December 30-31, 2026, Virtual Conference
Nabila Zine and Cherkaoui Leghris, Hassan II University of Casablanca, Morocco
The Internet of Things (IoT) has become a critical infrastructure exposed to an increasing diversity of cyberattacks, amplified by protocol heterogeneity, the resource constraints of embedded devices and the decentralisation of deployment architectures. Traditional reactive mechanisms, based on a posteriori detection and static signatures, no longer suffice against sophisticated, multi-stage and evolving threats. This paper reviews twenty contributions published between 2019 and 2026 on IoT security architectures and compares them along seven transversal criteria: target layers, main approach, proactivity, adaptivity, decentralisation, edge integration and compatibility with embedded constraints. The analysis shows a clear shift from static layered taxonomies towards intelligent architectures that combine deep learning, Zero Trust governance, decentralised trust mechanisms and, more recently, digital twins and large language models for threat anticipation. No existing architecture, however, simultaneously satisfies proactivity, adaptivity without full retraining, and deployability under embedded resource constraints.
Internet of Things, Security architecture, Intrusion detection, Zero Trust, Federated learning, Blockchain, Digital twin, Large language models, Edge computing.
Davit Kapanadze, Georgian National University SEU, Georgia and Institute of Mathematics and its Applications (IMA), United Kingdom
This paper proposes a cognitive-constructivist model for teaching the half-derivative, motivated by a student’s question: if we can speak of first, second and higher-order derivatives, why should there not be a derivative of order one half? The question is interpreted as an attempt to generalize a familiar operation; no empirical claims are made about its neural origins. The instructional model links, in a coherent sequence, prior knowledge, an operator hypothesis, the choice of a monomial space, a recurrence for the coefficients, the discovery of a free normalization parameter, and the derivation of the Gamma-function formula. The author’s previously developed mathematical results erve as the foundation; the independent contribution of the present paper is the instructional sequence built upon them, together with a set of diagnostic questions and a comparative system of problems. The Riemann–Liouville and Caputo definitions are compared on monomials: their agreement for positive exponents, their disagreement on constants, and the role of the domain of operator composition are demonstrated. An Abel equation, an initial value problem, a normalized heat-flux model and a derivative of order 3/2 are examined. The applied part is deliberately restricted to monomials and their finite linear combinations: a polynomial Abel equation and a Caputo initial value problem with a polynomial righthand side are considered, in which the constructive route reduces to finitely many algebraic operations. Finally, a dialogic AI protocol and a finite example in a Gamma-normalized basis are proposed. We claim neither a universal computational advantage nor an experimentally confirmed effectiveness of the model.
Half-Derivative; Graded Monomial Space; Fractional Differentiation; Constructivist Teaching; Apos; Cognitive Load; Artificial Intelligence
Virgil Ganescu and David Kaplan, USA
The rapid integration of Artificial Intelligence (AI) into higher education has generated both enthusiasm and skepticism among stakeholders. This study evaluates the perceived usefulness of six categories of AI-assisted learning tools through a comparative survey of students and faculty. Using a five-point Likert scale, respondents assessed AI functionalities ranging from theoretical summarization to step-by-step problem-solving and exam preparation. Results indicate a clear divergence between student and faculty perceptions, with students demonstrating moderate acceptance of AI tools—particularly those supporting procedural learning—while faculty responses reflect more cautious attitudes. The findings suggest that AI is most effective when deployed as a supplementary tool for applied learning rather than conceptual instruction. This study contributes to the growing body of research on AI in education by offering a structured comparison of specific AI functionalities and their perceived pedagogical value.
AI, Higher Education, Learning, Tools
Ayse Arslan, Oxford Alumni, USA
This paper examines how an engineering-led approach to information architecture, sprint planning, and quick-start flows can materially improve developer productivity and reliability outcomes. In the initial state, a fragmented and weakly structured knowledge base required AI developers to infer platform behaviour from tribal knowledge, ad hoc Slack threads, and outdated implementation guides, increasing onboarding time and error rates. This paper describes a re-design in which documentation is treated as a system with an explicit architecture—comprising Quick Start, Introduction, Environment Setup, First Use Case, Core Capabilities, Observability, and Special Topics—mirroring established systems-engineering practices for modularization and interface definition. By applying classic systems engineering principles to knowledge flows, the team transformed documentation from a passive artifact into an actively engineered component of the developer platform, yielding more predictable first-success paths, reduced support load, and a foundation for future automation and AI-assisted guidance.
Agentic AI; AI Engineering; AI Documentation; Information Architecture; Developer Experience; Responsible AI; AI Observability
Mohammad Ali, Atlantic International University, USA
Artificial neural networks (ANNs) promise to scale personalized instruction, early identification of at-risk learners, and formative feedback, yet their adoption in education raises technical, ethical, pedagogical, and institutional problems that are rarely evaluated together. This paper presents an integrated engineering study of ANNs for the education system. A five-layer reference architecture, the ANNbased Intelligent Education System (ANN-IES), is proposed whose assurance layer calibrates, explains, audits, and thresholds every model output before it reaches educators. Its components are formalized as a structural model of learner engagement and dropout, multilayer-perceptron and long short-term memory (LSTM) learners, knowledge tracing, mastery-based sequencing, calibration and fairness metrics, a decisiontheoretic intervention threshold, and an analytic hierarchy process (AHP) for prioritizing challenges. A reproducible simulation with 8,000 synthetic learners and a ten-skill tutoring environment shows that a deep ANN matches, but does not surpass, logistic regression and tree ensembles for mid-semester at-risk prediction (AUC 0.939 versus 0.942), whereas an LSTM deep-knowledge-tracing network implemented from first principles outperforms Bayesian knowledge tracing on sequential data (AUC 0.751 versus 0.706). The study further quantifies the accuracy–lead-time trade-off, the cold-start penalty of ANNs, a proxy-driven fairness gap and its mitigation, a cost-optimal alert threshold that matches theory, and the finding that more accurate knowledge tracing does not by itself improve closed-loop instruction. A consistent AHP ranks data privacy and algorithmic fairness as the most critical challenges. Opportunities, a 2026–2035 outlook, and recommendations conclude the paper.
Adaptive Learning, Analytic Hierarchy Process, Artificial Neural Networks, Deep Knowledge Tracing, Early-Warning Systems, Educational Data Mining, Explainable Artificial Intelligence, Fairness, Learning Analytics, Long Short-Term Memory