Hot off the Press: Bayesian Inverse Transfer in Evolutionary Multiobjective Optimization
Abstract
This Hot Off the Press paper provides a brief summary of our recent work "Bayesian Inverse Transfer in Evolutionary Multiobjective Optimization". In this paper, we propose a novel concept termed inverse transfer for multiobjective optimization. Unlike conventional transfer approaches, inverse transfer leverages the shared objective functions commonly found across diverse application domains, even when the corresponding decision spaces are misaligned. This capability enables the effective integration of knowledge from heterogeneous source tasks. Furthermore, a key byproduct of inverse transfer is the construction of high-precision inverse models, which can not only enhance diversity by querying previously unexplored regions of the solution space but also facilitate the on-demand generation of customized solutions tailored to user preferences. Building upon this idea, we introduce the first Inverse Transfer Evolutionary Multiobjective Optimizer (invTrEMO). The source code of the invTrEMO is made available at https://github.com/LiuJ-2023/invTrEMO.
Authors: Jiao Liu, Abhishek Gupta, Yew-Soon Ong
Published in: Genetic and Evolutionary Computation Conference Companion (GECCO Companion) (2025)