Abstract

To develop an Arabic text to Moroccan Sign Language (MSL) translation product through building two corpora of data on Arabic texts for the use of translation into MSL. The collected corpora of data will train Deep Learning Models to analyze and map Arabic words and sentences against MSL encodings.

Introduction

Over 5% of the world’s population (466 million people) has disabling hearing loss. 4 million are children [1]. They can be hard of hearing or deaf. Hard of hearing people usually communicate through spoken language and can benefit from assistive devices like cochlear implants. Deaf people mostly have profound hearing loss, which implies very little or no hearing.

The main impacts of deaf is on the individual’s ability to communicate with others in addition to the emotional feelings of loneliness and isolation in society. Consequently, they can not equally access public services, mostly education and health and have not equal rights in participating at the active and democratic life. This leads to a negative impact in their lives and the lives of the people surrounding them. Over the world, deaf people often communicate using a sign language with gestures of both hands and facial expressions.

Sign languages are full-fledged natural  languages with their own grammar and lexicon. However, they are not universal although they have striking similarities. In Morocco, deaf children receive very few education assistance. For many years, they were learning the local variety of sign language from Arabic, French, and American Sign Languages [2]. In April 2019, the governement standardized the Moroccan Sign Language (MSL) and initiated programs to support the education of deaf children [3]. However, the involved teachers are mostly hearing, have limited command of MSL and lack resources and tools to teach deaf learn from written or spoken text. Schools recruit interpreters to help the student understand what is being teached and said in class. Otherwise, teachers use graphics and captioned videos to learn the mappings to signs, but lack tools that translate written or spoken words and concepts into signs. This project comes to solve this particular issue.

Objectives

We propose an Arabic Sppech-to-MSL translator. The translation could be divided into two parts, the speech-to-text part and the text-to-MSL part. In a previous work [4], we was interested in the arabic Speech-to-Text translation. We conducted a research and a comparison on the existing Speech-to-Text APIs. A web application was built for this end [check web app here]. Our main focus in this current work is to take the results from the Speech-to-Text module and perform Text-to-MSL translation.

Up to now, there is not enough data of arabic words with their translations into MSL. This is of course challenging because we have to bring interpreters and linguists together in order to create this initial corpus. Each word or concept in the arabic corpus could be mapped to a time series of hand gestures and facial expressions in the MSL corpus. Our main objective is to find the best possible mapping between these two corpuses. The collected data will allow us to train big Deep Learning Models. In fact, we aim to explore the existing deep learning pretrained architectures suitable for analyzing arabic words and sentences and find their mappings with the MSL encodings. Recurrent Neural Networks using GRU and LSTM units and their variants are proved to be the most suitable and powerful when dealing with sequential data [5][6]. We believe that tuning these state of the art models will allow us to achieve good generalization performances.

Expected outcomes

With this work we expect building the MSL and the Arabic text corpuses that we aim keeping free and open for public use. For this end, we will develop an interactive web application for the creation of the two corpuses. The Text-to-MSL translation product will be hosted on a web and mobile application.

Abstract

To test the feasibility of the deployment of Unmanned Ground Vehicles (UGVs) for automated intelligent patrol, detection, wildlife monitoring, identification across the national parks and reserves in Kenya.

Introduction

Wildlife tourism is a significant and growing contributor to the economic and social development in the African region through revenue generation, infrastructure development and job creation. According to a recent press release by the World Travel and Tourism Council [1], travel and tourism contributed $194.2
billion (8.5% of GDP) to the African region in 2018 and supported 24.3 million jobs (6.7% of total employment). Globally, travel and tourism is a $7.6 trillion industry, and is responsible for an estimated 292 million jobs [2]. Tourism is also one of the few sectors in which female labor participation is already above parity, with women accounting for up to 70% of the workforce [2].

