no code implementations • 29 Aug 2023 • Coenraad Mouton, Marthinus W. Theunissen, Marelie H. Davel
While margins have been shown to be correlated with the generalization ability of a model when measured at its hidden representations (hidden margins), no such link between large margins and generalization has been established for input margins.
no code implementations • 14 Feb 2023 • Marthinus W. Theunissen, Coenraad Mouton, Marelie H. Davel
A global estimate of margin size is usually used in the literature.
no code implementations • 3 Oct 2022 • Walter Heymans, Marelie H. Davel, Charl van Heerden
The GAN is used to enhance the features of mismatched data prior to decoding, or can optionally be used to fine-tune the acoustic model.
Automatic Speech Recognition Automatic Speech Recognition (ASR) +2
no code implementations • 15 Feb 2022 • Walter Heymans, Marelie H. Davel, Charl van Heerden
Mismatched data is a challenging problem for automatic speech recognition (ASR) systems.
Automatic Speech Recognition Automatic Speech Recognition (ASR) +2
no code implementations • 1 Feb 2022 • Coenraad Mouton, Marelie H. Davel
While deep neural networks (DNNs) have become a standard architecture for many machine learning tasks, their internal decision-making process and general interpretability is still poorly understood.
no code implementations • 13 Apr 2021 • Johannes C. Myburgh, Coenraad Mouton, Marelie H. Davel
Our main focus is to investigate the effect of different architectural components of a standard CNN on that network's sensitivity to translation.
no code implementations • 18 Mar 2021 • Coenraad Mouton, Johannes C. Myburgh, Marelie H. Davel
We also observe that this characteristic is dataset-specific and dictates the relationship between pooling kernel size and stride required for translation invariance.
no code implementations • 14 Mar 2021 • Arthur E. W. Venter, Marthinus W. Theunissen, Marelie H. Davel
When training neural networks as classifiers, it is common to observe an increase in average test loss while still maintaining or improving the overall classification accuracy on the same dataset.
no code implementations • 17 Jan 2020 • Marelie H. Davel, Marthinus W. Theunissen, Arnold M. Pretorius, Etienne Barnard
A robust theoretical framework that can describe and predict the generalization ability of deep neural networks (DNNs) in general circumstances remains elusive.