Privacy-Preserving Optimization in Time-Critical Settings

Description

Solving some real-world optimization problems requires optimization without leaking sensitive information, i.e., privacy-preserving optimization. In time-critical settings, solutions for optimization problems must be found within the time constraints imposed by the application context. However, the use of methods for privacy-preserving computation causes a considerable runtime overhead. Therefore, privacy-preserving implementations of exact algorithms may not find an optimal solution within the time available for optimization. In contrast to exact algorithms, evolutionary algorithms allow the separation of the search for solutions and the evaluation of the solutions. Thus, only the evaluation needs to be implemented with methods for privacy-preserving computation, which reduces their impact on runtime. In addition, evolutionary algorithms provide valid solutions even if they are terminated at any point in time. Motivated by the Horizon Europe industrial research project HARMONIC, which deals with flight prioritization in Air Traffic Flow Management, we investigate the combination of evolutionary algorithms for solution search with methods for privacy-preserving computation to evaluate the solutions. We also explore multi-objective optimization to balance the interests of different stakeholders and ensure long-term equity between competing stakeholders. In this presentation, we present experimental results of using genetic algorithms for privacy-preserving optimization of multi-objective assignment problems in time-critical settings.

Period	12 Jul 2024
Event title	unbekannt/unknown
Event type	Other
Location	AustraliaShow on map