However, the wildlife tourism industry in Africa is being increasingly threatened by rising human population and wildlife crime. As poaching becomes more organised and livestock incursions become frequent occurrences, shortages in ranger workforce and shortcomings in technological developments in this space put thousands of species at risk of endangerment, and threaten to collapse the wildlife tourism industry and ecosystem. According to The National Wildlife Conservation Status Report, 2015 – 2017, presented by the Ministry of Tourism and Wildlife of Kenya [3], there is currently a shortage of 1038 rangers, from the required 2484 rangers in Kenyan national parks and reserves, a deficit of over 40%. With tourism in Kenya contributing a revenue of $1.5 billion in 2018 [4], and with the wildlife conservancies in Kenya supporting over 700,000 community livelihoods [3], the recession of the wildlife tourism industry could have major adverse economic and social impacts on the country. It is thus critical that sustainable solutions are reached to save the wildlife tourism industry, and further research is fuelled in this area.

The national parks, reserves and conservancies in Kenya span thousands of square kilometers and make it difficult for rangers to track down all possible poaching activities. Poachers normally use guns, snares, and poison to capture wild animals. By collecting real world data on poaching activities, better learning of adversarial behavior is achieved and optimal strategies for anti-poaching patrols can be employed [5]. According to [5], a large number of security games research lacks actual adversary data and does not consider heterogeneity among large populations of adversary which makes it difficult to build accurate models of adversary behavior. Other problems inherent in past predictive models neglect the uncertainty in crime data, use coarse-grained crime analysis and propose time-consuming techniques that cannot be directly integrated into low-resource outposts [6].

To address shortages in ranger workforce, carry out monitoring activities more effectively, and detect criminal or endangering activities with greater precision, we propose the development of an open dataset containing georeferenced data on poaching incidents from the past 10 years as well as historical data on tagged elephant and rhino movements. We aim to observe correlations between the data using machine learning models and effectively model poaching trends and behavioural patterns to predict the location of the next poaching attack and suggest better patrol routes. The study will be carried out over a period of 4 months at Nairobi National Park in Kenya which covers a total area of 117 square kilometers and hosts many of the endangered wildlife species listed in the IUCN Red List of Threatened Species, such as the African Elephant and Black Rhinoceros.

Objectives

1. To generate a real world dataset that maps poaching activities within the park.
2. To develop a hybrid model that predicts the behavior of poachers by capturing their heterogeneity.
3. To improve the accuracy of the hybrid model by creating novel algorithms in determining poaching
   activities and hotspots.
4. To investigate the computation challenges faced when learning the behavioral model of poachers.

Vision

Our future vision is to test the feasibility of the deployment of Unmanned Ground Vehicles (UGVs) for automated intelligent patrol and wildlife monitoring across the national parks and reserves in Kenya. In
addition to carrying out automated patrol using the models learned in this study, the UGVs would be fitted
with an array of cameras and sensors that would enable it to navigate autonomously within the parks, and run multiple deep learning and computer vision algorithms that carry out numerous monitoring activities such as detection of poaching, livestock incursions, human wildlife conflict, distressed wildlife, and species
identification.

References

[1] “African tourism sector booming – second-fastest growth rate in the world”, WTTC press release , Mar. 13, 2019. Accessed on Jul. 11, 2019. [Online]. Available: https://www.wttc.org/about/media-centre/press-releases/press-releases/2019/african-tourism-sector-booming-second-fastest-growth-rate-in-the-world/

[2] “Supporting Sustainable Livelihoods through Wildlife Tourism”, World Bank Group , 2018.

[3] “The National Wildlife Conservation Status Report, 2015 – 2017”, pp. 75, 131, Ministry of Tourism and
Wildlife, Kenya , 2017.

[4] “Tourism Sector Performance Report – 2018” , Hon. Najib Balala, 2018.

[5] R. Yang, B. Ford, M. Tambe, and A. Lemieux, “Adaptive resource allocation for wildlife protection against illegal poachers,” in Proceedings of the 2014 International Conference on Autonomous Agents and Multi-agent Systems , May 2014, pp. 453-460.

