Diagnosing Infeasible Optimization Problems Using Large Language Models

In the scientific paper, "Diagnosing Infeasible Optimization Problems Using Large Language Models", authors Hao Chen, Gonzalo E. Constante-Flores, and Can Li introduce a revolutionary tool named OptiChat. This system is designed to assist in understanding and interpreting complex mathematical optimization models, which are widely used in various fields such as economics, engineering, transportation, and health care.

Traditionally, the interpretation of these models, especially when they're infeasible, has been a significant challenge, often requiring extensive background knowledge in optimization. However, OptiChat, with its natural language-based system, simplifies this process, making it accessible even to non-expert users.

OptiChat's primary function is to identify the Irreducible Infeasible Subset (IIS) - a minimal subset of constraints and/or variable bounds that render the entire optimization model infeasible. It then offers suggestions to make the model feasible. The system uses an optimization solver to identify the IIS and is built on the GPT-4 model.

The authors have incorporated several innovative techniques to enhance OptiChat's reliability. Few-shot learning, for instance, provides a set of exemplary demonstrations, each comprising both the input and the anticipated output for the target task. The expert chain-of-thought (CoT) prompting method generates a series of concise sentences, each detailing a step in the reasoning process, leading to the final answer. Key-retrieve is a technique specifically designed for OptiChat, where the keys of the parameter and the constraint names are retrieved from the Pyomo code to improve robustness. Lastly, sentiment prompting is used to understand the user's sentiment.

The experiments carried out by the authors demonstrate that OptiChat significantly assists both expert and non-expert users in improving their understanding of optimization models, enabling them to quickly identify the sources of infeasibility. This innovative tool, thus, holds immense potential in the realm of decision-making problems and optimization models.

Read the whole article here: http://arxiv.org/abs/2308.12923v1