Tf-ELM-RVFL: A TensorFlow-based library for extreme learning machines and random vector functional link networks with deep and ensemble architectures

We present Tf-ELM-RVFL, a Python library for Extreme Learning Machines (ELM) and Random Vector Functional Link (RVFL) networks using TensorFlow as a computational backend. The library consolidates multiple model variants organized into categories including basic models, kernel-based methods, deep architectures, residual compensation models, and ensemble methods. A distinctive feature is the systematic implementation of RVFL counterparts for each ELM architecture, enabling direct comparisons between models with and without direct input-to-output connections. The library leverages TensorFlow’s efficient linear algebra operations for matrix computations, particularly the Moore-Penrose pseudoinverse calculation, enabling GPU acceleration for large-scale applications while maintaining full compatibility with scikit-learn. All models support classification, regression, and unsupervised learning tasks with competitive performance and the characteristic training efficiency of randomized neural networks.