[6] S. Gholami, S. McCarthy, B. Dilkina, A. Plumptre, M. Tambe, et. al., “Adversary models account for
imperfect crime data: Forecasting and planning against real-world poachers,” in Proceedings of the 17th
International Conference on Autonomous Agents and MultiAgent Systems , July 2018, pp. 823-831.

Abstract

To determine the effectiveness of Long Short Term Memory Network in the prediction of pregnant mothers at high risk of developing pre-eclampsia and the effectiveness of prophylaxis of preeclampsia.

Background

The Sustainable Development Goal (SDG) 3 aims to reduce the global maternal mortality ratio to less than 70 per 100,000 live births. These deaths are caused by a number of conditions experienced during pregnancy and childbirth. Preeclampsia has adverse effects on maternal health especially in low and middle-income countries. Many challenges persist in the prediction, prevention and management of preeclampsia. Prophylytic measures such as low dose aspirin and calcium supplementation has been used western countries however more evidence is needed before it can be used in developing countries. The current management is timely diagnosis; proper management, timely delivery and good follow up after birth. This study therefore seeks to explore the use of wearable devices to continuously measure blood pressure in pregnant mothers. The obtained blood pressure data will be used to predict future maternal blood pressures using Long Short Term Memory (LSTM) recurrent neural networks on mobile devices. Those whose blood pressures will be predicted to be high, therefore risking development of preeclampsia will be put into two groups: one will receive the usual care in the high risk clinic while the second group will be supplemented with low dose asprin and calcium from second trimester. It is expected that the prediction will serve to identify those at risk early and management instituted immediately and those supplemented with low dose aspirin and calcium will not develop preeclampsia. Additionally the data collected will be valuable for future studies in the area of preeclampsia prediction using machine learning.

Introduction

Preeclampsia is a pregnancy complication characterized by persistent high blood pressure. It usually begins after 20 weeks of pregnancy in women whose blood pressure (BP) has been normal. If left untreated it will progress to eclampsia that is often fatal to both mother and baby. (Macdonald-Wallis, C et al 2015). Preeclampsia is often diagnosed when a mother goes to the health care facility for routine check where BP measurement is taken. The first sign of preeclampsia is a BP reading exceeding 140/90 in two or more occasions, at least four hours apart at 20 or more week’s gestation. Most pregnant mothers in Low and Middle Income Countries do not have personal BP machines to take regular BP readings thus they depend on BP reading during the antenatal clinic visits, which are 4-5 for the entire pregnancy. Early detection of preeclampsia is often missed during these visits because the BP measurement is often taken once unless otherwise indicated during the visit.

The detection and management of preeclampsia in out of clinic settings has however become much easier in the recent past through the development of smart blood pressure monitors. These devices that are now readily available on the market use a variety of non intrusive methods such as a cuff that inflates slightly to measure systolic and diastolic pressure via the oscillometric method as is the case with the Omron Smart watch (Omron 2019) and using a combination of optical sensors and clinically validated software algorithms as is the case with a number of smart watches such as the one developed by Aktiia (2018) and Bpro by MedTach Inc (2018). These devices are not only able to take readings and generate alarms but are also capable of transmitting this data to other devices such as mobile phones for use in further analysis using techniques such as machine learning cite.

The use of machine learning techniques for blood pressure prediction is a practice that is steadily growing using techniques such as Artifical Neural Networks (Hao et al, 2015), as well as classification and  regression trees (Zhang et al, 2018). Additionally Long Short Term Memory (LSTM) networks are increasingly being considered in studies such as the ones by Su el al (2017), Zhao et al (2019), Lo et al (2017) and Radha et al (2019). A majority of these techniques, current studies and solutions are developed and deployed on devices with significant computing and storage power such as servers and super computers which presents a major challenge overall in the potential utility of machine learning for individuals who increasingly prefer to access services, content and solutions on their mobile devices.

