Infrequent Resolving Algorithm for Online Linear Programming
How a handful of well-timed LP solves can match an every-period policy while cutting computation dramatically.
如何通过少量、精准安排的 LP 求解，在大幅降低计算量的同时接近逐期重算策略。
Research briefs & papers研究简报与论文
Guokai Li
Operations, revenue management, and machine learning research, presented as visual briefs with the full papers close at hand.
运营管理、收益管理与机器学习研究。每项研究均以可视化简报呈现，并可直接获取完整论文。
Visual research briefs可视化研究简报
Plain-language explanations of the research question, method, evidence, and managerial implications. Each brief is available in English and Chinese.
以通俗语言解释研究问题、方法、证据与管理启示。每篇简报均提供中英文版本。
From Small to Large: A GCN Approach for Assortment Optimization
A graph model trained on small instances transfers its structural knowledge to assortments one hundred times larger.
在小规模实例上训练的图模型，将结构知识迁移到规模大百倍的商品组合问题。
Self-Improving Neural-Guided Pruning for Mixed Bundle Pricing
A GNN identifies where exact optimization effort matters, then improves itself as product catalogs grow.
图神经网络识别精确优化应聚焦的位置，并随商品目录扩大持续自我改进。
Optimal Emission Regulation under Market Uncertainty
When should a regulator choose an emissions tax, and when is a quantity cap the stronger instrument?
监管者何时应选择排放税，何时数量上限才是更有效的政策工具？
Dealership or Marketplace with Fulfillment Services
A dynamic comparison of two business models, and why their economics converge in sufficiently thick markets.
动态比较两种商业模式，并解释为何在足够厚的市场中两者经济性趋于一致。
When to Push Ads under Markov Customer Dynamics
An optimal timing rule balances hidden customer state, elapsed silence, campaign history, and advertising fatigue.
最优推送时机同时权衡顾客隐状态、沉默时长、活动历史与广告疲劳。
Misspecification Loss in Inventory Systems with Correlated Demand
A closed-form account of what inventory systems lose when correlated demand is modeled as independent.
精确刻画将相关需求误建模为独立需求时，库存系统会付出多大代价。
Primer & latest research入门导读与最新研究
Start with the visual primer, or explore the newest paper through its practitioner brief and full manuscript.
可先阅读可视化入门导读，也可通过实践简报与完整论文了解最新研究。
What Is Online Resource Allocation?
An interactive introduction to capacity, arriving requests, irreversible decisions, and regret.
互动式介绍容量、动态到达、不可逆决策与遗憾值。
Asymptotically Optimal Sequential Testing with Heterogeneous LLMs
Why a high-confidence binary decision needs at most two complementary specialists, selected from cost, latency, and state-specific information.
为何高置信度二元决策最多只需两位互补专家，并依据成本、延迟与分状态信息能力选出他们。