Machine Learning-Based Prediction of New Pareto-Optimal Solutions From Pseudo-Weights
- 28 September 2023
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Evolutionary Computation
- Vol. 28 (5) , 1351-1365
Abstract
Owing to the stochasticity of Evolutionary Multi-objective Optimization (EMO) Algorithms and an application with a limited budget of solution evaluations, a perfectly converged and uniformly distributed Pareto-optimal (PO) front cannot be always guaranteed. Thus, a subsequent decision-making step or a curiosity on the part of the optimization researcher may demand solutions at regions not well-represented by the obtained PO front. In this study, we propose to train Machine Learning (ML) models to capture the mapping between unique identifiers of PO solutions – pseudo-weight vectors, computed from the existing PO front data, and their corresponding decision variable vectors. These learned models can then be used to predict PO decision variables for any new desired pseudo-weight vector. We evaluate the proposed approach with two different ML methods on a variety of multi-and many-objective test and real-world problems. This procedure can also be incorporated into an EMO algorithm to find a better converged set of PO solutions, attempt to fill apparent gaps, and find more non-dominated solutions at preferred regions of the PO front, facilitating a number of key advances for multi-objective optimization and decision-making tasks.Keywords
Funding Information
- Koenig Endowed Chair position at ECE Department of Michigan State University
This publication has 40 references indexed in Scilit:
- Learning to Predict Pareto-Optimal Solutions from Pseudo-weights2023
- Machine learning-based framework to cover optimal Pareto-front in many-objective optimizationComplex & Intelligent Systems, 2022
- A benchmark test suite for evolutionary many-objective optimizationComplex & Intelligent Systems, 2017
- Adaptive Reference Vector Generation for Inverse Model Based Evolutionary Multiobjective Optimization with Degenerate and Disconnected Pareto Fronts2015
- Large-scale MV efficient frontier computation via a procedure of parametric quadratic programmingEuropean Journal of Operational Research, 2010
- Multiobjective optimization for crash safety design of vehicles using stepwise regression modelStructural and Multidisciplinary Optimization, 2007
- Scalable Test Problems for Evolutionary Multiobjective Optimization2005
- A Scalable Multi-objective Test Problem Toolkit2005
- Nonlinear Multiobjective Optimization2001
- A new method to solve generalized multicriteria optimization problems using the simple genetic algorithmStructural and Multidisciplinary Optimization, 1995