Preeclampsia remains a significant public health problem for both the developed and developing countries contributing to both maternal morbidity and mortality globally (McClure, Saleem, Pasha, & Goldenberg,  2009; Shah et al., 2009) however the impact of the disease is felt more severely in the developed countries (Prakash et al., 2010) where unlike other causes of mortality, medical intervention may be ineffective due to late presentation (Jido & Yakasai, 2013) . The problem is confounded by continuous mystery of the aetiology and unpredictable nature of the disease (Jido & Yakasai, 2013) . In developing countries supplementation of low dose aspirin calcium is used (Anderson & Schmella, 2017) are used as prophyxis for preeclampsia, however, further evidence is needed before it can be adopted in developing countries such as Kenya. The aim of this study is two fold: to determine the effectiveness of Long Short Term Memory Network in the prediction of those at high risk of developing preeclampsia and effectiveness of low dose aspiring and calcium in the prophylaxis of preeclampsia in those at risk

Objectives

1.  To determine the effectiveness of Long Short Term Memory Network in the prediction of those at high risk of developing preeclampsia.
2. To determine the effectiveness of low dose aspirin and calcium supplementation in the prophylaxis of preeclampsia in those at risk

Abstract

To develop a methodology for a semi-automatic classification of judgments disseminated by the High Court Library of the Malawi Judiciary with the purpose of enabling ‘intelligent searching’ within this body of knowledge.

Introduction

Challenges of Legal Research in Malawi Malawi faces a serious problem when it comes to law reporting [5]. The Official Law Reports have been discontinued; the African Law Reports Malawi Series and the Malawi Law Reports cover only the period 1923 – 1993. The MalawiLII website [9], which is the Malawi section on the Southern Africa Legal Information Institute SAFLII, is an online resource contains court judgments issued since 1993 and some statutory laws. However, it is not complete and not easily searchable. Paid services such as Blackhall’s Laws of Malawi contain all the statutory laws (Principal and Subsidiary Legislation) of Malawi in force available at one source on the Internet in an updated and consolidated form. However this is only accessible to paid members and it comes at a substantial cost. The High Court of Malawi maintains a section with printed judgments organised in folders by year and court. However, the indexing used is too rough. The High Court Library also has a paid email subscription service, though which members received scanned images of judgments. However, these are not in a searchable form. Commentaries and digests are very rare and most sections of the law do not have any such publications, e.g., the Criminal Law. There are also private libraries that may be maintained by various law firms.

Problem Statement

In Malawi, the legal research faces significant challenges in accessing and searching for relevant information. On one hand are the issues of accessibility and the availability or the scattered nature of the official reports. On the other hand are the challenges coming from the fact that the current document structure of Malawi legal text, e.g., court judgments, does not support a system of citation that makes it possible to link statutory law, case law and secondary law or to search by “legal terms” and their specific interpretations. This research tackles the specific problem of classifying court judgments disseminated by the High Court Library. The court judgments disseminated via the Malawi High Court Library are not classified according to useful categories, such as courts, topics of the law, statues they refer to. They do not have an index and the structure of the documents is not uniform. The internal structure of judgments impacts the efficiency of a search [2,4,6].

Objectives

The aim of this research is to develop a methodology for an semi-automatic
classification of judgments disseminated by the High Court Library of the Malawi Judiciary with
the purpose of enabling ‘intelligent searching’ within this body of knowledge. Specifically, we have
the following sub-objectives.

1. To test the efficiency of the search tool available at the moment in the MalawiLII website.
2. To build an automatic tool for identifying and extracting the general structure of court judgments in Malawi.
3. To build a semi-automatic tool for extracting key meta-data from court judgments: type of case, involved parties, key legal terms, and laws and statues referred to in the judgment.

References

[1] V. R. Benjamins, P. Casanovas, J. Breuker, and A. Gangemi. Law and the semantic web, an
introduction. In Law and the Semantic Web, pages 1–17. Springer, 2005.

