A Syntactic Adaptive Problem Solver Learning Landscape Structures for Scheduling in Clinical Laboratory

TL;DR

This paper develops a syntactic adaptive problem solver for clinical laboratory scheduling by analyzing landscape structures, aiming to improve solution efficiency and reduce algorithm tuning through landscape-aware strategies.

Contribution

It introduces a novel adaptive solver leveraging landscape structure analysis and meta-learning to enhance scheduling in complex, heterogeneous environments.

Findings

Smaller perturbations lead to smoother landscapes for medium-sized instances.

Larger perturbations increase landscape ruggedness, complicating search for large instances.

Meta-Lamarckian learning improves reliability of neighborhood management.

Abstract

This paper attempts to derive a mathematical formulation for real-practice clinical laboratory schedul-ing, and to present a syntactic adaptive problem solver by leveraging landscape structures. After formulating scheduling of medical tests as a distributed scheduling problem in heterogeneous, flexible job shop environment, we establish a mixed integer programming model to minimize mean test turn-around time. Preliminary landscape analysis sustains that these clinics-orientated scheduling instances are difficult to solve. The search difficulty motivates the search for an adaptive problem solver to reduce repetitive algorithm-tuning work, but with a guaranteed convergence. Yet, under a search strategy, relatedness from exploitation competence to landscape topology is not transparent. Under strategies that impose different-magnitude perturbations, we investigate changes in landscape struc-ture and find that disturbance amplitude, local-global optima connectivity, landscape's ruggedness and plateau size fairly predict strategies' efficacy. Medium-size instances of 100 tasks are easier un-der smaller-perturbation strategies that lead to smoother landscapes with smaller plateaus. For large-size instances of 200-500 tasks, existing strategies at hand, having either larger or smaller perturba-tions, face more rugged landscapes with larger plateaus that impede search. Our hypothesis that me-dium perturbations may generate smoother landscapes with smaller plateaus drives our design of this new strategy and its verification by experiments. Composite neighborhoods managed by meta-Lamarckian learning show beyond average performance, implying reliability when prior knowledge of landscape is unknown.