Extension of Stochastic Point Location for Multimodal Problems.

Area of Science:

• Optimization Algorithms
• Computational Mathematics
• Machine Learning

Background:

• Stochastic Point Location (SPL) is a fundamental optimization problem where a learning mechanism (LM) finds an optimal point using only directional stochastic information.
• Existing SPL methods are limited to unimodal search spaces, assuming only a single optimal solution.
• Multimodal Optimization Problems (MMOPs) feature multiple optimal solutions, necessitating extensions to existing SPL techniques.

Purpose of the Study:

• To extend Stochastic Point Location (SPL) from unimodal to multimodal optimization problems (MMOPs).
• To propose and validate a novel Parallel Partition Search (PPS) algorithm for addressing MMOPs within the SPL framework.
• To develop a method capable of identifying and locating multiple optimal points in stochastic environments.

Main Methods:

• The Parallel Partition Search (PPS) algorithm is introduced, extending SPL to handle multimodal search spaces.
• PPS extracts features from historical sampling data to identify subintervals likely containing optimal points.
• The search space is adaptively partitioned and sampled in parallel, with subintervals adjusted based on sampling statistics.

Main Results:

• The proposed PPS method successfully extends SPL to address multimodal optimization problems.
• Numerical testing demonstrates the effectiveness and power of the PPS scheme in locating multiple optimal points.
• The ϵ-optimal property of the proposed solution is mathematically proven.

Conclusions:

• The Parallel Partition Search (PPS) offers a robust solution for Stochastic Point Location in multimodal search spaces.
• This work significantly advances the field of optimization by enabling SPL in complex, multi-solution environments.
• The findings pave the way for broader applications of SPL in scenarios with multiple optima.