[2] Atefeh Farzindar and Guy Lapalme. ‘LetSum, an automatic Legal Text Summarizing system’ in T. Gordon (ed.), Legal Knowledge and Information Systems. Jurix 2004: The Seventeenth Annual Conference. Amsterdam: IOS Press, 2004, pp. 11-18.

[3] Heinrich H. Dzinyemba, Subject Index of Cases Unreported: Civil and Criminal Cases 1997 – 2003’, Malawi High Court Manuscript.

[4] H. Igari, A. Shimazu, and K. Ochimizu. Document structure analysis with syntactic model and parsers: Application to legal judgments. In JSAI International Symposium on A.I., pages 126–140, 2011.

[5] Judge Kapindu’s description of the Malawi Legal System notes in 2014 http://www.nyulawglobal.org/globalex/Malawi1.html#_edn70

[6] Marios Koniaris George Papastefanatos Yannis Vassiliou, Towards Automatic Structuring and Semantic Indexing of Legal Documents, PCI ’16, November 10 – 12, 2016, Patras, Greece.

[7] Q. Lu, J. G. Conrad, K. Al-Kofahi, and W. Keenan. Legal document clustering with built-in topic segmentation. In Proceedings of CIKM ’11, pages 383–392, 2011.

[8] Daniel Locke, G. Zuccon, & H. Scells. Automatic query generation from legal texts for case law retrieval. In 13th Asia Information Retrieval Societies Conference (AIRS 2017), 2017, Jeju, Korea.

[9] MalawiLII Website

[10] Xiaojun Wan and Jianguo Xiao. Single Document Keyphrase Extraction Using Neighborhood Knowledge, Proceedings of the Twenty-Third AAAI Conference on Artificial Intelligence, 2008.

[11] Adam Wyner, Raquel Mochales-Palau, Marie-Francine Moens, and David Milward, Approaches to Text Mining Arguments from Legal Cases, JURIX 2008.

Abstract

Pest monitoring by using a data-driven computer vision technique in directing the extension officers support services across sub-Sahara Africa in a real-time pest damage assessment and recommendation support system for small scale tomato farmers.

Problem situation

Agriculture is a vital tool for sustainable development in Africa. A high yielding crop such as tomato with high economic returns can greatly increase smallholder farmers income when well managed. Despite the socio-economic importance of tomato that produce market opportunity, food and nutritional security for smallholder grower, it is apparently constrained by the recent invasion of tomato pest Tuta absoluta that is devastating tomato yield causing loss of up to 100% hence jeopardizing livelihoods of millions of growers in sub-Sahara Africa [1]. This puts small scale farmers at risk of losing income. Tuta absoluta, has swept across Africa, leading to the declaration of a state of emergency [2][3] in some of the continent’s main tomato producing areas. Furthermore, the lack of adequate capacity to detect and implement management measures. A shift from a reactive to a more proactive intervention based on the internationally recognized threestage approach of prevention, early detection and control is needed to be adopted. This work focus on early detection, a novel approach in initiatives to strengthen phytosanitary capacity and systems to help solve Tuta absoluta devastation.

Objective

This work will radically transform Tuta absoluta pest monitoring by using a data-driven computer vision technique in directing the extension officers support services [4] across sub- Sahara Africa in a real-time pest damage assessment and recommendation support system for small scale tomato farmers. To the best of our knowledge, it will be the first alternative approach using computer vision to help aviate the current alarming situation of invasive tomato pest Tuta absoluta by providing solutions that could help in early management and control. We aim to increase the effectiveness of limited farm-level extension support by leveraging emerging technological [5] and extension support to targeted affected areas (based on damage status maps) using our developed models based on quantified images of pest damage.

