MMMark Abraham Magumba

Mark Abraham Magumba

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Mark Abraham Magumba has a total of 18 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    N for Parameter: Efficient Multi-Scale Neural Audio Super Resolution with GAN

    by Mark Magumba, Steven Davy, Owais bin Zuber

    We introduce N-GAN, an end-to-end GAN architecture for neural audio super resolution that can accommodate multiple input sample rates. We refer to our approach as “Wave-to-Wave” to distinguish it from the “Wave-to-Spectrogram-to-Wave” and “Wave-and-Spectrogram-to-Wave” approaches upon which the state-of-the-art results on this task are based. Our proposed “Wave-to-Wave” architecture produces models that are orders of magnitude smaller than current state-of-the-art models whilst matching or exceeding their performance. In addition, our approach improves inference speed by at least 150% (2.5x speedup) over previous similarly performant models. We show that our model obtains state-of-the-art performance on a target sample rate of 48kHz and input sample rates of 8kHz, 16kHz and 24kHz.

  • Open Access

    Neural-Semantic Prediction of Prenatal EPDS Subscale 3A Categories Using Socio-Demographic Features

    by Mark Abraham Magumba, Hasifah Kasujja, Tonny Justus Oyana

    In this work we demonstrate the use of data-driven Machine learning approaches to predict prenatal anxiety. We rely on a large dataset of Edinburgh Postnatal Depression Scale (EPDS) scores compiled from women in 9 hospitals within Kampala, Uganda between January 2022 and April 2022. A subset of the EPDS scale, the EPDS subscale 3A, can be employed to measure anxiety. Respondents who score 5 and above on this subscale are flagged as possibly anxious whereas those below this cut-off are considered normal. We show that given enough data it may be possible to use socio-demographic markers to predict EPDS subscale 3A categories. In addition, we introduce a lexical-semantic method for handling socio-demographic features allowing us to reformulate the problem as a text classification problem and obtain markedly higher recall (sensitivity) performance with deep neural networks.


  • Open Access

    E-Screening for Prenatal Depression in Kampala, Uganda Using the Edinburgh Postnatal Depression Scale: Survey Results

    by Hasifah Kasujja Namatovu, Mark Abraham Magumba, Dickens Akena

    Background: Perinatal depression remains a substantial public health challenge, often overlooked or incorrectly diagnosed in numerous low-income nations. Objective: The goal of this study was to establish statistical baselines for the prevalence of perinatal depression in Kampala and understand its relationship with key demographic variables. Methods: We employed an Android-based implementation of the Edinburgh Postnatal Depression Scale (EPDS) to survey 12,913 women recruited from 7 government health facilities located in Kampala, Uganda. We used the standard EPDS cutoff, which classifies women with total scores above 13 as possibly depressed and those below 13 as not depressed. The χ2 test of independence was used to determine the most influential categorical variables. We further analyzed the most influential categorical variable using odds ratios. For continuous variables such as age and the weeks of gestation, we performed a simple correlation analysis. Results: We found that 21.5% (2783/12,913, 95% CI 20.8%‐22.3%) were possibly depressed. Respondents’ relationship category was found to be the most influential variable (χ21=806.9, P<.001; Cramer’s V=0.25), indicating a small effect size. Among quantitative variables, we found a weak negative correlation between respondents’ age and the total EPDS score (r=−0.11, P<.001). Similarly, a weak negative correlation was also observed between the total EPDS score and the number of previous children of the respondent (r=−0.07, P<.001). Moreover, a weak positive correlation was noted between weeks of gestation and the total EPDS score (r=0.02, P=.05) Conclusions: This study shows that demographic factors such as spousal employment category, age, and relationship status have an influence on the respondents’ EPDS scores. These variables may serve as proxies for latent factors such as financial stability and emotional support.


