Effective feedback is a central component of successful student learning, with extensive research examining how best to implement it in educational settings. Increasingly, feedback is being generated by artificial intelligence (AI), offering scalable and adaptive responses. Two widely studied approaches are directive feedback, which gives explicit explanations and reduces cognitive load to speed up learning, and metacognitive feedback which prompts learners to reflect, track their progress, and develop self-regulated learning (SRL) skills. While both approaches have clear theoretical advantages, their comparative effects on engagement, confidence, and quality of work remain underexplored. This study presents a semester-long randomised controlled trial with 329 students in an introductory design and programming course using an adaptive educational platform. Participants were assigned to receive directive, metacognitive, or hybrid AI-generated feedback that blended elements of both directive and metacognitive feedback. Results showed that revision behaviour differed across feedback conditions, with Hybrid prompting the most revisions compared to Directive and Metacognitive. Confidence ratings were uniformly high, and resource quality outcomes were comparable across conditions. These findings highlight the promise of AI in delivering feedback that balances clarity with reflection. Hybrid approaches, in particular, illustrate how AI-generated feedback can be structured to support both clarity and reflection, although more work is required to evaluate their broader impact.
We study the computational complexity of exact cardinality-constrained minimum Riesz s-energy subset selection in finite metric spaces: given n points, select k<n points of minimum Riesz s-energy. The objective sums inverse-power pair interactions and therefore promotes well-separated subsets; as s becomes large, it increasingly approaches a bottleneck criterion governed by the closest selected pair, linking it to minimum pairwise distance (MPD). Building on the general-metric NP-hardness result of Pereverdieva et al. (2025), we prove that NP-hardness persists for point sets in the Euclidean plane when s is part of the input. In contrast, finite ultrametric spaces form an exact tractable regime: on rooted binary ultrametric trees with n leaves, an optimal size-k subset can be computed by dynamic programming in O(nk^2) time. We also discuss the ordered one-dimensional Euclidean case, where the classical MPD objective admits simple dynamic programming, but the additive Riesz energy does not appear to allow the same state compression. Finally, we explain why one natural route to fixed-s Euclidean hardness does not close: Fowler-style 3SAT gadgets, together with zeta-function bounds for far-field interactions, show why this approach still requires an exponent depending on k. Together, these results provide a compact complexity landscape for a natural diversity or dispersion objective, distinguishing Euclidean hardness, ultrametric tractability, and the ordered one-dimensional case.
Abstract Higher education institutions worldwide face mounting pressure to implement outcome-based education (OBE) at scale while ensuring quality and accountability. Yet OBE often remains resource-intensive, and blended learning implementations frequently lack systematic outcome alignment. Meanwhile, artificial intelligence (AI) applications in education are usually isolated tools (e.g., auto-grading, chatbots) rather than integrated components of curriculum governance. This paper proposes an integrated OBE–AI enhanced blended learning framework, operationalized through the intelligent blended learning system (iBLS). The architecture consists of five layers: (1) OBE curriculum design, (2) granular outcome decomposition, (3) constructive alignment, (4) AI-enhanced blended learning, and (5) continuous quality improvement (CQI). Two AI subsystems are embedded: TutorAI, an assessment engine for question generation, auto-grading, and analytics dashboards; and BotAI, a RAG-based conversational engine providing student support through institutional knowledge retrieval. Both subsystems share a common AI/NLP technological core but are tailored to distinct educational functions. A case study at Thai Nguyen University of Information and Communication Technology, Vietnam (ICTU) implemented the framework across five courses with approximately 2,500 students. Evaluation combined baseline comparison and simulated scalability analysis. Results show that improved outcome transparency (78% of students reported clear weekly targets vs. 42% baseline), stronger alignment ratings (4.3/5 vs. 3.1/5), higher efficiency in assessment (25% reduction in grading workload, 82% item acceptance), enhanced student support (91% query coverage, 88% accuracy), and actionable CQI feedback with 14 course learning outcomes (CLOs) revisions, 12% program learning outcomes (PLOs) updates. Overall pass rates increased from approximately 60% to approximately 80%. The study demonstrates that AI can support the scalability and sustainability of OBE in resource-constrained contexts, bridging theoretical accreditation frameworks and practical implementation. Implications are drawn for institutions, policymakers, and system designers, with future research directions addressing scalability, AI robustness, human adoption, and ethical considerations.
