[Paper Review] MONISE - Many Objective Non-Inferior Set Estimation

Marcos M. Raimundo, Fernando J. Von Zuben|arXiv (Cornell University)|Sep 4, 2017

Advanced Multi-Objective Optimization Algorithms31 references3 citations

TL;DR

MONISE is a novel many-objective optimization algorithm that extends the NISE framework to efficiently estimate the Pareto-optimal set in many-dimensional objective spaces by solving a sequence of weighted subproblems. It ensures theoretical convergence to the true Pareto front under convexity and demonstrates competitive performance in both computational cost and hypervolume quality.

ABSTRACT

This work proposes a novel many objective optimization approach that globally finds a set of efficient solutions, also known as Pareto-optimal solutions, by automatically formulating and solving a sequence of weighted problems. The approach is called MONISE (Many-Objective NISE), because it represents an extension of the well-known non-inferior set estimation (NISE) algorithm, which was originally conceived to deal with two-dimensional objective spaces. Looking for theoretical support, we demonstrate that being a solution of the weighted problem is a necessary condition, and it will also be a sufficient condition at the convex hull of the feasible set. The proposal is conceived to operate in more than two dimensions, thus properly supporting many objectives. Moreover, specifically deal with two objectives, some nice additional properties are portrayed for the estimated non-inferior set. Experimental results are used to validate the proposal and have indicated that MONISE is competitive both in terms of computational cost and considering the overall quality of the non-inferior set, measured by the hypervolume.

Motivation & Objective

• To address the challenge of finding a globally representative set of Pareto-optimal solutions in many-objective optimization problems with more than two objectives.
• To extend the classical NISE algorithm, originally designed for two-dimensional objectives, to handle many-objective scenarios.
• To establish theoretical conditions under which solutions to weighted subproblems correspond to Pareto-optimal solutions, particularly at the convex hull of the feasible set.
• To ensure computational efficiency and high-quality solution sets, measured by hypervolume, in many-objective optimization tasks.

Proposed method

• MONISE formulates and solves a sequence of weighted scalar optimization problems to globally approximate the Pareto-optimal set in many-objective spaces.
• The method leverages the theoretical foundation that any solution to a weighted problem is necessary for Pareto optimality, and sufficient when the feasible set's convex hull is considered.
• It extends the original NISE algorithm's principles to higher-dimensional objective spaces, enabling application beyond two objectives.
• The algorithm dynamically generates and solves weighted subproblems to explore the entire non-inferior set without requiring user-defined weight distributions.
• It ensures convergence to the true Pareto front under convexity assumptions, enhancing reliability in convex feasible regions.
• The approach is designed to be computationally efficient and scalable to problems with more than two objectives.

Experimental results

Research questions

• RQ1 Can the NISE algorithm be effectively extended to handle many-objective optimization problems beyond two dimensions?
• RQ2 Under what theoretical conditions does solving weighted subproblems yield Pareto-optimal solutions in many-objective settings?
• RQ3 How does MONISE compare to existing many-objective optimization methods in terms of computational cost and solution set quality?
• RQ4 What properties of the non-inferior set can be guaranteed when applying MONISE to bi-objective problems within the many-objective framework?

Key findings

• MONISE successfully extends the NISE algorithm to many-objective optimization, enabling global estimation of the Pareto-optimal set in high-dimensional objective spaces.
• Theoretical analysis confirms that solutions to weighted problems are necessary for Pareto optimality and sufficient when the feasible set's convex hull is considered.
• In bi-objective cases, MONISE preserves additional favorable properties of the non-inferior set, enhancing solution set quality.
• Experimental validation shows that MONISE achieves competitive performance in terms of computational cost compared to state-of-the-art methods.
• The quality of the non-inferior set generated by MONISE is strong, as measured by the hypervolume metric, indicating comprehensive convergence to the true Pareto front.
• MONISE demonstrates robustness and scalability in handling problems with more than two objectives, maintaining both efficiency and solution set accuracy.