This week in computer science, energy, and health education, we digest 10 new open-access studies, many with African authors.
📊 This week at a glance
🌍 African-led research
Explainable AI (XAI) is moving from built-in transparency to large language models that explain decisions in plain language.
The review by Mumuni et al. in Array (2024) maps how XAI has evolved from inherently interpretable models to methods that use large language models to generate explanations. This changes the assumption that explainability must be designed into the model itself; now, post-hoc explanation via LLMs is a major direction. For African healthcare and other mission-critical sectors, this matters because trust and adoption of AI depend on stakeholders understanding decisions, not just accepting them.
Two conjectures about domination in cubic graphs are settled, lowering the girth condition from 83 to a smaller value.
Dorbec and Henning in the Journal of Graph Theory (2024) resolve the 13-conjectures, which asked whether the girth condition in a 2008 result on domination numbers could be reduced. This changes what graph theorists thought was the necessary girth for a cubic graph to have domination number at most 13. The result matters for African mathematics because it strengthens the global theoretical toolkit, and the authors are African-based, showing leadership in pure mathematics.
South African rehabilitation students report negative perceptions of neuroanatomy, a phenomenon known as neurophobia.
Human et al. in Anatomical Sciences Education (2024) surveyed undergraduate physiotherapy, occupational therapy, and speech-language pathology students in South Africa, finding that neurophobia is present in these groups. This changes the assumption that neurophobia mainly affects medical students; it also affects rehabilitation-focused disciplines. For African health education, the implication is that teaching methods for neuroanatomy may need to address anxiety and relevance to improve learning outcomes.
Aragonite from cuttlebone waste can be 3D-printed and shows promising structural and thermal properties.
Hamdan et al. in the Journal of Thermal Analysis and Calorimetry (2024) characterised aragonite (a form of calcium carbonate) extracted from cuttlebone waste and demonstrated its use in 3D printing. The abstract is empty, so details are limited, but the title indicates a waste-to-resource approach. This matters for African materials science because it turns abundant marine waste into a valuable biomaterial, with potential for bone scaffolds and sustainable manufacturing.
Integrating AI control, self-healing materials, and thermoelectric energy harvesting could make photocatalytic water treatment self-regenerating.
Zourif et al. in Applied Energy (2024) review how these three technologies can be combined to overcome catalyst deactivation, high energy use, and variable water quality in photocatalytic treatment. This changes the design paradigm from passive systems to autonomous, self-repairing ones. For African water treatment, where energy and maintenance are often constrained, such systems could improve resilience and reduce costs.
Flue gas recirculation in ammonia-hydrogen MILD combustion changes NO formation and improves stability.
Wu et al. in Combustion and Flame (2024) experimentally and kinetically investigated how flue gas recirculation affects NO emissions and flame stability in NH3-H2 MILD combustion. This changes the understanding of how to burn ammonia, a carbon-free fuel, while controlling nitrogen oxides. For African energy transitions, this matters because ammonia is a promising hydrogen carrier, and cleaner combustion supports decarbonisation goals.
Mapping thermodynamic phase boundaries can stabilise 3.5-valent vanadium electrolytes for flow batteries.
Sun et al. in the Journal of Power Sources (2024) used thermodynamic phase-boundary mapping to design initial electrolyte compositions that prevent thermal precipitation of VO2+ in vanadium flow batteries. This changes the approach from adding additives after preparation to designing the electrolyte upfront. For African renewable energy storage, this could lead to more durable and cost-effective batteries for grid-scale storage.
🔬 Global breakthroughs
M2SNet, a multi-scale subtraction network, improves medical image segmentation by reducing redundant information.
Zhao et al. in Machine Intelligence Research (2024) propose M2SNet, which uses subtraction instead of addition or concatenation to fuse features across scales, leading to sharper boundaries and better localisation. This changes the common U-shape decoder design that often blurs edges. For African healthcare, better segmentation means more accurate diagnosis from medical images, especially where expert radiologists are scarce.
Large language model-based intelligent agents are defined and classified, showing advanced reasoning, planning, and tool use.
Cheng et al. in Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery (2024) review the definitions, methods, and prospects of LLM-based agents that use natural language as a universal interface. This changes the view of LLMs from text generators to autonomous agents that can act. For African researchers, this opens opportunities to build local language agents for education, health, and public services.
MambaTS improves selective state space models for long-term time series forecasting.
Cai et al. in Pattern Recognition (2024) propose MambaTS, an improved selective state space model for forecasting long sequences. The abstract is empty, so details are limited, but the title indicates a methodological advance. For African data science, better time series forecasting can aid climate, health, and economic planning, though the lack of abstract means we cannot yet assess its specific gains.
All papers are open access. Explore more Technology & Engineering research on FRELIP · discover open scholarship at frelip.org and search 36,000+ open works at search.frelip.org. FRELIP — born in Nigeria, built for African scholarship, serving the world.