South African Tuberculosis Vaccine Initiative
UCT Computer Science Research Document ArchiveNot a member yet
1270 research outputs found
Sort by
Scalable Evolutionary Hierarchical Reinforcement Learning
Full text linkThis paper investigates a novel method combining Scalable Evolution
Strategies (S-ES) and Hierarchical Reinforcement Learning (HRL).
S-ES, named for its excellent scalability, was popularised with demonstrated
performance comparable to state-of-the-art policy gradient
methods. However, S-ES has not been tested in conjunction with
HRL methods, which empower temporal abstraction thus allowing
agents to tackle more challenging problems. We introduce a novel
method merging S-ES and HRL, which creates a highly scalable
and efficient (compute time) algorithm. We demonstrate that the
proposed method benefits from S-ES’s scalability and indifference
to delayed rewards. This results in our main contribution: significantly
higher learning speed and competitive performance compared
to gradient-based HRL methods, across a range of tasks
Measuring Cloud Latency in Africa
Full text linkInternet and Public Cloud adoption has been growing all over the world, and major Cloud Providers have a growing presence in Africa. However, little research has been conducted on performance of the Cloud in Africa, particularly from end users' point of view. This study measures network latencies experienced in Africa when accessing public Cloud infrastructure, and compares this with what is achievable in Europe. We use the RIPE Atlas platform to run latency and traceroute measurements to CDN endpoints and servers in datacenters in Africa and Europe. Reverse measurements are also conducted from the Virtual servers to non-RIPE endpoints in both Africa and Europe. Our results show that clients in Africa mostly use CDN nodes located outside the continent, resulting in higher latencies. We also observed some clients making use of circuitous routes to cloud destinations within Africa. In Europe, we found that a majority of CDN endpoints used were local, which resulted in lower latencies. We also find that using CDN nodes in Africa provides up to 87 lower latencies than accessing the Data Centres directly. In Europe, CDN access provided up to 142 lower latencies than accessing Data Centers. Results of this study should motivate cloud providers to continue increasing their CDN presence in Africa and to work with local ISPs to optimise routing and content delivery from their cloud infrastructure
Comparative molecular modelling of capsular polysaccharide conformations in Streptococcus suis serotypes 1, 2, 1/2 and 14 identifies common epitopes for antibody binding.
Full text linkStreptococcus suis is an encapsulated, commensal, potentially pathogenic bacterium that infects swine globally and causes sporadic life-threatening zoonotic septicemia and meningitis infections in humans. The capsular polysaccharide is a primary virulence factor for S. suis. As S. suis serotype 2 is the most prevalent serotype globally, the serotype 2 CPS is the primary target of current efforts to develop an effective glycoconjugate veterinary vaccine against S. suis. Possible cross-protection with related serotypes would broaden the coverage of a vaccine. The CPS in serotypes 2 and 1/2 differ at a single residue (Gal versus GalNAc), and both are similar to serotypes 1 and 14: all contain a terminal sialic acid on a side chain. However, despite this similarity, there is complex pattern of cross-protection for these serotypes, with varying estimations of the importance of sialic acid in a protective epitope. Further, a pentasaccharide without the terminal sialic acid has been identified as minimal epitope for serotype 2. Here we use molecular simulation to model the molecule conformations of the CPS in serotypes 2, 1/2, 1 and 14, as well as three vaccine candidate oligosaccharides. The common epitopes we identify assist in rationalizing the apparently contradictory immunological data and provide a basis for rational design of S. suis vaccines in the future
Foundational Ontologies: From Theory to Practice and Back
Full text linkThis is a commentary on the article by Augusto (2022; this issue) on cate-
gories and foundation ontology (FO). We agree that the notion of categories of
kinds of elements to devise a FO deserves more attention than it has received
to date. From a practical point of view sensu developing domain ontologies,
however, it probably does not matter much as long as a FO is used and that
that one was understood
BFO Classifier: Aligning domain ontologies to BFO
Full text linkFoundational ontologies are known to have a steep learning curve, which hampers casual use by domain ontology developers to use them for domain ontology development. Foundational ontology developers have not provided methods or tools to lower the barriers of uptake beyond offering, at best, a computational version. We investigate an approach to bridge this gap through the development of a decision diagram for BFO, which offers the modeller a series of questions with closed answer options in order to step-wise arrive at a suitable entity to align the domain entity to. This diagram was implemented in a tool, the BFO Classifier, that keeps track of the question and answer trace and with the click of a button the alignment axiom can be added to the ontology. It was evaluated with two BFO-aligned ontologies, which showed that in at least half of the alignment axioms, a more precise BFO entity could be selected, and a minority corrected
