I'm uploading all of my publications here. This is to help me keep track of stuff that I've done, as well as make it easier for others to obtain the PDFs in case of paywall.
Jade Z. Abbott, Laura Martinus NeurIPS 2018 Workshop on Machine Learning for the Developing World
Given that South African education is in crisis, strategies for improvement and sustainability of high-quality, up-to-date education must be explored. In the migration of education online, inclusion of machine translation for low-resourced local languages becomes necessary. This paper aims to spur the use of current neural machine translation (NMT) techniques for low-resourced local languages. The paper demonstrates state-of-the-art performance on English-to-Setswana translation using the Autshumato dataset. The use of the Transformer architecture beat previous techniques by 5.33 BLEU points. This demonstrates the promise of using current NMT techniques for African languages.
[https://arxiv.org/abs/1811.05467]
Laurie Butgereit, Laura Martinus 2018 International Conference on Advances in Big Data, Computing and Data Communication Systems (icABCD)
Tourism is a major contributor to employment in southern Africa and a major contributor to gross domestic products of many southern African countries. One of the major tourist attractions in many southern African countries is the wild animals. Major national parks such as Etosha in Namibia and Central Kalahari in Botswana often have rangers available to assist tourists on their game safaris by recognising animals and describing their habitats. Many of the smaller reserves, however, do not have the luxury of rangers available to tourists. At such smaller reserves, tourists are left on their own to recognise the various animals. This paper describes the use of Google's TensorFlow to create an image recogniser trained for southern African mammals. The recogniser was embedded in an Android mobile app and could then assist tourists at smaller reserves.
[https://ieeexplore.ieee.org/document/8465441]
Laurie Butgereit, Laura Martinus 2018 IST-Africa Week Conference (IST-Africa)
Tourism contributes approximately 3% to the GDP in Africa. One of the big attraction to tourists visiting Africa is the wildlife - both the fauna and flora. The larger game reserves and national parks have a wide variety of supporting facililtes such as good access roads, accommodation, food services, and professional game rangers to assist tourists in identifying plants and animals. The smaller less known game reserves, however, often do not have the luxury of these supporting facilities and often lack professional game rangers. These reserves are often avoided by tourists. The surrounding communities, therefore, miss out on the additional economic advantage of having tourists in the area who also need to purchase petrol, buy food, and find accommodation. The paper investigates the use of Google TensorFlow in identifying African mammals. The actual TensorFlow model is trained using traditional desktop workstations and thousands of photographs. Once the model is created, it can be downloaded to an Android device and used in offline mode. This would allow tourists visiting smaller less known game reserves to identify animals and plants in areas where there is no Internet connectivity. The project was a proof-of-concept and the idea can be expanded to include bird watching clubs, fishing clubs, in addition to national parks. The results are very positive. The Android app developed for this research could reguarly identify 35 common African game mammals.
[https://ieeexplore.ieee.org/document/8417191]
L. J. Martinus, J. J. Hanekom 2017 IEEE AFRICON
In this paper, the problem of speech recognition in noise is considered. The focus of this project was developing a system that uses voice instructions to control a robot in a noisy environment. Speech enhancement of the noisy speech takes place through the use of a Kalman filter, which provides a significant improvement in the quality and intelligibility of the signal. Speech recognition takes place through the use of Hidden Markov models with mel-frequency cepstral coefficients used for feature extraction. The overall accuracy of the developed system is 64% and a 0% error rate when tested with input signals with signal-to-noise ratios varying between -15 dB and 30 dB of additive white Gaussian noise. This paper provides the detailed design and implementation of the system.
[https://ieeexplore.ieee.org/document/8095688]
Fighting obesity: A proposed formula for calculating gamified airtime rewards for using public exercise equipment
Laurie Butgereit, Laura Martinus 2017 IST-Africa Week Conference (IST-Africa)
Health agencies around the world are documenting a growing obesity crisis with the world's population becoming more and more overweight. There have been a number of attempts to tackle the growing obesity problem using smart devices such as walking apps on GPS enabled devices and wearable devices. Such suggested solutions, however, require enough internal motivation on the part of the overweight person to download the app or purchase a wearable device. Another possible solution is to give real world rewards on public exercise equipment. This would provide opportunities to impulse exercise (as opposed to impulse buy). One of the difficulties here, however, is how to price these real world rewards. This paper proposes a generalised formula or algorithm for providing real world rewards on public exercise.
[https://ieeexplore.ieee.org/document/8102310]
AirCycle proof-of-concept: Work towards using gamification and IoT to fight the global obesity crisis
Laurie Butgereit, Laura Martinus 2016 International Conference on Advances in Computing and Communication Engineering (ICACCE)
The world is becoming overweight and obese. A multitude of health organisations provide dire statistics about the prevalance of obesity in modern society. It is a problem which spans both the first world and the third world. To counter this problem, there is a growing number of apps for smart devices and wearable devices which encourage people to exercise. These apps and wearables, however, are often not focused on the poor, the aged, and the infirm who can not afford such devices. This paper looks at the possibility of installing exercise equipment in public places which provide real-world value for exercising. The participant would not need to purchase any devices.