Institution profile

University of the Witwatersrand

Academic institutionafrica · za
Official website
Research library83linked papers
Opportunities0open roles
Selected work

Representative Papers

Prevalence of food and housing insecurity among direct support professionals in New York.

Dec 01, 2024Disability and Health Journal

This study presents the first systematic assessment of the prevalence of food and housing insecurity among Direct Support Professionals (DSPs) in New York State and its association with social determinants of health. Drawing on a statewide cross-sectional survey of 2,766 DSPs and employing chi-square tests and logistic regression analyses, the research reveals that 62.6% of respondents experience food and/or housing insecurity, with disproportionately high rates among individuals with disabilities, people of color, and those with low incomes. The findings conceptualize basic living insecurity as a critical occupational hazard and a fundamental threat to the stability of the care system. The study underscores the urgent need for policy interventions to improve DSP compensation and ensure access to basic living necessities, providing empirical evidence to inform efforts aimed at enhancing both working conditions for DSPs and the quality of care they deliver.

2 citationsRead paper

Make Haste Slowly: A Theory of Emergent Structured Mixed Selectivity in Feature Learning ReLU Networks

Mar 08, 2025

Existing theoretical frameworks inadequately explain how structured representations emerge in finite-width ReLU neural networks, particularly for multi-context tasks. Method: The authors establish, for the first time, dynamical equivalence between ReLU networks and gated deep linear networks under multi-context settings, leveraging nonlinear dynamical analysis, inductive bias modeling, and carefully designed controllable tasks. Contribution/Results: They reveal that node reuse and learning-rate dynamics jointly drive the emergence of hidden-layer representations; moreover, increasing network depth or context count systematically enhances *mixed selectivity*—a non-decomposable yet highly structured latent representation. This work provides the first interpretable and predictive theory of feature learning for finite-width ReLU networks, removing restrictive assumptions such as infinite width, single-layer architectures, or unstructured data.

1 citationsRead paper

A Theory of Initialisation's Impact on Specialisation

Mar 04, 2025

This work challenges the necessity of neuron specialization for mitigating catastrophic forgetting in continual learning, revealing that specialization is primarily governed by network initialization rather than intrinsic task properties. Method: Through theoretical analysis and empirical validation, the authors demonstrate that weight imbalance and high weight entropy actively induce localized representations; they provide the first theoretical proof that specialization is not inherent but contingent. They further derive a quantitative relationship between specialization degree and initialization parameters, and reproduce the monotonic relationship between task similarity and forgetting rate even in non-specialized networks. Contribution/Results: Specialized initialization significantly enhances Elastic Weight Consolidation (EWC) performance, an effect attributable to initialization-induced prior shaping of representation structure. These findings establish a novel theoretical foundation for regularization design in continual learning and yield principled guidelines for initialization strategy selection.

1 citationsRead paper

Investigating the Impact of Language-Adaptive Fine-Tuning on Sentiment Analysis in Hausa Language Using AfriBERTa

Jan 19, 2025

To address the suboptimal adaptation of pretrained models for sentiment analysis in low-resource languages like Hausa, this paper proposes Language-Adaptive Fine-Tuning (LAFT): first performing unsupervised domain- and language-specific adaptation of AfriBERTa on unlabeled Hausa corpora, followed by supervised fine-tuning on the NaijaSenti dataset. This work represents the first application of LAFT to Hausa sentiment analysis, explicitly accounting for linguistic characteristics of informal social media text. Experiments demonstrate consistent, modest performance gains from LAFT; AfriBERTa substantially outperforms multilingual baselines without language-specific adaptation, underscoring the critical role of language-targeted pretraining in low-resource settings. All data and code are publicly released to advance NLP research for African languages.

1 citationsRead paper
Recent publications

Latest Papers

Programmable Cellular Automata

Sep 05, 2026

研究通过将元胞自动机表示为Python代码并模块化，解决了创建有效局部规则困难的问题，并探索了全局函数的作用以减少解决问题的迭代次数。

0 citationsRead paper