Justification

Pests and diseases are major threat to smallholder farmers [6] however, Tuta absoluta control still rely on low-speed inefficient manual identification and a few on the support of limited number of agriculture extension officers [7]. With application of computer vision based image recognition technology, early identification and quantification of Tuta absoluta damage status using recent improvements of tele-infrastructure and information technology will give new tools to deploy the start-of-art of computer vision [8] [9] [10] therefore giving a more targeted control needs to be taken in phytosanitary measures of Tuta absoluta.

Preliminary works and Expected outcomes

Our hypothesis is that current emerging technology can be integrated into a decision platform for tomato pest management and can provide diagnostics in real-time at minimal human capacity training. However, we leverage to extend and integrate alternative support such as recent discovery of a promising pesticide by our fellow team member, Ms. Never. We also anticipate that advice from limited extension service can be delivered to large numbers of smallholder farmers. We fully expect the proposed work to succeed. To achieve this, the first steps of this work have already been completed over the last 12 months through field work and in-house experiment to collect data using cameras and drones in affected areas of Arusha and Morogoro, Tanzania. We have taken and labeled over 4,000 images of tomatoes and multispectral images (RGB, infra-red, red edge allowing for vegetation indices data collection like NDVI) and trained convolutional neural network model. The models can classify Tuta absoluta damage cases. This work also emerged as computer vision for global challenge workshop (CV4GC) winner presented at Computer Vision Pattern Recognition (CVPR). The multidisciplinary research team and links to major key players such as Sokoine University of Agriculture, NM-AIST, agriculture extension officers have helped in initial works. A combination of different technical skills and background could be the best approach in tackling the  apparent state-of-emergency of Tuta absoluta invasion. Since we expect our work to have major impact, we will test how Tuta absoluta pest damage map assessment could increase yield in tomato value chains in Tanzania and sub-Sahara Africa.

Reference

[1] Z. Never, A. N. Patrick, C. Musa, and M. Ernest, “Tomato Leafminer, Tuta absoluta (Meyrick 1917), an emerging agricultural pest in Sub-Saharan Africa: Current and prospective management strategies,”African J. Agric. Res., vol. 12, no. 6, pp. 389–396, 2017.

[2] Nigeria’s Kaduma state declares ‘tomato emergency’ [Online] Available https://www.bbc.com/news/world-africa-36369015 Accessed: 30th July, 2018.

[3] Invasive Africa: Tuta absoluta. [Online] Available https://www.youtube.com/watch?v=_dubR2qoW8k Accessed: 24th September, 2018.

[4] T. J. Maginga, T. Nordey, and M. Ally, “Extension System for Improving the Management of Vegetable Cropping Systems,” vol. 3, no. 4, 2018.

[5] Zahedi, Seyed Reza, and Seyed Morteza Zahedi. “Role of information and communication technologies in modern agriculture.” International Journal of Agriculture and Crop Sciences 4, no. 23 (2012): 1725- 1728.

[6] V. Mutayoba, T. Mwalimu, and N. Memorial, “Assessing tomato farming and marketing among smallholders in high potential agricultural areas of Tanzania Venance ,” no. July, pp. 0–17, 2017.

[7] R. Y. A. Guimapi, S. A. Mohamed, G. O. Okeyo, F. T. Ndjomatchoua, S. Ekesi, and H. E. Z. Tonnang, “Modeling the risk of invasion and spread of Tuta absoluta in Africa,” Ecol. Complex., vol. 28, pp. 77–93, 2016.

[8] Y. Lecun, Y. Bengio, and G. Hinton, “Deep learning,” 2015.
[9] K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,”
pp. 1–14, 2015.
[10] H. Kaiming, Z. Xiangyu, R. Shaoqing, and S. Jian, “Deep Residual Learning for Image Recognition.”
[11] H. Peng, Y. Bayram, L. Shaltiel-Harpaz, F. Sohrabi, A. Saji, U.T. Esenali, A. Jalilov . “Tuta absoluta
continues to disperse in Asia: damage, ongoing management and future challenges.” Journal of Pest
Science(2018): 1-11.