I work across learning systems, language models, climate informatics, and medical AI. The thread stays the same. Build methods that hold up in the real world, then put them to work on problems that count.
My PhD thesis at the University of Victoria. An adaptive learning method picks the best detection model for each satellite tile of kelp coastline worldwide, with the British Columbia coast as the case study. Labels arrive late and rarely, and conditions change with seasons and climate. The method keeps learning from sparse feedback and beats a single fixed model as the kelp shifts.
Learning automata that stay reliable when some nodes act maliciously. The work builds Byzantine detection and resistance into peer-to-peer content distribution, so the network keeps serving content under attack.
A language model for Ejagham, a low-resource Nigerian language. The goal is to preserve the language and give its speakers working text tools.
Land-cover segmentation from Landsat satellite imagery for Nigeria. The maps feed a Tier-3 greenhouse gas inventory for land use and forestry.
A diagnostic model for pigmented and non-pigmented skin lesions in ethnic skin types. It supports dermatology for communities with little specialist access.
aidermatology.orgGreenhouse gas accounting for the Nigerian waste sector. The analysis measures how circular economy actions change emissions, and supports Nigeria’s national climate targets.
Self-organizing lists that reorder as queries arrive. The research uses pursuit-based object partitioning to cut access cost on hierarchical list structures.