Research Output
Our research advances AI for agriculture, healthcare, and digital sovereignty across East Africa and the Global South.
2025-2026
When Synthetic Data Hurts: DCGAN Augmentation Failure Under Extreme Data Scarcity for Cassava Disease Classification in East Africa
DSA 2026 Workshop, Makerere University, Kampala, Uganda (23–24 July 2026)
Investigates failure modes of DCGAN-based data augmentation under extreme data scarcity in cassava disease classification, revealing conditions under which synthetic augmentation degrades model performance in East African field settings.
Auditing Representation in Crop Disease Benchmark Datasets Intended for Sub-Saharan African Deployment
DSA 2026 Workshop, Makerere University, Kampala, Uganda (23–24 July 2026) — Poster Presentation
Audits geographic, class, and demographic representation in widely-used crop disease benchmark datasets intended for Sub-Saharan African deployment, highlighting critical gaps that reduce model transferability in local field conditions.
Few-Shot Transfer Learning for Cassava Leaf Disease Detection Under East African Field Conditions
DSA 2026 Workshop, Makerere University, Kampala, Uganda (23–24 July 2026) — Poster Presentation
Evaluates few-shot transfer learning strategies for cassava leaf disease detection, specifically addressing the domain shift challenges arising from real-world East African field image conditions compared to curated benchmark data.
From Digital Consumerism to Strategic Sovereignty: Analyzing the Draft EAC Regional AI Strategy (2026)
AAIAC 2026 — 2nd International Conference on Advancements of AI in the African Context, Mlimani City Conference Centre, Dar es Salaam, Tanzania (22–23 July 2026)
Critically analyzes the Draft East African Community Regional AI Strategy through a digital sovereignty lens, identifying gaps between aspirational policy goals and structural constraints facing the region's AI ecosystem.
Leakage-Free Fraud Detection in Mobile-First Banking: A PR-AUC Benchmark of XGBoost, Random Forest, and Isolation Forest for East Africa
AAIAC 2026 — 2nd International Conference on Advancements of AI in the African Context, Mlimani City Conference Centre, Dar es Salaam, Tanzania (22–23 July 2026)
Establishes a rigorous leakage-free PR-AUC benchmarking methodology for mobile-first banking fraud detection in East Africa, comparing XGBoost, Random Forest, and Isolation Forest under realistic data imbalance conditions.
A Physics-Informed Random Forest Framework for Multi-Decadal Solar Irradiance Forecasting in Tanzanian Off-Grid Mini-Grids
AAIAC 2026 — 2nd International Conference on Advancements of AI in the African Context, Mlimani City Conference Centre, Dar es Salaam, Tanzania (22–23 July 2026)
Develops a physics-informed random forest model for long-horizon solar irradiance forecasting tailored to Tanzanian off-grid mini-grid deployments, integrating domain physical constraints to improve reliability under data-scarce conditions.
Resilient Cyber Defence for African Military 5G Networks: A Multi-Stage Hybrid ML Framework for Zero-Day Detection
AAIAC 2026 — 2nd International Conference on Advancements of AI in the African Context, Mlimani City Conference Centre, Dar es Salaam, Tanzania (22–23 July 2026)
Proposes a multi-stage hybrid machine learning framework for zero-day threat detection in African military 5G network environments, combining anomaly detection with signature-based classifiers for resilient cyber defence.
Predicting Insurgency Hotspots in the Sahel: Leveraging GTD Data for Regional Defence Resource Allocation
AAIAC 2026 — 2nd International Conference on Advancements of AI in the African Context, Mlimani City Conference Centre, Dar es Salaam, Tanzania (22–23 July 2026)
Uses Global Terrorism Database (GTD) data with machine learning models to predict insurgency hotspots across the Sahel region, enabling data-driven regional defence resource allocation for conflict prevention.
CLOVES-4603: Benchmarking Classical Texture Features and Fine-Tuned Deep Models for Clove Quality Grading
7th Workshop on Vision for Agriculture (CV4A) @ CVPR 2026
Introduces the CLOVES-4603 dataset and benchmarks classical texture features (LBP, GLCM, HOG) against fine-tuned deep learning models for automated clove quality grading, addressing a critical post-harvest challenge for Zanzibar's primary export crop.
Unifying Perspectives on Learning Biases: A Data-Centric Intervention for Holistic Fairness, Robustness, and Generalization
ICLR 2026 Workshop on Principled Design for Trustworthy AI-Interpretability
Proposes a unified data-centric framework that simultaneously addresses fairness, robustness, and generalization through principled bias interventions, bridging previously siloed perspectives in trustworthy ML research.
AI Sovereignty in the Global South: Power, Dependency, and Strategic Futures
AAAI 2026 Conference - Workshop Empowering Global South AI
Analyzes structural power asymmetries in AI development across the Global South, proposing multipolar policy frameworks involving the African Union, ASEAN, and South-South cooperation to counter digital dependency on foreign platforms.
Digital Sovereignty or Digital Serfdom? AI Strategies, Geopolitics, and the Quest for an Equitable Global South
PAIG Workshop @ EurIPS 2025 (Beyond Regulation: Governance of AI in the Real World)
Examines whether Global South countries are advancing toward digital sovereignty or drifting into digital serfdom. Uses a seven-dimensional framework and four case studies (Singapore, Vietnam, Kenya, and Syria) to reveal asymmetries in computing, talent, and regulation.
BAFNet: Deep contour-aware features for colorectal polyps segmentation
Expert Systems with Applications, October 2025
Introduces a boundary-aware feature fusion neural network for improved colorectal polyp segmentation through gating mechanisms that select high-quality contour features and enhance semantic accuracy.
CARA-FMs: A Framework for Context-Aware and Responsible AI in East Africa
Presented theoretical framework for culturally attuned foundation models for East African agriculture. Winner, Best Poster Award (Publication & Dataset Category).
Towards Culturally Attuned and Resource-Aware Foundation Models for East African Agriculture: A theoretical framework and research agenda
Deep Learning Indaba (DLI) 2025 Conference - Responsible AI Workshop
Oral presentation on theoretical framework for culturally attuned foundation models for East African agriculture. To be published in PMLR Conference Proceedings (Tentatively January 2026).
Prime Directives for Responsible AI for Africa: A Manifesto for Inclusive Technology
Data Science Africa (DSA) 2025, Ibadan University, Nigeria
We present a holistic framework for developing inclusive, ethical, and sustainable AI that addresses Africa's unique cultural, social, and economic contexts. We emphasize participatory design, data equity, and local empowerment to ensure AI benefits all African communities.
Rectifying the extremely weakened signals for cassava leaf disease detection
Computers and Electronics in Agriculture, Volume 232, 2025
Developed an artificial intelligence approach to accurately detect cassava leaf diseases by enhancing weak signal patterns in agricultural imaging.
From Art to Algorithms: Co-Designing AI for Clove Grading with Zanzibar's Indigenous Experts
Proposes a socio-technical approach to modernizing Zanzibar's clove grading system. Validates a segmentation-first deep learning architecture, achieving 99.0% accuracy, while advocating for co-design with Indigenous experts to develop a low-cost mobile tool that fosters transparency and trust.
Computing at the Edge of Possibility: Building Agricultural AI for Resource Constrained East Africa
NPTEL CRC 302 Studio, IIT Madras Chennai
Faculty talk on developing agricultural AI solutions for resource-constrained environments in East Africa.
How can Innovation Modernize the Public Service Delivery and Create Inclusive Digital Economy
4th ZRCP Joint Annual Conference, Zanzibar
A talk on innovation's role in modernizing public service delivery and fostering an inclusive digital economy.