Interpretable and Personalized Apprenticeship Scheduling: Learning Interpretable Scheduling Policies from Heterogeneous User Demonstrations

Resource scheduling and coordination is an NP-hard optimization requiring an\nefficient allocation of agents to a set of tasks with upper- and lower bound\ntemporal and resource constraints. Due to the large-scale and dynamic nature of\nresource coordination in hospitals and factories, human domain experts manually\nplan and adjust schedules on the fly. To perform this job, domain experts\nleverage heterogeneous strategies and rules-of-thumb honed over years of\napprenticeship. What is critically needed is the ability to extract this domain\nknowledge in a heterogeneous and interpretable apprenticeship learning\nframework to scale beyond the power of a single human expert, a necessity in\nsafety-critical domains. We propose a personalized and interpretable\napprenticeship scheduling algorithm that infers an interpretable representation\nof all human task demonstrators by extracting decision-making criteria via an\ninferred, personalized embedding non-parametric in the number of demonstrator\ntypes. We achieve near-perfect LfD accuracy in synthetic domains and 88.22\\%\naccuracy on a planning domain with real-world, outperforming baselines.\nFinally, our user study showed our methodology produces more interpretable and\neasier-to-use models than neural networks ($p < 0.05$).\n