Forecasting stock prices remains a central challenge in financial modelling, as markets are influenced by market sentiment, firm-level fundamentals and complex interactions between macroeconomic and microeconomic factors, for example. This study evaluates the predictive performance of both classical statistical models and advanced attention-based deep learning architectures for daily stock price forecasting. Using a dataset of major U.S. equities and Exchange Traded Funds (ETFs) covering 2012–2024, we compare traditional statistical approaches, Seasonal Autoregressive Integrated Moving Average (SARIMA) and Exponential Smoothing (ES) in the Error, Trend, Seasonal (ETS) framework, with deep learning architectures such as the Temporal Fusion Transformer (TFT), and a novel hybrid model, the TFT-Graph Neural Network (TFT-GNN), which incorporates relational information between assets. All models are assessed under consistent experimental conditions in terms of forecast accuracy, computational efficiency, and interpretability. Our results indicate that while statistical models offer strong baselines with high stability and low computational cost, the TFT outperforms them in capturing short-term nonlinear dependencies. The hybrid TFT-GNN achieves the highest overall predictive accuracy, demonstrating that relational signals derived from inter-asset connections provide meaningful enhancements beyond traditional temporal and technical indicators. These findings highlight the advantages of integrating relational learning into temporal forecasting frameworks and emphasise the continued relevance of statistical models as interpretable and efficient benchmarks for evaluating deep learning approaches in high-frequency financial prediction.

The search for edge-disjoint Hamilton cycles in star graphs is important for the design of interconnection network topologies. We define automorphisms for star graphs St _n of degree n−1, for every positive odd integer n, which yield permutations of labels for the edges of St _n taken from the set of integers between 1 and ⌊ n/2 ⌋. By decomposing these permutations into permutation cycles, we are able to identify edge-disjoint Hamilton cycles that are automorphic images of a known Hamilton cycle in St _n . Our method produces a better than two-fold improvement from ⌊ ϕ (n)/10 ⌋ to ⌊ 2ϕ (n)/9 ⌋, where ϕ is the Euler function, for the known number of edge-disjoint Hamilton cycles in St _n for all odd integers n. For prime n, the improvement is from ⌊ n/8 ⌋ to ⌊ n/5 ⌋, and we can extend this result to the case when n is the power of a prime greater than 7.

Conversational agents in universities are mostly screen-based and lack the presence needed for confident, back-and-forth teaching. We introduce life-sized, AI-powered Holographic Conversational Agents (HCAs) that run on Proto displays and portray Ada Lovelace and Averroes for use in education and public engagement. The paper makes three contributions. First, it sets out a documented, reproducible pipeline that takes still or video assets to an interactive, source-grounded, safety-checked avatar suitable for live education. Second, it details an EDI-informed persona design, pairing a woman in computing with a non-Western scholar, with deliberate choices about voice, accent and knowledge sources to support inclusive engagement. Third, it reports a mixed-methods evaluation with staff and students in real settings, showing strong perceived usefulness, engagement and social presence, with clear intent to reuse in classes and exhibitions. Results confirmed high perceived value, across 51 participants, 85.7\% of all ratings were positive; Engagement averaged 4.31/5 (85.3\% positive) and recommendation 4.37/5 with 88.2\% rating 4–5. We close with practical guidance for building new subject-specific HCAs from the pipeline and deployment notes for classrooms, labs and galleries, positioning life-sized holography as a viable format for authoritative, dialogic teaching with historical figures.

Artificial Intelligence (AI) is rapidly changing higher education, offering new opportunities to support diverse learners, including students with dyslexia, while raising important questions around ethics and pedagogy. This study introduces a novel, research-informed framework that supports students with dyslexia in reading, writing, and academic workflow planning, grounded in the principles of Universal Design for Learning (UDL) and constructive alignment. Using mixed methods, including student surveys and interviews with the Student Support and Wellbeing Services (SSWS) team, we identify key barriers such as information overload, linear writing demands, and uncertainty around AI use. The framework offers a step-by-step guide for responsible AI use aligned with UDL's principles of multiple means of engagement, representation, and expression. This paper’s primary contribution is a novel, dual-component pedagogical system designed to resolve this uncertainty. It comprises an academic-facing model, \textit{the AI Engagement Spectrum}, which shifts responsibility for ethical AI use from student policing to proactive instructional design, and a student-facing workflow, the \textit{AI Scaffold} that applies UDL principles for AI-supported academic tasks. The framework provides a process that embeds ethical checkpoints designed to support AI literacy, ensure human-in-the-loop agency, and mitigate the risks of uncritical cognitive offloading. Our findings suggest that with the right design, AI can become a practical and equitable support strategy, enhancing autonomy while addressing concerns like skill atrophy and academic integrity.

Forecasting stock prices remains a central challenge in financial modelling, as markets are influenced by complex interactions between macroeconomic and microeconomic factors, firm-level fundamentals, and market sentiment. This study evaluates the predictive performance of classical statistical models and advanced attention-based deep learning architectures for daily stock price forecasting. Using a dataset of major U.S. equities and Exchange Traded Funds covering 2012–2024, we compare traditional statistical approaches, Seasonal Autoregressive Integrated Moving Average (SARIMA) and Exponential Smoothing, with deep learning architectures such as the Temporal Fusion Transformer (TFT), and the novel TFT-Graph Neural Network (TFT-GNN) hybrid that incorporates relational information between assets. All models are assessed under consistent experimental conditions in terms of forecast accuracy, computational efficiency, and interpretability. Results indicate that whilst statistical models offer strong baselines with high stability and low computational cost, the TFT outperforms them in capturing short-term nonlinear dependencies. The hybrid TFT-GNN achieves the highest overall predictive accuracy, demonstrating that relational signals derived from inter-asset connections provide meaningful enhancements beyond traditional temporal and technical indicators. These findings underscore the benefits of integrating relational learning into temporal forecasting frameworks and highlight the continued relevance of statistical models as interpretable, efficient benchmarks for evaluating deep learning approaches in high-frequency financial prediction.