DMGSO: A Deterministic Memory-Guided Sensing Framework for Derivative-Free Optimization

Derivative-free optimization (DFO) is widely used for black-box problems where analytical derivatives are unavailable or expensive to obtain. While stochastic methods often provide strong exploration, their performance may vary due to randomized search operators. Deterministic approaches offer reproducible behavior but frequently lack adaptive exploration in complex landscapes. To address these limitations, this study introduces Deterministic Memory-Guided Sensing Optimization (DMGSO), a deterministic optimization framework based on directional sensing, adaptive memory-guided search, and deterministic relocation. The framework formulates optimization as a knowledge-guided decision process that integrates local and long-range observations with accumulated search experience. Unlike population-based stochastic optimizers, DMGSO generates reproducible search trajectories through deterministic evidence accumulation and adaptive decision regulation. The framework was evaluated on the CEC2014 and COCO/BBOB benchmark suites using ablation studies and statistical analyses. Results show that the complete framework consistently outperformed reduced variants and achieved competitive performance on complex hybrid and composition functions. Although scalability decreased in high-dimensional problems due to sensing sparsity, the findings demonstrate that deterministic sensing and memory-guided evidence accumulation provide an effective and reproducible alternative for derivative-free optimization.