Zero-Inflated Gaussian Distributions Enable Parameter-Space Sparsity in Estimation-of-Distribution Algorithms

The article presents a novel approach using multivariate zero-inflated Gaussian (ZIG) distributions in estimation-of-distribution algorithms (EDAs) to effectively handle sparse parameter spaces in black-box optimization. This method allows for joint optimization of sparsity patterns and active values without relying on hand-crafted operators, demonstrating superior performance on the Lunar Lander benchmark by achieving faster convergence and higher returns compared to traditional dense Gaussian EDAs and existing sparse algorithms. This advancement is significant for practitioners as it enhances the efficiency of optimization in scenarios where many parameters are inactive, thus reducing computational resources and improving solution quality.