Enhancing accuracy of deep learning algorithms by training with low-discrepancy sequences
This provides an efficient method for building inexpensive surrogates in scientific computing, addressing a domain-specific bottleneck.
The paper tackles the problem of improving deep learning accuracy by using low-discrepancy sequences for training sets, demonstrating significant performance gains over random training data in moderately high-dimensional problems through theoretical and experimental evidence.
We propose a deep supervised learning algorithm based on low-discrepancy sequences as the training set. By a combination of theoretical arguments and extensive numerical experiments we demonstrate that the proposed algorithm significantly outperforms standard deep learning algorithms that are based on randomly chosen training data, for problems in moderately high dimensions. The proposed algorithm provides an efficient method for building inexpensive surrogates for many underlying maps in the context of scientific computing.