Tinghan (Joe) Ye

Learning Optimization Proxies for Sequential Contextual Stochastic Programs: An Order Fulfillment Application

Wed, 24 Jun 2026 00:00:00 +0000

Decision-Focused On-Policy Learning for Contextual Linear Optimization with Partial Feedback

Sun, 31 May 2026 00:00:00 +0000

This preprint develops an on-policy decision-focused learning method for sequential contextual linear optimization with partial feedback.

The approach learns a stochastic predict-then-optimize policy and updates it with a hybrid gradient estimator combining score-function and decision-focused plug-in components. Experiments cover top-k selection, shortest path, combinatorial pricing, and energy scheduling.

Democratizing Large-Scale Re-Optimization with LLM-Guided Model Patches

Wed, 27 May 2026 00:00:00 +0000

This preprint introduces an agentic re-optimization framework in which an LLM translates user requests into structured optimization model patches, selects re-optimization tools, and solves updated instances.

The experiments span supply-chain re-optimization and university exam scheduling, showing efficiency gains from primal-based and solver-aware methods while preserving traceable model changes.

HawkesLLM: Semantic Uncertainty Propagation in Agentic Text Simulation

Thu, 21 May 2026 00:00:00 +0000

This preprint studies path-dependent uncertainty in agentic text-simulation systems. HawkesLLM separates temporal influence modeling from text generation: a multivariate Hawkes process selects compact prompt memory, and a language model writes each new event from that memory.

LISTEN to Your Preferences: An LLM Framework for Multi-Objective Selection

Fri, 02 Jan 2026 00:00:00 +0000

Accepted to the 35th International Joint Conference on Artificial Intelligence (IJCAI-ECAI 2026).

We introduce an LLM-powered decision aid that captures qualitative preferences for multi-objective selection.

Paratransit Optimization with Constraint Programming: A Case Study in Savannah, Georgia

Thu, 01 Jan 2026 00:00:00 +0000

Accepted to the 32nd International Conference on Principles and Practice of Constraint Programming (CP 2026).

Contextual Stochastic Optimization for Omnichannel Multicourier Order Fulfillment under Delivery Time Uncertainty

Tue, 14 Oct 2025 00:00:00 +0000

The paper studies a large-scale order fulfillment problem for a leading e-commerce company in the United States. The challenge involves selecting fulfillment centers and shipping carriers with observational data only to efficiently process orders from a vast network of physical stores and warehouses. The company’s current practice relies on heuristic rules that choose the cheapest fulfillment and shipping options for each unit, without considering opportunities for batching items or the reliability of carriers in meeting expected delivery dates.

The paper develops a data-driven contextual stochastic optimization framework that integrates distributional forecasts of delivery time deviations with stochastic and robust order fulfillment optimization models. The framework optimizes the selection of fulfillment centers and carriers, accounting for item consolidation and delivery time uncertainty. Validated on a real-world dataset containing tens of thousands of products, each with hundreds of fulfillment options, the proposed framework significantly enhances the accuracy of meeting customer-expected delivery dates compared to current practices. It provides a flexible balance between reducing fulfillment costs and managing delivery time deviation risks, emphasizing the importance of contextual information and distributional forecasts in order fulfillment.

Recognition: Honorable Mention, M&SOM Practice-based Research Competition, 2025.

Boosting Column Generation with Graph Neural Networks for Joint Rider Trip Planning and Crew Shift Scheduling

Wed, 01 Oct 2025 00:00:00 +0000

Optimizing service schedules is pivotal to reliable, efficient, and inclusive on-demand mobility. This challenge is especially acute for complex paratransit systems that must jointly coordinate rider trip planning and crew scheduling under tight real-time constraints.

This work develops a graph-neural-network-informed column generation approach for the Joint Rider Trip Planning and Crew Shift Scheduling Problem. The key idea is to reduce the number of paths explored in the pricing problem, accelerating the most time-consuming part of column generation while preserving solution quality. The method was evaluated on a real-world dataset from the paratransit system of Chatham County in Georgia and produced substantial improvements over baseline approaches. Additional project context is available on the SAM lab site.

