Study questions whether random projections preserve landscape analysis features needed for high-dimensional optimization

Exploratory Landscape Analysis (ELA) struggles with high-dimensional optimization problems due to sparsity effects and high computational costs

Researchers tested whether dimensionality reduction via Random Gaussian Embeddings (RGEs) maintains the integrity of ELA features

Linear random projections were found to significantly alter geometric and topological properties of optimization landscapes

The findings suggest that features computed in reduced-dimension spaces may not reliably reflect the true characteristics of original high-dimensional problems