Zero-Inflated Gaussian Distributions Enable Parameter-Space Sparsity in Estimation-of-Distribution Algorithms
Researchers have developed a novel approach to enhance estimation-of-distribution algorithms (EDAs) for optimization problems with sparse parameter spaces. By employing multivariate zero-inflated Gaussian (ZIG) distributions, these algorithms can now effectively handle scenarios where many solution coefficients are zero. This method jointly optimizes sparsity patterns and active parameter values without hierarchical assumptions, leading to improved convergence and performance on benchmarks like Lunar Lander compared to existing sparse optimization techniques. AI
IMPACT Introduces a new method for optimizing sparse parameter spaces in machine learning algorithms.