Cornell University Uses Integer Programming to Optimize Final Exam Scheduling

Tue, 30 Sep 2025 00:00:00 +0000

This paper presents an integer programming framework designed to effectively address the complex final exam scheduling challenges encountered at Cornell University. With a high degree of flexibility, the framework is specifically tailored to accommodate diverse constraints, including the front-loading of large courses and the exclusion of specific time slots during the exam period.

By generating multiple scheduling model variants and incorporating heuristic approaches, the framework enables comprehensive comparisons of different schedules. This empowers the University Registrar to make informed decisions, considering trade-offs in schedule comfort measured by different levels of exam conflicts. The results demonstrate significant time and effort savings for the university administration while enhancing student and faculty satisfaction.

Recognition: Finalist, INFORMS Undergraduate Operations Research Prize, 2023.

Conformal Predictive Distributions for Order Fulfillment Time Forecasting

Thu, 22 May 2025 00:00:00 +0000

Accurate estimation of order fulfillment time is critical for e-commerce logistics, yet traditional rule-based approaches often fail to capture the inherent uncertainties in delivery operations. This paper introduces a novel framework for distributional forecasting of order fulfillment time, leveraging Conformal Predictive Systems and Cross Venn-Abers Predictors, model-agnostic techniques that provide rigorous coverage or validity guarantees.

The proposed machine learning methods integrate granular spatiotemporal features, capturing fulfillment location and carrier performance dynamics to enhance predictive accuracy. Additionally, a cost-sensitive decision rule is developed to convert probabilistic forecasts into reliable point predictions. Experimental evaluation on a large-scale industrial dataset demonstrates that the proposed methods generate competitive distributional forecasts, while machine learning-based point predictions significantly outperform the existing rule-based system, achieving up to 14% higher prediction accuracy and up to 75% improvement in identifying late deliveries.

Evaluating Solvers for Linearly Constrained Simulation Optimization

Sun, 15 Dec 2024 00:00:00 +0000

This paper is part of SimOpt, an open-source testbed of simulation-optimization algorithms. I contributed test problems such as a COVID-19 testing frequency problem and an emergency medical volunteer response problem, and I implemented solvers including STRONG and ADAM. This work extends that effort by studying solver performance on problems with linear constraints and by strengthening the benchmark suite available to the community.

Managed Residential Electric Vehicle Charging Minimizes Electricity Bills while Meeting Driver and Community Preferences

Mon, 01 Apr 2024 00:00:00 +0000

Transitions to electric vehicles are expected to increase electricity use in residences, where most drivers tend to recharge. We develop a mathematical programming framework for shifting residential EV charging during low electricity pricing hours to minimize the additional electricity costs that a household incurs from charging its vehicle. The model also accounts for household and community preferences through four secondary objectives: charging as soon as possible on arrival, charging as late as possible before departure, charging for valley filling and peak shaving of residential load, and charging in a shared community hub by using a fast charging station.

We analyze granular residential energy data from a sample of Austin households in 2018 and compare electricity bills under four pricing schemes, including flat rates and time-of-use rates, both with and without a separate meter for EV charging. The findings show that all four secondary charging objectives avoid on-peak charging periods and reduce households’ overall daily electricity costs while surfacing trade-offs in flexibility, charger utilization, and grid load smoothing.

Recognition: 2nd Place, INFORMS Mini Poster Competition, 2021.

A Min-Max Theorem for the Minimum Fleet-Size Problem

Mon, 01 May 2023 00:00:00 +0000

A taxi routing problem can be solved via bipartite matching, where a maximum cardinality matching corresponds to the minimum number of taxis needed to cover all trips. We prove a min-max theorem: the maximum number of pairwise incompatible trips equals the minimum number of taxis needed. We also demonstrate the problem on an NYC taxi dataset and obtain a 35% reduction in the total number of taxis needed.

Part of the work is integrated into a lab for ENGRI 1101: Engineering Applications of OR.