Throwing bones

by siobhan | Apr 30, 2025 | Sauti Yetu

With a good enough model, could we throw the bones and predict the impact of the Deep Learning Indaba on ourselves, and on the continent?

Rarely, with the benefit of hindsight, there is a moment that stands out as wildly impactful. For me, attending the Deep Learning Indaba (DLI) in 2022 was one of those milestones. A year into a tough PhD programme overseas with a heavy theoretical machine learning course load, I was struggling with self-doubt and imposter syndrome. The Indaba in Tunisia was my first after five years, yet it felt like I belonged. The DLI re-rooted my commitment to Africa and my goal of giving back to our community.

I like to think of my Indaba re-rooting using the analogy of “throwing bones”, which has three meanings for me: the bones that form the core knowledge skeleton; the bones of prediction; and the bones we use to help others. First, as a medical doctor, the skeleton is what holds the body together: the fundamentals. Second, as a theoretical neuroscience and machine learning researcher, the throwing of the bones (or amathambo) for divination by Zulu and Xhosa traditional healers (sangomas), speaks strongly to the problem of prediction. And third, the act of throwing someone a bone – offering a small piece of help – is a meaning of amathambo that has become clear to me in reflection.

The company I co-created out of the 2022 Indaba’s Ideathon, Amathambo AI, has put all three of these concepts together in our quest to predict patient load and manage staff rostering in healthcare. The beautiful name was suggested by my fellow conspirators and founders, two long-time Indaba stalwarts and friends, Simphiwe Zitha and Sicelukwanda Zwane. Our journey embodies the three amathambo principles:

  1. They have thrown me many bones! Being able to trust and lean on one’s team is essential.
  2. We have been cautious to stick to the skeleton in the product we are building – we have kept the tech simple, because we know that we are solving an important problem.
  3. We have been mindful to not get too caught up in our researcher-brains: to take time to predict not just patient load data, but also our future and our business’s future.

While these sound straightforward, I am reminded that there would be no Amathambo AI without the Indaba community, who have thrown me so, so many bones. From organisers and keynotes giving up time, to sharing a lift, to eating together, to exchanging both cultural and scientific ideas, to enabling internship opportunities, to being able to access sustained mentorship and advice – these small puzzle pieces have played a huge role in growing me as a researcher and participant in the African machine learning community. Importantly, I think I owe a large part of my journey to simply being at the Indaba in 2022.

“The family is like the forest: if you are outside it is dense; if you are inside you see that each tree has its own position.”

– Akan Proverb, shared by Yaa Gyasi in Homegoing, an excellent book contrasting the author’s birthplace in Ghana to modern USA across generations of slavery.

There’s something special about the rootedness and connection of going home, and it's how I felt when I left Tunisia. “Homegoing” to the Indaba gets better with every mindful interaction. Choosing to be involved, to put the community’s interests above one’s own, to be present: these are how we continue to grow our space, and throw bones of knowledge and support to one another.

I encourage everyone to get involved in the Indaba. This does not need to look like leaping to join the main organising committee from the get-go. For me, it means reflecting on your path and finding the next meaningful step (see James Allingham’s suggestions in last month’s edition), including those that also help another member of the community. I started by simply attending – joining an Ideathon team sharing my interests – then slowly realised where I could make more and more impact.

So, throw someone a bone, be sure to build and nourish your own knowledge skeleton, and may the bones continue to point towards an AI future with strong, interconnected roots across the continent.

- Kira Düsterwald. 2025 Programme Chair (Keynotes), 2024 Programme Chair (Practicals), multi-year practicals developer and tutor, and long-time attendee. April 2025.

Cheikh Anta Diop Award

The Cheikh Anta Diop Award recognises, encourages, and celebrates excellence in research, teaching, and community service by early- to mid-career academics and researchers at African universities in any area of artificial intelligence, as well as computational and statistical sciences. Its recipients are those who uphold Cheikh Anta Diop’s legacy as a multidisciplinary scientist and visionary intellectual.

The DLI 2026 Cheikh Anta Diop Award is awarded to:

Winner: Prof Omneya Attallah 

Country: Egypt

Affiliation: Arab Academy for Science, Technology and Maritime Transport (AASTMT)

Essence of the Award: Prof. Omneya Attallah says that this award reminds her that every late night, sacrifice, and challenge she has faced as a working mother and African researcher has been worthwhile. It proves that perseverance truly pays off.

