Current browse context:
math.DS
Change to browse by:
References & Citations
Computer Science > Machine Learning
Title: A Letter on Convergence of In-Parameter-Linear Nonlinear Neural Architectures with Gradient Learnings
(Submitted on 25 Nov 2021)
Abstract: This letter summarizes and proves the concept of bounded-input bounded-state (BIBS) stability for weight convergence of a broad family of in-parameter-linear nonlinear neural architectures as it generally applies to a broad family of incremental gradient learning algorithms. A practical BIBS convergence condition results from the derived proofs for every individual learning point or batches for real-time applications.
Link back to: arXiv, form interface, contact.