Cardiac arrhythmias are a leading cause of life-threatening cardiac events, highlighting the urgent need for accurate and timely detection. Electrocardiography (ECG) remains the clinical gold standard for arrhythmia diagnosis; however, manual interpretation is time-consuming, dependent on clinical expertise, and prone to human error. Although deep learning has advanced automated ECG analysis, many existing models abstract away the signal's intrinsic temporal and morphological features, lack interpretability, and are computationally intensive-hindering their deployment on resource-constrained platforms. In this work, we propose two novel lightweight 1D convolutional neural networks, ArrhythmiNet V1 and V2, optimized for efficient, real-time arrhythmia classification on edge devices. Inspired by MobileNet's depthwise separable convolutional design, these models maintain memory footprints of just 302.18 KB and 157.76 KB, respectively, while achieving classification accuracies of 0.99 (V1) and 0.98 (V2) on the MIT-BIH Arrhythmia Dataset across five classes: Normal Sinus Rhythm, Left Bundle Branch Block, Right Bundle Branch Block, Atrial Premature Contraction, and Premature Ventricular Contraction. In order to ensure clinical transparency and relevance, we integrate Shapley Additive Explanations and Gradient-weighted Class Activation Mapping, enabling both local and global interpretability. These techniques highlight physiologically meaningful patterns such as the QRS complex and T-wave that contribute to the model's predictions. We also discuss performance-efficiency trade-offs and address current limitations related to dataset diversity and generalizability. Overall, our findings demonstrate the feasibility of combining interpretability, predictive accuracy, and computational efficiency in practical, wearable, and embedded ECG monitoring systems.
Feature selection (FS) is a crucial pre-processing step that is frequently applied in a variety of domains to enhance the efficiency of learning algorithms. Numerous features are utilized in medical data mining to diagnose illnesses; however, many of these features have redundant characteristics and weak connections that are irrelevant, which leads to several issues that negatively impact diagnostic forecasting accuracy. Due to the growing need for techniques that may decrease the multi-dimensionality of data by choosing the optimal selection of components based on predetermined standards required for optimized accuracy of predictions and minimized irrelevant characteristics, work on FS has expanded significantly across several domains. To try and find a solution that is nearly optimal within a limited amount of time, metaheuristics have recently been chosen instead of traditional optimization applications regarding FS problems, as they produce solutions that are appropriate and time-effective. Metaheuristics are optimization methods that are considered algorithms for general use that may be utilized in solving nearly any optimization problem. This is particularly helpful when trying to solve complex problems. By comparing metaheuristics’ success in addressing familiar concerns in comparison to other algorithms, numerous well-known applications have shown the effectiveness of metaheuristics in different manners. The literature contains a significant quantity of metaheuristic methods which include Ant Colony and Particle Swarm Optimizer. Those methods use swarm intelligence principles. This study mainly employs a metaheuristic approach that is an algorithm known as the Heap Based Optimizer (HBO), in various tests to find a solution for FS problems with a higher degree of precision. To improve the selection of features process, Histogram of Oriented Gradients is integrated with a classifier referred to as the K-Nearest Neighbor, using the wrapper approach. The proposed approach’s effectiveness is then assessed opposed to seven established techniques from previous associated studies using nine high-dimensional datasets which contain small sample numbers as well as a large number of classes. In comparison to the other methods, the results show that the HBO achieves good accuracy in two datasets and reduces the amount of characteristics required for classification tasks. In addition, it surpassed the other competing techniques in terms of the speed of convergence. As such, the FS process can be optimized using the suggested HBO approach, regardless of the selection size or classification accuracy.