This week’s digest covers artificial intelligence, graph theory, health professions education, materials science, energy and combustion engineering, and time-series forecasting, drawing on studies from Array, Journal of Graph Theory, Anatomical Sciences Education, Journal of Thermal Analysis and Calorimetry, Applied Energy, Combustion and Flame, Journal of Power Sources, Machine Intelligence Research, Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, and Pattern Recognition.
📊 This week at a glance
🌍 African-led research
Explainable AI (XAI) — systems that show how they reach decisions — is mapped from built-in transparency to large language models, with trust in high-stakes fields like healthcare as the central problem.
Mumuni et al. survey the field in Array (2024), organising XAI methods from inherently interpretable models to explanations for large language models (LLMs). The practical implication is that opaque decision logic remains a barrier to adopting machine learning in mission-critical domains such as healthcare, where clinicians and regulators need to understand outputs. For African scholarship, this matters because explainability is a prerequisite for locally deploying AI in health systems where accountability and trust are essential.
A 2008 result on domination in cubic graphs — where every vertex is either in a chosen set or adjacent to one — required girth (shortest cycle length) of at least 83; the 13-conjectures ask whether that girth condition can be lowered.
Dorbec and Henning, in the Journal of Graph Theory, address the natural question of whether the girth bound of 83 can be reduced, giving rise to the 13-conjectures. This changes the frontier from a single high-girth theorem to a structured set of conjectures that guide where smaller girth might still permit the same domination bound. For African mathematics, it signals active contribution to foundational graph theory with implications for network design and combinatorial optimisation.
South African undergraduate rehabilitation students report negative perceptions of neuroanatomy — the phenomenon called neurophobia — with limited prior data for physiotherapy, occupational therapy, and speech-language pathology.
Human et al. conducted a cross-sectional study in Anatomical Sciences Education (2024) to describe perceptions among these rehabilitation-focused health professional students in South Africa. The finding confirms that neurophobia extends beyond medicine into rehabilitation disciplines, implying that teaching approaches may need targeted support to reduce anxiety and improve learning. For African health professions education, it provides locally generated evidence to inform curriculum design where such data were previously scarce.
Aragonite (a form of calcium carbonate, CaCO3) derived from cuttlebone waste was characterised for structural-thermal behaviour and 3D printing, turning a biological discard into a printable material.
Ait Hamdan et al. report in the Journal of Thermal Analysis and Calorimetry on the physicochemical characterisation and 3D printing of aragonite from cuttlebone waste. The abstract is not available, so detail is limited to the title, which indicates a waste-to-material route for a common marine by-product. For African materials science, it points to low-cost, locally available feedstocks for additive manufacturing and thermal applications.
A review proposes integrating cognitive AI control, bio-inspired self-healing materials, and thermoelectric energy harvesting to make photocatalytic water treatment systems more autonomous and resource-efficient.
Zourif et al. examine this integration in Applied Energy (2024), targeting persistent limitations of catalyst deactivation, energy consumption, and adaptability to variable influent quality. The implication is a framework for self-regenerating treatment systems that could reduce downtime and energy demand. For African water treatment, where influent quality and power reliability vary, such autonomous, energy-harvesting designs could improve resilience.
In ammonia-hydrogen MILD combustion (a low-oxygen dilution technique), flue gas recirculation chemically influences NO formation and stability by affecting ammonia cracking to hydrogen.
Wu et al. report experimental and kinetic findings in Combustion and Flame (2024) on the kinetic interactions in NH3-H2 MILD combustion. This changes understanding of how recirculated flue gas chemistry, not just dilution, shapes emissions and flame stability. For African energy transitions, ammonia as a hydrogen carrier could enable carbon-neutral systems, making control of NO emissions critical.
Mapping thermodynamic phase boundaries allows stabilisation of 3.5-valent vanadium electrolytes, addressing the poor thermal stability of VO2+ that hinders vanadium flow batteries.
Sun et al. report in the Journal of Power Sources (2024) a design approach based on the thermodynamic properties of the initial electrolyte composition, rather than adding stabilisers after preparation. This changes the strategy from post-hoc additives to upfront phase-boundary mapping, with implications for longer-lasting large-scale energy storage. For African grid storage, more stable vanadium flow batteries could support renewable integration where dispatchable storage is needed.
🔬 Global breakthroughs
M2SNet, a multi-scale subtraction network, improves medical image segmentation by reducing redundant information from standard addition or concatenation feature fusion.
Zhao et al. present M2SNet in Machine Intelligence Research (2024), targeting inaccurate localisation and blurred edges caused by redundant features in U-shaped decoders. The implication is more precise segmentation for early medical diagnosis. For African healthcare, better segmentation tools can support diagnosis where specialist radiology access is limited.
Large language model (LLM)-based intelligent agents now exhibit advanced reasoning, planning, and tool use, with natural language as a universal interface for cognition and interaction.
Cheng et al. review definitions, methods, and prospects in Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery (2024). This changes the agent paradigm from narrow, hand-coded systems to LLM-driven agents that generalise across domains. For African AI research, it highlights an accessible entry point via natural language interfaces, though infrastructure and language coverage remain considerations.
MambaTS improves selective state space models for long-term time series forecasting, addressing limitations in capturing long-range dependencies.
Cai et al. present MambaTS in Pattern Recognition (2024). The abstract is not available, so detail is limited to the title, which indicates an architectural improvement for forecasting over extended horizons. For African data science, better long-term forecasting supports applications in climate, health, and economics where planning depends on reliable projections.
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.