Technology Showcases

Join conference exhibitors as they discuss innovations and best practices in their fields. Technology Showcase presentations are educational (not commercial) and feature case studies which may include use of exhibitor products and services.

Professional Development Units (PDUs) are available to those who attend these sessions. All full conference registrants are welcome to join during the scheduled time.

Sunday, November 1

11-11:35AM
Room 101

Optimization in Practice for Sustainability with
COPT 9.0

Speakers: Tiancheng Zhang & Helen Yang

Presentation

Rising electricity demand is straining grid capacity just as decarbonization commitments accelerate, and shifting trade policy continues to reshape global supply networks. Against this backdrop, this session presents industrial deployments of the COPT solver in power grid operations, supply chain management, and the circular economy. We share solver best practices, implementation lessons from industry, and a look at where solver technology is heading.

11:40AM-12:15PM
Room 101

Teaching Business Students to Build AI Applications

Speaker: Tingting “Rachel” Chung

Presentation

Rachel Chung will provide an extended presentation of various labs, teaching aids, and challenges in teaching business students to develop AI applications. Learning objectives regarding advancing critical thinking and responsible behavior are covered. This presentation is drawn from her new textbook, “AI for Business: Building Applications using Python” from Prospect Press.

1:15-1:50pm
Room 101

Your Optimization Copilot: Accelerating Better Decisions with Gurobi Intelligence Hub

Speakers: Caroline Weinberg & Nicholas Parham

Presentation

Imagine giving every stakeholder the ability to interact with optimization through natural language and AI-guided assistance. This session showcases Gurobi Intelligence Hub and the intelligent capabilities designed to simplify model development, improve collaboration, and help organizations turn optimization expertise into business impact at scale. See how AI can help bridge the gap between complex models and confident decision-making.

1:55-2:30pm
Room 101

GPU-Accelerated Optimization Takes Flight: FICO® Xpress and Southwest Airlines

Speakers: Rusty Burlingame & Tristan Gally

Presentation

GPU-accelerated optimization has generated enormous promise over the past few years, including the PDHG algorithm for solving large-scale linear programs. While PDHG is just one Linear Programming method that runs well on GPUs, its properties make it amenable for hybrid approaches beyond LP solves. In this session, we’ll show how Southwest Airlines combines FICO Xpress’s GPU-accelerated PDHG algorithm with its commercial MIP solver to tackle one of the largest and most consequential combinatorial optimization problems in the airline industry. We’ll cover the algorithm, its practical implementation, the trade-offs, and then walk through the engineering effort involved, including the challenges encountered along the way and the measurable before-and-after results in performance, memory usage, and business impact. Attendees will leave with a practical roadmap for evaluating and pursuing GPU-accelerated optimization in their own organization.

2:45-3:20pm
Room 101

GAMS in Production: Tools, Trade-offs, and
Real-World Results

Speakers: Atharv Bhosekar, Adam Christensen, & Steve Dirkse

Presentation

This session explores how the GAMS ecosystem supports modern optimization workflows across Python integration, cloud deployment, and
data exchange. Presenters will highlight key tools — GAMSPy for Python-
based modeling, GAMS Connect for moving data between systems, GAMS Engine for scaling computationally demanding models, and MIRO for delivering results through web-based interfaces — illustrated through real-world case studies drawn from production deployments. The session also touches on emerging AI integrations where they complement existing workflows. Attendees will come away with a clear picture of how these tools fit together and the trade-offs involved in solver selection, performance tuning, and long-term model maintainability.

3:25-4pm
Room 101

Introducing the new Nonlinear Modeler and Concurrent Solver in Artelys Knitro 16.0

Speaker: Richard Waltz

Presentation

Artelys Knitro is the leading solver focused on large-scale, nonlinear optimization problems. Knitro offers multiple algorithms for continuous models, as well as tools for handling problems with integer variables. This technology showcase will highlight two new features introduced in the latest Knitro 16.0 release. Firstly, we will highlight the new Nonlinear Modeler (NLM). This tool allows modeling general nonlinear optimization problems in a straightforward way, while also automatically computing and providing all needed derivatives for nonlinear functions. Secondly, we will demonstrate the new Concurrent Solver in Knitro, which runs several Knitro configurations in parallel to quickly and robustly find solutions.

Monday, November 2

8:40-9:15am
Room 101

From Trusting AI to Testing AI: Addressing Common Concerns About LLMs in Text Analysis

Speaker: Normand Péladeau

Presentation

How transparent and reproducible are analyses based on generative AI and large language models (LLMs)? How much confidence should researchers place in results that may vary across runs, models, or providers? What happens when privacy, cost, or institutional restrictions limit access to generative AI? This presentation examines these and other practical and methodological challenges surrounding the use of AI and LLMs in text analysis. It will present approaches developed to address them and illustrate how these ideas have been implemented in QDA Miner and WordStat, while also considering situations in which generative AI may not be the most appropriate solution.

11-11:35am
Room 101

From Compact Combinatorial Models to Strong Dual Bounds: Automatic Decomposition in Practice

Speaker: Fred Gardi

Presentation

Many combinatorial problems in routing, scheduling, and packing are most naturally expressed with set and permutation variables. In a vehicle routing problem, for instance, the sequence of visits of each vehicle is a single list variable. Such compact models are easy to write and well suited to finding good solutions, but they raise a fundamental question: how can strong dual bounds be derived from them? A direct MILP reformulation introduces a quadratic number of binary variables and big-M constraints, which yields weak linear relaxations and scales poorly.

This tutorial-style session explains how extended formulations close this gap. We first review why compact MILP formulations of sequencing problems are weak, and how Dantzig-Wolfe-type extended formulations over routes, schedules, or patterns provide much tighter relaxations. We then discuss how the required structures can be detected automatically in a compact model, and how the resulting formulation can be solved with branch-and-cut-and-price, column elimination, and no-good search, so that the modeler does not need to design the decomposition by hand.

We illustrate these ideas with three case studies drawn from our implementation in the Hexaly solver: capacitated vehicle routing on the CVRPLib X instances (up to 1,000 customers, average gap of 4.5% in 10 minutes), flexible job shop scheduling (average gap of 2.9% in 1 minute), and open-pit mine planning on MineLib (up to 112,000 items and 3 million precedences, with all Ultimate Pit Limit instances solved to optimality in under 10 seconds and gaps below 4.8% in 10 minutes for the Constrained Pit Limit). For each case, we highlight which modeling choices drive bound quality and computational performance. We conclude with practical lessons for practitioners and open research questions in automatic decomposition.

11:40am-12:15pm
Room 101