Enhancing accuracy of deep learning algorithms by training with\n low-discrepancy sequences

2020/05/26 by Siddhartha Mishra, T. Konstantin Rusch, Mishra, Siddhartha +1 · 3 citations
Computer Science · Engineering · Physics and Astronomy · #Advanced Numerical Analysis Techniques #FOS: Computer and information sciences #FOS: Mathematics #FOS: Physical sciences #Fluid Dynamics (physics.flu-dyn) #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Model Reduction and Neural Networks #Numerical Analysis (math.NA) #Numerical Methods and Algorithms

paper · pdf · doi:10.48550/arxiv.2005.12564

openalex publication_date 2020/05/26 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

We propose a deep supervised learning algorithm based on low-discrepancy\nsequences as the training set. By a combination of theoretical arguments and\nextensive numerical experiments we demonstrate that the proposed algorithm\nsignificantly outperforms standard deep learning algorithms that are based on\nrandomly chosen training data, for problems in moderately high dimensions. The\nproposed algorithm provides an efficient method for building inexpensive\nsurrogates for many underlying maps in the context of scientific computing.\n

Cited by

• Neural Low-Discrepancy Sequences
• Optimality of quasi-Monte Carlo methods and suboptimality of the sparse-grid Gauss--Hermite rule in Gaussian Sobolev spaces
• On the optimization of discrepancy measures

Related

• Artificial Neural Networks generated by Low Discrepancy Sequences
• Neural Low-Discrepancy Sequences
• Learning to Generate Synthetic Training Data using Gradient Matching and Implicit Differentiation
• A Multi-level procedure for enhancing accuracy of machine learning algorithms
• Deep Learning Training Procedure Augmentations