Overview
Kurt Hornik is a statistician and computational scientist (Professor, WU Vienna University of Economics and Business), known for foundational results on neural-network approximation and as a central maintainer of the R project (CRAN, R core).
Key Contributions / Features
- Universal approximation (Hornik-Stinchcombe-White 1989): multilayer feed-forward networks with enough hidden units can approximate any Borel-measurable function to arbitrary accuracy — a foundational result for neural networks.
- Linear autoencoders and PCA (Baldi-Hornik 1989): with Pierre Baldi, characterised the linear-network error landscape as PCA with a unique minimum and only saddles otherwise (Low-Rank Approximation).
- Long-standing contributions to the R ecosystem (CRAN, R core, machine-learning packages).
Related