The aims of this research work are to develop a feature detection, description, and matching system for low-resource applications. This work was motivated by the need for a vision sensor to assist the flight of a quad-rotor UAV. This application presented a real-world challenge of autonomous drift stabilization using vision sensors. The initial solution implemented a basic feature detector and matching system on an FPGA. The research then pursued ways to improve the vision system. Research began with color feature detection, and the Color Difference of Gaussians feature detector was developed. CDoG provides better results than gray scale DoG and does not require any additional processing than gray scale if implemented in a parallel architecture. The CDoG Scale-Invariant Feature Transform modification was developed which provided color feature detection and description to the gray scale SIFT descriptor. To demonstrate the benefits of color information, the CDSIFT algorithm was applied to a real application: library book inventory. While color provides added benefit to the CDSIFT descriptor, CDSIFT descriptors are still computationally intractable for a low-resource hardware implementation. Because of these shortcomings, this research focused on developing a new feature descriptor. The BAsis Sparse-coding Inspired Similarity (BASIS) descriptor was developed with low-resource systems in mind. BASIS utilizes sparse coding to provide a generic description of feature characterstics. The BASIS descriptor provided improved accuracy over SIFT, and similar accuracy to SURF on the task of aerial UAV frame-to-frame feature matching. However, basis dictionaries are non-orthogonal and can contain redundant information. In addition to a feature descriptor, an FPGA-based feature correlation (or matching) system needed to be developed. TreeBASIS was developed to answer this need and address the redundancy issues of BASIS. TreeBASIS utilizes a vocabulary tree to drastically reduce descriptor computation time and descriptor size. TreeBASIS also obtains a higher level of accuracy than SIFT, SURF, and BASIS on the UAV aerial imagery task. Both BASIS and TreeBASIS were implemented in VHDL and are well suited for low-resource FPGA applications. TreeBASIS provides a complete feature detection, description, and correlation system-on-a-chip for low-resource FPGA vision systems.

The recently developed state-of-the-art models for Named Entity Recognition are heavily dependent upon huge amounts of available annotated data. Consequently, it is extremely challenging for data-scarce languages to obtain significant result. Several approaches have been proposed to circumvent this issue, including cross-lingual transfer learning, which is the leveraging of knowledge obtained by available resources in the source language and transfer it to a target low-resource language.        Maltese is one of the many majorly underresourced languages. The main purpose of this project is to research how recently developed transformer multilingual models (Multilingual BERT and XLM-RoBERTa) perform and to ultimately set up an evaluation benchmark in zero-shot cross-lingual transfer learning for Maltese Named Entity Recognition. The models are fine-tuned on Arabic, English, Italian, Spanish and Dutch. The experiments evaluated the efficacy of the source languages and the use of multilingual data in both the training and validation stages.         The experiments demonstrated that feeding multilingual data to both the training and the validation phases was mostly beneficial to the performance. However, adding it to the validation phase only was generally detrimental. Furthermore, XLM-R achieved overall better scores however, employing mBERT and English as the source language yielded the best performance.

School feeding programs enhance the efficiency of the education system by improving enrollment, reducing dropouts and increasing perseverance. They also have the potential to reach the poor, directly making them an effective social safety net. In many low-resource countries, school feeding programs are designed to protect children from the effects of hunger. Unfortunately, the continuity of such programs is threatened by over-reliance on external funding. Given the patterns of withdrawal of external support, countries that rely on donor funds to implement such programs need to develop plans that will move them from external to localized support. It is well documented that programs that involve community members are self-sustaining. Regrettably, even though community members are involved in school feeding programs in Malawi, their participation is restricted to food storage and preparation and doesn't include decision making. Thus the transition plan for Malawi has to deliberately involve community members and influence them to take ownership of the school feeding programs. This dissertation explored the use of Social Marketing, a strategy for influencing behavior change that applies traditional marketing techniques to persuade a target audience to adopt, adapt, maintain or reject a behavior for the benefit of individuals, groups, or society as a whole to plan school food programs in Malawian primary schools. Using focus groups and individual interview techniques, I carried out a qualitative study at a primary school in Malawi where the community has initiated a school feeding program with the aim of understanding the barriers and benefits of supporting such an initiative from the community members' perspective. The results show that the cost of producing food, particularly the use of chemical fertilizer, is the main barrier whilst ensuring that all children regardless of social-economic status have access to a meal at school is the drive behind this initiative. The Social Marketing campaign therefore focuses on promoting the use of eco-san toilets whose output is humanure in this school community so as to minimize the cost of producing food to ensure sustainability of this initiative.

