Speaker: Nate Thach, PhD student, School of Computing at UNL
When: Friday, October 17th, 10:50am-11:50am
Where: KH A510 (Lincoln), PKI 160 (Omaha)
Zoom: https://unl.zoom.us/j/94281831713
Title: Large Language Models for Combinatorial Optimization Problems in Transportation
Abstract: Solving NP-hard combinatorial optimization problems (COPs), such as traveling salesman problems (TSPs) and capacitated vehicle routing problems (CVRPs), traditionally involves handcrafting heuristics or formulating linear programs, which require substantial domain knowledge and implementation efforts from human experts. Recently, advances in large language models (LLMs) such as GPT and Gemini have enabled new approaches for algorithmic design. In this talk, I will present my recent and ongoing work on leveraging LLMs for several practical COPs in transportation. In the first part, I will talk about our newest LLM-based framework for automating the design of novel heuristics for TSPs and CVRPs. In the second part, I will demonstrate how LLMs can be employed for solving a challenging COP in the wild, namely the generalized facility accessibility improvement problem, which has no effective heuristics in the literature.