Artificial Intelligence
Guokai Li
李国凯 🤓
Postdoctoral Researcher, Desautels Faculty of Management, McGill University
I am passionate about developing practical and deployable algorithms for real-world decision-making — working at the intersection of operations research and artificial intelligence, in revenue management and online resource allocation.
I am currently a Postdoctoral Researcher at the Desautels Faculty of Management, McGill University, advised by Prof. Maxime Cohen, Prof. Murray Lei, and Prof. Alys Liang. Before that, I was a Postdoctoral Fellow at Queen's University, advised by Prof. Stefanus Jasin, Prof. Murray Lei, and Prof. Alys Liang.
I received my Ph.D. degree (with Presidential Fellowship) from The Chinese University of Hong Kong, Shenzhen, where I was very fortunate to be supervised by Prof. Zizhuo Wang and Prof. Pin Gao. Prior to that, I obtained my bachelor's degree (ranked 1st) in Industrial Engineering from Xi'an Jiaotong University. I am passionate about developing practical and deployable algorithms for real-world decision-making. My research is broadly related to revenue management and online resource allocation, and I am gradually broadening the scope of my research.
Currently, I am working on topics at the intersection of operations research (OR) and artificial intelligence (AI). In particular, I focus on leveraging AI methodologies to solve complex OR problems (AI4OR) and implementing OR algorithms or insights to improve AI system performance (OR4AI). If you are interested in my research and would like to discuss it further, please feel free to catch me at a conference or email me. 😁
Research interests
News
-
Our paper “Self-Improving Neural-Guided Pruning: A Graph Neural Network Framework for Scalable Mixed Bundle Pricing” is now available on arXiv.
In this paper, we train GNNs to prune the exponentially large bundle space and then use an MILP solver to determine the final bundles and prices. Since exact labels exist only at small scales, we also propose a novel iterative self-improving approach to overcome the scalability bottleneck. For practitioners, see the industry note. Many thanks to Liangyu and Chenghan for their excellent work!
-
Our paper “Infrequent Resolving Algorithm for Online Linear Programming” has been accepted by Mathematics of Operations Research.
In this paper, we develop an efficient algorithm for online resource allocation problems; a non-technical introduction is available here. Notably, the algorithm guarantees constant regret without requiring the non-degeneracy assumption or prior demand information, while solving only a small number of linear programs. This yields an excellent balance between revenue performance and computational efficiency. For practitioners, see the industry note.
-
Our paper “Misspecification Loss in Inventory Systems with Correlated Demand” is now available on SSRN.
In this paper, we analyze the operational costs of ignoring demand correlation and explore how this misspecification loss varies with lead time, correlation coefficient, and backlog penalty. Here is an industry note for practitioners.
-
Our paper “From Small to Large: A Graph Convolutional Network Approach for Solving Assortment Optimization Problems” is now available on arXiv.
In this paper, we illustrate how to use graph neural networks to learn patterns of optimal assortments from small-scale samples, and then utilize the learned patterns to efficiently solve large-scale constrained assortment problems in seconds, which is referred to as the “from small to large” idea. Both theoretical and numerical evidence are provided. Here is an industry note for practitioners. The code is available at GitHub.