Omar Hujran, Department of Statistics and Business Analytics, United Arab Emirates University, Al Ain, United Arab Emirates
This research report provides a comprehensive analysis of Compact Composite Descriptors (CCDs) as a highly ef icient alternative to deep learning embeddings for Content-Based Image Retrieval (CBIR) in resource-constrained environments. While Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) of er superior semantic performance, their computational overhead and storage requirements—often exceeding 8KB per image—limit their applicability in Edge AI and IoT scenarios. In contrast, engineered descriptors such as the Color and Edge Directivity Descriptor (CEDD), Fuzzy Color and Texture Histogram (FCTH), and Joint Composite Descriptor (JCD) utilize fuzzy inference systems to encode visual features into ultra-compact vectors ranging from 54 to 72 bytes. The study explores the algorithmic foundations of these descriptors, their implementation within the LIRE (Lucene Image Retrieval) framework, and benchmarks demonstrating their competitive retrieval accuracy against MPEG-7 standards. Finally, the report highlights the strategic utility of CCDs for privacy-preserving, low-bandwidth visual search on edge devices, proposing hybrid architectures that leverage the speed of fuzzy composites with the semantic power of neural re-ranking.
ChatGPT, Education, Conversational quality, Learning outcomes, Ethical concerns
Razwan Ahmed Tanvir and Greg Speegle, Department of Computer Science, Baylor University Waco, Texas, USA
Collaborative blockchain ecosystems allow diverse groups to cooperate on tasks while providing properties such as decentralization and transaction security. We provide a model that uses a repository blockchain to manage hard forks within a collaborative system such that a single process (assuming that it has knowledge of the requirements of each fork) can access all of the blocks within the system. The repository blockchain replaces the need for Inter Blockchain Communication (IBC) within the ecosystem by navigating the networks. The resulting construction resembles a tree instead of a chain. A proof-of-concept implementation performs a depth-first search on the new structure.
Hard Fork, Shared Governance, Inter Blockchain Communication (IBC), Blockchain Ecosystem
Jesus Antonio Motta1 and Pedro David Gomez 2, 1Laval University ,Quebec (Canada), 2 Foundation University of Health Sciences, Bogota (Colombia)
In this work, we present a highly efficient machine learning method for identifying DNA sequences that code for genes. The learning process is based on Human Genome Build 38 (GRCh38) sequences extracted from various specialized databases. The sequences were then translated into amino acid sequences and used to build matrices that facilitate the extraction of features with the TF*IDF metric for the creation of the training space. The prediction functions are learned using a convolutional neural network (CNN) deep learning model. The training spaces were created using the 24 chromosomes of the human genome and approximately 36,000 genes and pseudogenes whose names were fetched from the HUGO Gene Nomenclature Committee (HGNC). Performance analysis was performed on 24 genes associated with genetic disorders, as well as the surrounding DNA regions. The metrics used were precision, recall, F_score measure, accuracy and ROC curves for the genes of interest. The results achieved exceed all our expectations and place the work at the level of the state of the art for gene prediction.
DNA, Amino-Acids, TF×IDF, CNN, Genetic Disorder, Learning Model
Maisie Mary Cu 1 Berlin School of Economics and Law, Germany
This paper examines manual rework in the construction-settlement (XDTT) workflow of WinCommerce, the modern-trade grocery-retail arm of Masan Group, which operates more than 3,000 neighbourhood stores in Vietnam. Drawing on an internal review of 100 stores and 7,913 work-item records, the study finds that 57.1% of submitted line items are flagged as exceptions (phát sinh); 50.7% of all items already exist in the in-force Bill of Quantities (BOQ) but were missed during bid preparation, while only 6.4% are genuinely new. The 6.4% trigger a 60-day price-negotiation and legal-approval cycle that blocks payment for the other 93.6%; 40% of files are also returned for material defects, adding 24 days. Findings are aggregated via Excel pivot tables over a restructured BOQ (5 categories, 15 subcategories, 156 group headers, 2,291 leaf items), and the five candidate interventions are ranked using the Analytic Hierarchy Process (CR = 0.006): separating the volume-check step from the price-approval cycle dominates (composite weight 0.336). The paper outlines an AI-assisted coordination design, semantic deduplication, automated exception triage and digitised parallel approval, as a domain-grounded blueprint for soft-computing application to large retail expansion programmes.