Machine Translation (MT) is the task of mapping a source language to a target language. The recent introduction of neural MT (NMT) has shown promising results for high-resource language, however, poorly performing for low-resource language (LRL) settings. Furthermore, the vast majority of the 7, 000+ languages around the world do not have parallel data, creating a zero-resource language (ZRL) scenario. In this thesis, we present our approach to improving NMT for LRL and ZRL, leveraging a multilingual NMT modeling (M-NMT), an approach that allows building a single NMT to translate across multiple source and target languages. This thesis begins by i) analyzing the effectiveness of M-NMT for LRL and ZRL translation tasks, spanning two NMT modeling architectures (Recurrent and Transformer), ii) presents a self-learning approach for improving the zero-shot translation directions of ZRLs, iii) proposes a dynamic transfer-learning approach from a pre-trained (parent) model to a LRL (child) model by tailoring to the vocabulary entries of the latter, iv) extends M-NMT to translate from a source language to specific language varieties (e.g. dialects), and finally, v) proposes an approach that can control the verbosity of an NMT model output. Our experimental findings show the effectiveness of the proposed approaches in improving NMT of LRLs and ZRLs.

Task-oriented dialogue (ToD) systems need to interpret the user's input to understand the user's needs (intent) and corresponding relevant information (slots). This process is performed by a Natural Language Understanding (NLU) component, which maps the text utterance into a semantic frame representation, involving two subtasks: intent classification (text classification) and slot filling (sequence tagging). Typically, new domains and languages are regularly added to the system to support more functionalities. Collecting domain-specific data and performing fine-grained annotation of large amounts of data every time a new domain and language is introduced can be expensive. Thus, developing an NLU model that generalizes well across domains and languages with less labeled data (low-resource) is crucial and remains challenging. This thesis focuses on investigating transfer learning and data augmentation methods for low-resource NLU in ToD. Our first contribution is a study of the potential of non-conversational text as a source for transfer. Most transfer learning approaches assume labeled conversational data as the source task and adapt the NLU model to the target task. We show that leveraging similar tasks from non-conversational text improves performance on target slot filling tasks through multi-task learning in low-resource settings. Second, we propose a set of lightweight augmentation methods that apply data transformation on token and sentence levels through slot value substitution and syntactic manipulation. Despite its simplicity, the performance is comparable to deep learning-based augmentation models, and it is effective on six languages on NLU tasks. Third, we investigate the effectiveness of domain adaptive pre-training for zero-shot cross-lingual NLU. In terms of overall performance, continued pre-training in English is effective across languages. This result indicates that the domain knowledge learned in English is transferable to other languages. In addition to that, domain similarity is essential. We show that intermediate pre-training data that is more similar – in terms of data distribution – to the target dataset yields better performance.