Professor Omneya Attallah is recognised for a career that integrates research excellence, teaching, and community service into a single mission. At AASTMT, she has built a research programme spanning breast cancer detection, paediatric epilepsy diagnosis, wearable biosensing, explainable AI, and CanSense, a breathomics AI system for non-invasive breast cancer screening developed on locally collected Egyptian clinical data in partnership with Egyptian clinicians. As a teacher she designed the curriculum for a new undergraduate Biomedical Engineering programme. She has supervised 13 postgraduate researchers to completion, the majority of them women. Through HERBioLab, WeBios, and her IEEE and editorial activities, she has built laboratories, curricula, training pathways, and partnerships that are shaping the next generation of African biomedical AI researchers

Wangari Maathai Impact Award

The Wangari Maathai Impact Award recognises and celebrates work by African innovators, thinkers, and advocates who demonstrate impactful work, including, but not limited to, technical, societal, environmental, and economic domains, around machine learning and artificial intelligence. This award reinforces the legacy of Wangari Maathai by acknowledging the capacity of individuals to be a positive force for change: by recognising ideas and initiatives that demonstrate that each of us, no matter how small, can make a difference.

The DLI 2026 Wangari Maathai Impact Award is awarded to:

Winner: Dr Hellina Hailu Nigatu

Country: Ethiopia

Affiliation: Artificial Intelligence Accountability Lab (AIAL) at Trinity College Dublin

Essence of the Award: Dr Hellina Hailu Nigatu says that having read Wangari Maathai's autobiography, Unbowed, it is a personal privilege to be recognized with an award in her name. She wants to give gratitude to her mentees on the project who carried the majority of the work. This recognition is as much, if not more, a fruit of their effort and hard work. She hopes this will also inspire more grassroots ventures in the African AI landscape that foster meaningful community engagement and participation.

Dr Hellina Hailu Nigatu is recognised for her work addressing gender bias in machine translation across three Ethiopian languages: Amharic, Afan Oromo, and Tigrinya. Rather than adapting English benchmarks, she built evaluation frameworks and datasets directly within African languages, preserving cultural nuances that translation from English cannot capture. Key to this work is the Yeswa Stories dataset, centered on African women's narratives and lived experiences. Seven Ethiopian female students participated as paid researchers, benefiting from funding, compute access, and substantive mentorship throughout. The project produced open-release benchmark datasets now available to the broader African NLP community. Her mentorship has been described by those she supervised as transformative, not nominal.

The DLI 2026 Wangari Maathai Impact Award runner-up is: 

Runner-up: Abdel-aziz Harane Abounounou 

Country: Chad

Affiliation: Chad AI Network

Essence of the Award: Abdel-aziz Harane says that he is deeply honored by this award. This recognition reflects his constant obsession: solving Chad's real problems through AI and software engineering. It renews his commitment to the Chad AI Network's mission of sustainable, locally driven transformation.

Abdel-aziz Harane Abounounou is recognised for his work building the first natural language processing tools for Chad's indigenous languages, a country with 123 languages yet only two represented in existing digital tools. Rather than waiting for a foreign lab to take interest, he set out to build the entire pipeline himself, working directly with the communities who speak these languages. The Kalam-na project anchors this work, collecting audio and text data across 16 dialects of the Sara language family. Native speakers contribute their own voices through Kalam-na Voice, a platform built to remove technical barriers to participation. The project has gathered over 68,000 transcribed audio clips and 72,000 aligned texts, laying the groundwork for open ASR, TTS, and MT models. He also founded the Chad AI Network, now over 200 members strong, and ran two editions of IndabaX Chad. This work reaches an estimated 4 million speakers of Chadian languages currently shut out of the digital world.

Thamsanqa Kambule Doctoral Award

The Thamsanqa Kambule Doctoral Award recognises, encourages, and celebrates excellence in research and writing by doctoral candidates at African universities in any area of computational and statistical sciences. Its recipients are those who uphold Thamsanqa Kambule’s legacy as a defender of learning, a seeker of knowledge, and an activist for equality.