  • Open Access

    A bird vocalisation dataset of birds in Uganda for automated bio-acoustic monitoring and analysis

    by Mark Abraham Magumba, Tom Denton, Mutesasira Bashir


  • Open Access

    Electronic Screening Of Perinatal Depression In Kampala, Uganda – Survey Results (Preprint)

    by Hasifah Kasujja Namatovu, Mark Abraham Magumba, Dickens Akena

    BACKGROUND New and expectant mothers face several unique health challenges related to the physical and psychological changes accompanying motherhood. However, in low-income settings where mothers grapple with meeting basic physical needs like access to safe water, proper nutrition, access to health facilities and health workers, psychological needs may end up being ignored. Psychological needs are an important factor that can impact the physical health of the mother and the development of the child. OBJECTIVE In this paper, we report key findings from a survey of perinatal depression in 7518 women in Kampala, Uganda. The goal of the exercise was to establish statistical baselines for the prevalence of perinatal depression in Kampala and understand its relationship with key demographic variables. METHODS We employed an Android implementation of the Edinburgh Postnatal Depression Scale (EPDS). The scale gives depression scores from 0 to 30. Results were transmitted to a web-based backend MySQL server. Two EPDS depression classification scales were employed. The more common coarse-grained binary classification flags respondents who score 13 on the EPDS and above as possibly depressed. RESULTS Based on this classification it was found that 21.83% (95% CI: 20.91%–22.79%) of respondents were possibly depressed. We also employed a more detailed fine-grained classification, which classifies scores of 6 or less as minimal or mild depression, scores of 13 or less as mild depression, scores of 19 or less as moderate depression and scores greater than 20 as severe depression. Based on this classification, 60.59% (95% CI: 59.47%–61.69%) of respondents were found to be mildly depressed. 19.59% (95% CI: 18.7%–20.51%) of respondents were flagged as moderately depressed and 17.58% (95% CI: 16.73%–18.46%) were flagged as not or minimally depressed and 2.23% (95% CI: 1.91%–2.59%) were found to be severely depressed. CONCLUSIONS Spousal employment category was found to be a key determinant for respondents’ overall depression and anxiety score and respondent’s relationship status was found to be an important factor for suicidal ideation. This conclusively means that financial stability greatly influences mental stability and as such, advocacy to have men support their spouses is crucial

  • Barriers and Facilitators of eHealth Adoption Among Healthcare Providers in Uganda – A Quantitative Study

    by Hasifah K. Namatovu, Agnes R. Semwanga, Vincent M. Kiberu, Livingstone Ndigezza, Mark A. Magumba, Swaib K. Kyanda


  • Open Access

    Evaluation of different machine learning approaches and input text representations for multilingual classification of tweets for disease surveillance in the social web

    by Mark Abraham Magumba, Peter Nabende

    Twitter and social media as a whole have great potential as a source of disease surveillance data however the general messiness of tweets presents several challenges for standard information extraction methods. Most deployed systems employ approaches that rely on simple keyword matching and do not distinguish between relevant and irrelevant keyword mentions making them susceptible to false positives as a result of the fact that keyword volume can be influenced by several social phenomena that may be unrelated to disease occurrence. Furthermore, most solutions are intended for a single language and those meant for multilingual scenarios do not incorporate semantic context. In this paper we experimentally examine different approaches for classifying text for epidemiological surveillance on the social web in addition we offer a systematic comparison of the impact of different input representations on performance. Specifically we compare continuous representations against one-hot encoding for word-based, class-based (ontology-based) and subword units in the form of byte pair encodings. We also go on to establish the desirable performance characteristics for multi-lingual semantic filtering approaches and offer an in-depth discussion of the implications for end-to-end surveillance.


  • Open Access

    Beyond Raw Performance: Neural Machine Translation Based Multi-lingual Classification of Tweets for Automated Disease Surveillance

    by Mark Abraham Magumba, Peter Nabende

    Twitter and social media as a whole have great potential as a source of disease surveillance data however the general messiness of tweets presents several challenges for standard information extraction methods. Most deployed systems employ approaches that rely on simple keyword matching and do not distinguish between relevant and irrelevant keyword mentions making them susceptible to false positives as a result of the fact that keyword volume can be influenced by several social phenomena that may be unrelated to disease occurrence. Furthermore, most solutions are intended for a single language and those meant for multilingual scenarios do not incorporate semantic context. In this paper we experimentally examine a translation based approach that allows for incorporation of semantic context in multi-lingual disease surveillance in the social web.