DSpace Repository

Communities in NWU Repository

Select a community to browse its collections.

Now showing 1 - 5 of 20

Recent Submissions

Data Pruning: Redundant, Problematic, and Interdependent Samples
(2026) Freese, Leon; Marthinus, W; Theunissen
The performance of deep learning models is affected by not only data quantity but also data quality. Data pruning is a process by which practitioners can reduce the size of a dataset by only keeping the most important training data points, thereby achieving similar test set performance. We empirically investigate two popular data pruning methods under noisy and noiseless conditions and show that these methods fail in the presence of significant label noise. We highlight that the success of data pruning is distinctly affected by three factors: redundancy in the dataset, the presence of problematic samples, and interdependence between samples. We perform a detailed investigation on commonly used benchmark classification datasets and neural network architectures. We find that our observa- tions are consistent across data distributions and training protocols
Feature extraction for plant growth estimation
(2025) Ngorima, SA; Helberg, ASJ; Davel, MH
Precision agriculture requires the estimation of plant growth stages in real-time. When the plant growth stage is known, the wastage of resources in cultivation, such as nutrients and water, is reduced as only the required resources need to be supplied. Plants at different growth stages, however, have similar morphological features, which can make autonomous growth stage estimation difficult. This paper presents two feature extraction methods for growth stage estimation: one that uses a bank of Gabor filters and morphological operations, and the other that uses pre-trained convolutional neural networks (CNNs) and trans- fer learning. We test these methods on a publicly available plant growth stage dataset ("bccr-segset") for two species, canola and radish, grown and captured under indoor conditions. The two proposed feature extrac- tion methods are compared, using support vector machines and boosted trees as classifiers. We find that both methods are suitable for real-time applications, and that CNN features outperform the hand-crafted fea- tures, both with regard to speed and accuracy. The best system (VGG-19 features, classified with a radial basis function support vector machine) obtained an accuracy of 98.4% for both species, processing an image in 0.08 seconds.
Useful nonrobust features are ubiquitous in biomedical images.
(2026) Mouton Coenraad; Rabe Randle; Koser Niklas; Hansen Christopher
We study whether deep networks for medical imaging learn useful nonrobust features - predictive input patterns that are not human interpretable and highly susceptible to small ad- versarial perturbations - and how these features impact test performance. We show that models trained only on non- robust features achieve well-above-chance accuracy across five MedMNIST classification tasks, confirming their predic- tive value in-distribution. Conversely, adversarially trained models that primarily rely on robust features sacrifice in- distribution accuracy but yield markedly better performance under controlled distribution shifts (MedMNIST-C). Overall, nonrobust features boost standard accuracy yet degrade out- of-distribution performance, revealing a practical robustness- accuracy trade-off in medical imaging classification tasks that should be tailored to the requirements of the deployment setting.
Investigating the relationship between diversity and generalization in over-parameterized deep neural networks
(2026) Van der Spoel; Ruan P; Randle Robe
In ensembles, improved generalization is frequently attributed to \emph{diversity} among members of the ensemble. By viewing a single neural network as an \emph{implicit ensemble}, we perform an exploratory investigation that applies well-known ensemble diversity measures to a neural network in order to study the relationship between diversity and generalization in the over-parameterized regime. Our results show that i) deeper layers of the network generally have higher levels of diversity-particularly for MLPs-and ii) layer-wise accuracy positively correlates with diversity. Additionally, we study the effects of well-known regularizers such as Dropout, DropConnect and batch size, on diversity and generalization. We generally find that increasing the strength of the regularizer increases the diversity in the neural network and this increase in diversity is positively correlated with model accuracy. We show that these results hold for several benchmark datasets (such as Fashion-MNIST and CIFAR-10) and architectures (MLPs and CNNs). Our findings suggest new avenues of research into the generalization ability of deep neural networks.
An investigation of the specific writing difficulties of tertiary level Business Communication students in Botswana
(North-West University, 2026) Plapparambil Nadaraja Pillai, Sivaraj Babu; Butler, HG; Siziba, L
The writing of professional business documents in English appears to be a challenging endeavour for tertiary students all over the world. Since English writing skills are seen as one of the priority skills required when graduates seek employment, the writing difficulties often experienced by such graduates could affect their chances of securing employment negatively. In this context, the current study investigated the writing ability of tertiary students at a specific tertiary college in Botswana (hereafter referred to as the ABCD) with specific reference to their writing of appropriate business documents. All these students have to complete a Business Communication course in their first year (An Introduction to Business Communication [IBC]), where the writing of business documents is addressed as part of the course. A comprehensive literature review was conducted as a first step in order to determine the general academic writing difficulties of tertiary students in Africa, as well as the specific writing difficulties of such students in writing professional documents related to a business context. The literature survey further wished to establish whether writing difficulties were limited to tertiary students in Africa or if such difficulties were experienced internationally. Secondly, the researcher wanted to determine empirically whether the IBC students at the ABCD experienced writing difficulties with respect to the writing of business (and general academic) documents both before and after they completed the first year IBC course. Because the Business Communication course needed to prepare the students adequately for the writing of appropriate business documents in the world of work, the study further determined whether students who were busy with their internships (during their 4th year) at various prospective employers in Botswana met the expectations of such employers concerning the writing of professional business documents. Various additional data sets were also collected in order to verify the appropriateness of the IBC course at the ABCD. These included an analysis of the learning outcomes of the IBC course compared with the typical outcomes of other Business Communication courses, an analysis of the final examination papers of the IBC course, relevant information collected from the lecturers who presented the course, as well as observations on how the writing aspect of the course was taught by these lecturers. The results of all these investigations were then used to propose a possible adjustment to the current business writing intervention at the ABCD to address the specific writing difficulties of the IBC students more effectively, as well as to respond more effectively to the expectations of employers.