Current browse context:
stat.ML
Change to browse by:
References & Citations
Statistics > Machine Learning
Title: The Universal Approximation Property
(Submitted on 8 Oct 2019 (v1), last revised 28 Nov 2020 (this version, v4))
Abstract: The universal approximation property of various machine learning models is currently only understood on a case-by-case basis, limiting the rapid development of new theoretically justified neural network architectures and blurring our understanding of our current models' potential. This paper works towards overcoming these challenges by presenting a characterization, a representation, a construction method, and an existence result, each of which applies to any universal approximator on most function spaces of practical interest. Our characterization result is used to describe which activation functions allow the feed-forward architecture to maintain its universal approximation capabilities when multiple constraints are imposed on its final layers and its remaining layers are only sparsely connected. These include a rescaled and shifted Leaky ReLU activation function but not the ReLU activation function. Our construction and representation result is used to exhibit a simple modification of the feed-forward architecture, which can approximate any continuous function with non-pathological growth, uniformly on the entire Euclidean input space. This improves the known capabilities of the feed-forward architecture.
Submission history
From: Anastasis Kratsios [view email][v1] Tue, 8 Oct 2019 11:30:33 GMT (142kb)
[v2] Thu, 20 Feb 2020 09:20:21 GMT (41kb)
[v3] Wed, 5 Aug 2020 20:52:07 GMT (41kb)
[v4] Sat, 28 Nov 2020 11:17:47 GMT (42kb)
Link back to: arXiv, form interface, contact.