“Exploring the Impact and Roles that Strategic Government Leadership Plays in the Adoption and Use of eHealth in Low Resource Countries: A Quantitative and Descriptive Study of the Medical and Dental Council of Nigeria as a Health Regulatory Agency”. MSc eHealth Defense Candidate: Victor Gbenro Governments of low resource countries (LRCs) have embraced and leveraged the potential benefits Information Computer Technology (ICT) brings to the healthcare sector, taking various steps to adopt and use eHealth to improve healthcare delivery despite recognized challenges. LRCs have been identified as most challenged with the implementation of policies that can drive the development of critical sectors. Despite the development of policies and frameworks in these countries, many still struggle to deliver on their health goals. It is yet to be fully understood to what extent professional health regulatory agencies (PHRAs) understand their roles in effective regulation, as it would relate to medical education, professional conduct and registration of practitioners. The Healthcare workforce is one of the core building blocks of any health system and the regulation of the workforce is central to the provision of quality healthcare services. PHRAs provide strategic leadership through existing legislation, policies, and frameworks and are themselves adopting the use of ICT in a range of applications. These include the registration and licensure process of practitioners, and training and retraining of practitioners through continuing professional development activities. In 2016, the Nigerian government approved the implementation of it National ICT Strategic Framework for health and empowered its agency, the National Information Technology Development Agency (NITDA) which is an agency in the ministry of communication to provide the required leadership, governance, and stewardship in coordinating and improve upon the use of ICT in all key sectors of the country. This study was undertaken to understand the role PHRAs like the Medical and Dental Council of Nigeria (MDCN) a regulatory body for the professions of medicine and dentistry in Nigeria provides Strategic Government Leadership (SGL) in the adoption and use of eHealth tools through policies and legislation in the health sector. The study also assessed the knowledge and perception of employees of the (MDCN) on existing eHealth policies and legislation and of their relevancy or adequacies in providing effective regulation. The study answers the research questions i) does a relationship exist between SGL and capacity for eHealth innovation and technological/ infrastructural development? ii) What are the measures taken and the importance of the security and privacy of practitioner records to PHRA? and iii) Does SGL, as demonstrated through policy development, affect the adoption and use of eHealth by employees of the PHRA? A systematic literature review was performed, and a structured questionnaire was administered to MDCN professional staff. The results were subjected to statistical analysis to investigate relationships between the dependent variable (SGL) and 14 independent variables representing the 15 constructs from the questionnaire. A regression model found four significant predictors of the value of the dependent variable. A study of other related PHRAs is recommended to improve the suitability of the framework proposed, considering the limitations of this study. / Thesis / Master of Science (MSc)

No / A culturally sensitive approach needs to be adopted in disseminating evidence-based preventive programs internationally, and very little is known about effective dissemination into low-resource settings such as low and middle income countries. Following guidelines on optimizing the fit of evidence-based parenting programs worldwide, a cultural relevance study was conducted in Panama, Central America. Parents (N = 120) from low-resource communities were surveyed to explore cultural relevance of material from the Triple P-Positive Parenting Program. Intention to participate and views on delivery formats and program features were also examined. Descriptive statistics and regressions were carried out to analyze the results. Parents found program materials highly relevant and reported that they would be willing to participate in a program if one was offered. A large proportion of the sample expressed a preference for self-directed formats such as books, articles and brochures (77.6 %). Regression analyses suggested that most parents considered material as relevant, interesting and useful, regardless of other factors such as socio-economic status, gender, the level of child behavioral difficulties, parental stress, parental confidence and expectations of future behavioral problems. The study provides a potential approach for dissemination of research and offers an insight into the needs and preferences of a particular segment of the world’s population—parents in low-resource settings. Strategies for meeting the needs and preferences of these parents in terms of service delivery are discussed.

No / Several international organizations have suggested the need for disseminating existing evidence-based parenting interventions into low-resource settings of the world in order to prevent societal difficulties such as violence. Before dissemination efforts take place, it is important to examine the fit of existing interventions in these contexts. In the present study, 80 practitioners from low-resource communities in Panama, Central America, were surveyed in order to explore their views on materials, principles and strategies of an evidence-based parenting program, the Triple P Positive Parenting Program. This study is part of a larger project in which cultural relevance was also explored from parents’ perspective, instruments were translated and validated, and a RCT was carried out to determine efficacy. Practitioners in the present study were psychologists, teachers, social workers and learning disability specialists based in school settings. Descriptive statistics were used to analyze the data and regression analyses were carried out in order to determine whether socio-demographic variables predicted acceptability scores. Scores for cultural relevance and usefulness of the program were high. A sample of material was found to be interesting, familiar, and acceptable. All practitioners (100 %) expressed a need to implement a parenting program in their community. Only being female and greater hours of consultation per week were associated with greater acceptability. These results have the potential to inform implementation efforts in Panama and the study offers a methodology which can be used to explore the relevance of other programs in other low-resource settings.

