From Natural Language to Simulations: Applying GPT-3 Codex to Automate Simulation Modeling of Logistics Systems

  • 2022-08-09 14:59:55
  • Ilya Jackson, Maria Jesus Saenz
  • 0

Abstract

Our work is the first attempt to apply Natural Language Processing toautomate the development of simulation models of systems vitally important forlogistics. We demonstrated that the framework built on top of the fine-tunedGPT-3 Codex, a Transformer-based language model, could produce functionallyvalid simulations of queuing and inventory control systems given the verbaldescription. In conducted experiments, GPT-3 Codex demonstrated convincingexpertise in Python as well as an understanding of the domain-specificvocabulary. As a result, the language model could produce simulations of asingle-product inventory-control system and single-server queuing system giventhe domain-specific context, a detailed description of the process, and a listof variables with the corresponding values. The demonstrated results, alongwith the rapid improvement of language models, open the door for significantsimplification of the workflow behind the simulation model development, whichwill allow experts to focus on the high-level consideration of the problem andholistic thinking.

 

Quick Read (beta)

loading the full paper ...