我要吃瓜

Citation

Catalano GAPI, Brownlee A, Cairns D, McCall J, Fyvie M & Ainslie R (2024) Explaining a Staff Rostering Problem using Partial Solutions. In: TBC. Lecture Notes in Artificial Intelligence. AI-2024 Forty-fourth SGAI International Conference on Artificial Intelligence, Cambridge, 17.12.2024-19.12.2024. Cham, Switzerland: Springer.

Abstract
There are many critical optimisation tasks that metaheuris-tic approaches have been shown to be able to solve effectively. Despite promising results, users might not trust these algorithms due to their intrinsic lack of interpretability. This paper demonstrates the use of ex-plainability to resolve this issue by producing human-interpretable insights that focus on simplicity, fitness and linkage. Our explainability approach revolves around the concept of Partial Solutions , which assist in breaking up the solutions of optimisation problems into smaller components. We first expand upon our previous research proposing the technique, and then provide a use case on the Staff Ros-tering task: a large and otherwise uninterpretable optimisation problem with ethical implications due to its direct impact on humans. The explanations consist in rota assignments for interacting groups of workers, along with the reasons why they are interacting. Lastly, some experiments are used to ascertain that the algorithms work as intended and for hyperparameter tuning. The results suggest that our methodology is capable of presenting in-sightful information for the Staff Rostering problem, by producing both local explanations of solutions and global explanations of the problem definition.

Keywords
Explainability; XAI; Job Scheduling; Metaheuristics

Notes
Output Status: Forthcoming