The DLI 2026 Thamsanqa Kambule Doctoral Award is awarded to:

Winner: Dr Everlyn Chimoto 

Country: Kenya

Affiliation: Lelapa AI & Riara University

Essence of the Award: “Receiving this award is deeply meaningful because it recognises not only the years of work invested in this thesis, but also the support of my supervisor, collaborators, family, and the African NLP community that has continually inspired and challenged me. To me, it also recognises the importance of pursuing research motivated by the needs of underrepresented language communities. I believe this work demonstrates that data-centric approaches can make meaningful progress in low-resource settings while remaining relevant to many other machine learning problems. I am encouraged to continue pursuing research that advances knowledge while creating opportunities for more languages and more people to participate in language technology,” says Dr Everlyn Chimoto.

Dr Everlyn Chimoto is recognised for a doctoral thesis on sentence alignment for the Marama dialect of Luhya, CAT data pruning which achieves 92% of full-data performance at 50% of the data with no additional compute cost, GrammaMT which delivers over 12 BLEU point improvements for endangered languages from as few as 21 grammatical examples, and COMET-QE active learning for efficient data selection. Together these form a framework immediately usable by African NLP practitioners working under real compute and data constraints. Her work is centred on improving model performance to make AI systems more inclusive for African language communities.

The DLI 2026 Thamsanqa Kambule Doctoral Award runner-up is:

Runner-up: Dr Devon Jarvis 

Country: South Africa

Affiliation: University of the Witwatersrand

Essence of the Award: “I started learning ML while volunteering at the Indaba in 2017. To be recognised nine years later is gratifying and reflects the impact this community has had on my career,” says Dr Devon Jarvis.

Dr Devon Jarvis is recognised for his doctoral work developing a theoretical framework for controlled semantic cognition using deep and nonlinear neural networks to explain how the human brain flexibly applies learned semantic concepts to new contexts. Rather than treating machine learning purely as an engineering tool, he built tractable theoretical models that shed light on the principles of human cognition itself. His dissertation extends deep linear network theory to explain systematic generalisation, introduces a novel framework (the Rectified Linear Network) to model ReLU network training dynamics, and shows that these networks can capture all six classical properties of controlled semantic cognition while still falling short of systematic generalisation. He then proposes a meta-learning model, the Meta-GDLN, as a path toward closing that gap. He has since remained on the continent to establish the CAandL Lab, one of Africa's first computational neuroscience labs.

Grace Alele-Williams Masters Award

The Grace Alele-Williams Masters Award recognises and celebrates excellence in research and writing by master's candidates at African universities in any area of computational and statistical sciences. Its recipients are those who uphold Grace Alele-Williams’ legacy as a defender of learning, champion of academic excellence, and activist for access to education.

The DLI 2026 Grace Alele-Williams Masters Award is awarded to:

Winner: Akinbobola Adegboyega

Country: Nigeria

Affiliation: Olabisi Onabanjo University

Essence of the Award: According to Akinbobola Adegboyega, “The Alele-Williams Award is a reminder that hard work, curiosity, and dedication matter, and it inspires me to keep making meaningful contributions for societal benefit.”

Akinbobola Adegboyega is recognised for a Master's thesis that introduces affective fidelity as a new evaluation lens for machine translation in low-resource tonal languages, reframing emotion as a core measure of translation faithfulness rather than a secondary concern. His work targets Yorùbá, a language representing just 0.008% of Common Crawl training data. He produced a culturally validated bilingual emotion dataset of over 200 annotated paragraph pairs verified by native speakers, created a reproducible benchmarking framework evaluating four major language models, and identified six distinct failure modes specific to tonal low-resource language translation. The dataset has been released openly, contributing usable infrastructure to the African NLP community.

The DLI 2026 Grace Alele-Williams Masters Award runner-up is:

Runner-up: Imen Habibi 

Country: Tunisia

Affiliation: LARIA, ENSI, University of Manouba, Tunisia

Essence of the Award: “This recognition is the reward for five years of dedication and perseverance. It honors my supervisors' support and inspires me to continue pursuing research with confidence,” says Imen Habibi.

Imen Habibi is recognised for her master's thesis designing a drone-based inspection system that integrates optimized UAV path planning with AI-powered imaging for industrial applications. She developed spiral and zigzag trajectory strategies for nuclear facility inspection (validated through 2D and 3D simulations) alongside a CNN-based anomaly detection pipeline for PV solar panels achieving 91% accuracy. Her work bridges trajectory optimization and intelligent image analysis into a unified system applied to real industrial use cases, improving inspection reliability, precision, and efficiency.