AI-assisted operational coordination, exception handling, Bill of Quantities (BOQ), construction settlement, semantic deduplication, retail expansion, process automation.
Sultan AlSultan and Walid Karamti1 1 Qassim University,Saudi Arabia
Early identification of students at risk of weak academic performance is essential for effective support and quality assurance in higher education. However, many educational prediction studies rely on variables that are direct components of the final outcome, which may lead to information leakage and overly optimistic per formance estimates. This study proposes an interpretable machine learning frame work for predicting student performance categories using academic, demographic, behavioral, health, engagement, and support-related variables. The target variable is represented by five ordered categories: Poor, Needs Improvement, Satisfactory, Good, and Excellent. A signal-enhanced modeling subset consisting of 16,892 student records and 42 variables was constructed through a correlation-guided filtering process. To support realistic evaluation, direct outcome-score variables were removed from the main training setting. Data preprocessing included type-specific encoding, imputation, standardization, and SMOTENC balancing applied only to the training set. Random Forest, Gradient Boosting, and LightGBM were trained and compared. LightGBM achieved the strongest overall performance with 44.4% accuracy, 44.0% precision, 40.0% recall, 41.5% F1-score, and 72.2% multiclass AUC. The confusion matrix showed that most errors occurred between neighboring categories, suggesting that the model captured meaningful ordinal structure even when exact five-class classification remained difficult. SHAP analysis identified GPA, attendance rate, research involvement, high school GPA, access to academic resources, and entrance examination score as the most influential predictors. The proposed framework provides interpretable evidence that can support academic advisors, instructors, and quality units in early intervention planning.
Student Performance Prediction, Machine Learning, Learning Analytics, Educational Data Mining, Academic Quality Assurance, Explainable AI
Mehul Vani 1 Westcliff University, Irvine, USA
Self-improving agentic systems must act in sequential environments while diagnosing failures, revising strategies, and adapting to non-stationary objectives. Conventional reinforcement learning (RL) improves a policy mainly through reward-driven updates and typically treats self-critique as an external supervision signal. This paper introduces RLM-RPTR, a self-improving agentic system that unifies (i) trust-region RL optimization, (ii) post-training consolidation via periodic offline re-optimization, and (iii) a reflection operator that generates process-level corrective signals from trajectory feedback. The methodology formalizes reflection as a bounded reward-shaping and policy-correction functional that regularizes policy updates under a KL trust region, yielding stable improvement dynamics. Experiments use a fully reproducible stochastic obstacle gridworld suite (12×12, slip=0.30, obstacle density=0.28) and a non-stationary goal-switch variant, spanning 10 random seeds and 2,000 training episodes per run. Compared with tabular Q-learning and SARSA, RLM-RPTR improves mean episodic return by 9.7% and 8.5% in the stationary regime and increases terminal success to 99.9%. Under goal switches at episode 1,000, RLM-RPTR reduces adaptation time to 0.9 success by 3.7% while preserving comparable asymptotic reward. These results support reflection-augmented post-training as a scalable mechanism for self-improving agents. We further specify a thesis-ready benchmark suite including WebShop, ALFWorld, Procgen, Habitat, and GSM8K for future scale-up and ablations.
agentic AI, reinforcement learning, reasoning language models, reflection, post-training, trust-region optimization, non-stationary environments.
Manuel Stoger, Mario Bernhart, and Thomas Grechenig, Research Group for Industrial Software (INSO), TU Wien, Vienna, Austria
Software repositories constitute rich, heterogeneous data sources whose effective exploitation is pivotalfor understanding software evolution and ensuring software quality. Knowledge graphs (KGs) and relatedgraph-based representations have emerged as a promising paradigm for structuring, querying, and rea-soning over repository data. This paper reviews 56 primary studies (2006–2025) to answer: “How haveknowledge graphs been applied to mining, analyzing, and visualizing software repositories?”. Follow-ing the Design Science Methodology by Wieringa, we classify proposed treatments, evaluate validationstrategies, and synthesize results into five application clusters: (1) ontology-based repository modeling,(2) code knowledge graph construction and querying, (3) developer and collaboration networks, (4) defect,maintenance, and traceability, and (5) software evolution and dependency analysis. The findings reveala clear evolution from early ontology-based approaches (2006–2012) through deep-learning-augmentedKGs (2017–2021) to LLM-integrated repository graphs (2023–2025). Open challenges include scalability,standardization, and the convergence of graph-based and neural approaches.