Machine learning can provide assistance to humans in making decisions, including linguistic decisions such as determining the part of speech of a word. Supervised machine learning methods derive patterns indicative of possible labels (decisions) from annotated example data. For many problems, including most language analysis problems, acquiring annotated data requires human annotators who are trained to understand the problem and to disambiguate among multiple possible labels. Hence, the availability of experts can limit the scope and quantity of annotated data. Machine-learned pre-annotation assistance, which suggests probable labels for unannotated items, can enable expert annotators to work more quickly and thus to produce broader and larger annotated resources more cost-efficiently. Yet, because annotated data is required to build the pre-annotation model, bootstrapping is an obstacle to utilizing pre-annotation assistance, especially for low-resource problems where little or no annotated data exists. Interactive pre-annotation assistance can mitigate bootstrapping costs, even for low-resource problems, by continually refining the pre-annotation model with new annotated examples as the annotators work. In practice, continually refining models has seldom been done except for the simplest of models which can be trained quickly. As a case study in developing sophisticated, interactive, machine-assisted annotation, this work employs the task of corpus-dictionary linkage (CDL), which is to link each word token in a corpus to its correct dictionary entry. CDL resources, such as machine-readable dictionaries and concordances, are essential aids in many tasks including language learning and corpus studies. We employ a pipeline model to provide CDL pre-annotations, with one model per CDL sub-task. We evaluate different models for lemmatization, the most significant CDL sub-task since many dictionary entry headwords are usually lemmas. The best performing lemmatization model is a hybrid which uses a maximum entropy Markov model (MEMM) to handle unknown (novel) word tokens and other component models to handle known word tokens. We extend the hybrid model design to the other CDL sub-tasks in the pipeline. We develop an incremental training algorithm for the MEMM which avoids wasting previous computation as would be done by simply retraining from scratch. The incremental training algorithm facilitates the addition of new dictionary entries over time (i.e., new labels) and also facilitates learning from partially annotated sentences which allows annotators to annotate words in any order. We validate that the hybrid model attains high accuracy and can be trained sufficiently quickly to provide interactive pre-annotation assistance by simulating CDL annotation on Quranic Arabic and classical Syriac data.

A Thesis submitted to The University of Arizona College of Medicine - Phoenix in partial fulfillment of the requirements for the Degree of Doctor of Medicine. / The role of enzyme‐assisted subcutaneous fluid administration (EASFA) in treating mild to moderate dehydration in pediatrics, geriatrics, and palliative care has been studied in developed countries. However, it has historically been underutilized due to widely available health care and alternative treatments, namely peripheral intravenous (IV) fluid administration. Fluid infusions in the subcutaneous tissue have a low risk of infection, are easy to administer, and have wide potential use. The use of EASFA in low resource settings to treat those with difficult IV access or where skilled healthcare workers are not as readily available could prove to be a live saving measure in many situations, including the care of patients in remote areas of the world, mass casualty events, or other disasters. Our objective was to determine if EASFA is a valid and appropriate technique to utilize in pediatric and elderly patients, and evaluate if it could be a safe and efficient way to provide fluid resuscitation in low resource settings. For this systematic review MEDLINE and Cochrane Library were searched from January 1950 to December 2015 to recover all available literature relevant to this topic. Studies that met the inclusion criteria were analyzed using Cohen’s D. This was calculated using the mean difference between intervention and control divided by the pooled standard deviation. For dichotomous outcome of the placement success rate the odds ratios were calculated with 95% confidence intervals. In reviewing 7 articles using Cohen’s D to compare mean differences to determine effect size, we found that catheter placement success rates and infusion rates were similar between EASFA and peripheral intravenous fluid administration. Additionally, it was found that the odds of correct initial needle placement was 7.19 times higher in EASFA versus intravenous administration. EASFA is a comparable alternative to intravenous fluid administration when delivering fluids to pediatric and elderly patients with mild to moderate dehydration. While infusion rates and total volume of fluids administered were similar, the high rate of success with placement of the subcutaneous catheter proves it to be more useful in some situations. Venous cannulation is difficult, even for a trained healthcare provider, and the ease of placement of subcutaneous catheters makes training lay people to administer subcutaneous fluids a possibility. Additionally, this type of fluid administration may lead to less psychological trauma to a child from multiple needle sticks, while still achieving a similar outcome of effective volume replacement. Based on the results of this study, further research is needed to evaluate the effectiveness of utilizing EASFA in low resource settings.