Towards Run-time Efficient Hierarchical Reinforcement Learning
Full text linkThis paper investigates a novel method combining
Scalable Evolution Strategies (S-ES) and Hierarchical Reinforcement
Learning (HRL). S-ES, named for its excellent scalability,
was popularised with demonstrated performance comparable to
state-of-the-art policy gradient methods. However, S-ES has not
been tested in conjunction with HRL methods, which empower
temporal abstraction thus allowing agents to tackle more challenging
problems. We introduce a novel method merging S-ES
and HRL, which creates a highly scalable and efficient (compute
time) algorithm. We demonstrate that the proposed method
benefits from S-ES’s scalability and indifference to delayed
rewards. This results in our main contribution: significantly
higher learning speed and competitive performance compared
to gradient-based HRL methods, across a range of tasks
Investigating the Usability and Quality of Experience of Mobile Video-Conferencing Apps Among Bandwidth-Constrained Users in South Africa
Full text linkDuring the COVID-19 pandemic and mandated global lockdowns, people and busi- nesses started the extensive use of video-conferencing applications for staying connected. This surge in demand and the usability of video-conferencing services has been severely overlooked in developing countries like South Africa, where one-third of adults rely on mo- bile devices to access the internet, and the per-gigabyte data cost is among the highest in Africa. Considering these numbers, we conduct a two-pronged study where 1) we measure data consumption of different Android apps through data measurement experiments and 2) we conduct interviews and usability assessments with bandwidth-constrained users to bet- ter understand the usability and Quality of Experience (QoE) of mobile video-conferencing apps. Usability is the degree to which specified users can use a product to achieve specified goals. In contrast, QoE measures the subjective perception of the quality of an application and the level of delight or annoyance with a service. The key benefit of this study will be to inform organisations that seek to be inclusive about these tools’ relative usability by letting them know about the factors influencing users’ QoE
A Survey of Multilingual OWL Ontologies in BioPortal
Full text linkThe internationalisation goal for OWL sought to offer support for multilingual ontologies. User-displayable labels were suggested as a way to realise this, by means of rdfs:label. However, because each label is a language-tagged string, this hampers accurate representation of strings in languages that require grammatical features such as inflected forms and gender. At least eight linguistic models have been proposed to address this key shortcoming, with OntoLex-Lemon now the de facto standard. The purpose of this survey was to determine if there has been any adoption of linguistic models within OWL ontologies. As OWL ontologies are widely used in the biomedical domain, the survey was limited to those ontologies in NCBO BioPortal, a biomedical repository. The results indicate that OntoLex-Lemon was not used in any production OWL ontology at time of review, nor that of any other linguistic model. In addition, the adoption rate of multilingualism in OWL ontologies in BioPortal was observed to be 5%, with English the primary language, followed by French and German
INVEST: Ontology Driven Bayesian Networks for Investment Decision Making on the JSE
Full text linkThis research proposes an architecture and prototype implementation of a knowledge-based system for automating share evaluation and investment decision making on the Johannesburg Stock Exchange (JSE). The knowledge acquired from an analysis of the investment domain for a value investing approach is represented in an ontology. A Bayesian network, developed using the ontology, is used to capture the complex causal relations between different factors that influence the quality and value of individual shares. The system was found to adequately represent the decision-making process of investment professionals and provided superior returns to selected benchmark JSE indices from 2012 to 2018
Subword Segmental Machine Translation: Unifying Segmentation and Target Sentence Generation
Full text linkSubword segmenters like BPE operate as a preprocessing step in neural machine translation and other (conditional) language models. They are applied to datasets before training, so translation or text generation quality relies on the quality of segmentations. We propose a departure from this paradigm, called subword segmental machine translation (SSMT). SSMT unifies subword segmentation and MT in a single trainable model. It learns to segment target sentence words while jointly learning to generate target sentences. To use SSMT during inference we propose dynamic decoding, a text generation algorithm that adapts segmentations as it generates translations. Experiments across 6 translation directions show that SSMT improves chrF scores for morphologically rich agglutinative languages. Gains are strongest in the very low-resource scenario. SSMT also learns subwords that are closer to morphemes compared to baselines and proves more robust on a test set constructed for evaluating morphological compositional generalisation