Knowledge Graph, Mining Software Repositories, Structured Literature Analysis, Ontology, Soft-ware Evolution
Chitti Srinivasa Phani1 Sedar Olmez 2 and Yokota Koichi 3 Antifragility Research Group, Fujitsu Research of Europe, Slough, United Kingdom
User-authored data-processing scripts are widely used for ETL workflows due to their flexibility and ease of development, but they frequently encode implicit assumptions about local file systems, execution order, and runtime context. These assumptions make such scripts fragile when migrated to cloud environments, where differences in storage semantics, resource constraints, and execution models can lead to silent failures or incorrect behaviour. Existing migration approaches typically rely on manual refactoring or narrowly scoped tooling, limiting scalability and reliability. This paper presents Platform Code Converter (PCC), a compiler-inspired middleware for transforming unstructured ETL scripts into portable, cloud-ready artefacts. PCC recovers operational semantics using a language-neutral Intent Grammar and a lightweight Intermediate Representation (IR), from which it derives backend-specific execution policies for chunking, batching, parallelism, and storage access. A tiered transformation engine applies correctnesspreserving rewrites, augmented by LLM-assisted, correctness-preserving refinement when deterministic rewriting is insufficient, while conservatively preserving code under semantic uncertainty. We evaluate PCC on forty heterogeneous ETL workloads, spanning synthetic patterns and real production scripts; across these, PCC applies 110 deterministic transformations, migrates over 300 file paths to cloud-safe representations, and validates all generated artefacts at execution time. These results demonstrate that intent-guided compilation provides a practical and reliable foundation for cloud migration of unstructured ETL code.
ETL migration, cloud computing, code transformation, intermediate representation, intent grammar
Usman Durrani, Mustafa Akpinar, Ghazi Alnaymat AAPoly, Higher Colleges of Technology, Ajman University ,Melbourne Australia, Sharjah UAE, Ajman UAE
Generative AI (GenAI) has fundamentally undermined the evidential basis of summative assessment in higher education. GPT-4 scores at the 90th percentile on the Uniform Bar Exam; a 2024 University of Reading study found AI-generated submissions evaded detection 94% of the time; and a longitudinal study recorded a 21.88 percentage-point rise in pass rates following ChatGPT's release, with all else held constant. Australia's regulator TEQSA concluded in September 2025 that AI-assisted cheating is "all but impossible to detect consistently," mandating structural redesign across the sector. We argue that classroom-based gamified problem-based formative assessment (GPBFA) is the response higher education needs — addressing integrity and pedagogy simultaneously. Live classroom activities cannot be AI-delegated, and formative assessment consistently produces larger learning gains than summative testing. We present an empirical study of saval.app, a GPBFA platform deployed in a university Business Analysis unit. Using the ARCS motivational model, pre-unit (N = 41) and post-unit (N = 44) surveys returned strongly positive outcomes across all four dimensions (all p < .0001; Cronbach's α > .93; 81% positive response rate). Qualitative findings confirm active engagement, deep conceptual understanding, and transferable self-efficacy that AI-assisted summative performance cannot confer.
Generative AI, Academic Integrity, Summative Assessment, Formative Assessment, Gamification, Problem-Based Learning, ARCS Model, Saval.app, Higher Education
Md. Ebna Sabed Chowdhury and Kamrul Hasan Talukder Computer Science and Engineering Discipline, Khulna University, Khulna - 9208, Bangladesh
Agriculture plays a critical and vital role in guaranteeing food security and supporting economic development. Selecting the most appropriate crop for cultivation necessitates careful consideration of soil attributes, environmental conditions, and market trends. While existing crop recommendation systems focus primarily on identifying appropriate and fitting crops, they frequently overlook future market conditions that may affect farmer’s cultivation decisions. This study proposes an intelligent agricultural decision support system that merges machine learning-based crop recommendation with crop price forecasting. Soil nutrients (N, P, and K), temperature, humidity, pH, and rainfall are used to suggest fitting crops, while future crop prices are forecasted employing the Prophet model. The proposed framework recommends the top two proper and right crops and identifies the most beneficial choice established on forecasted market prices. Experimental outcomes demonstrate that the integrated approach can support more knowledgeable and educated agricultural planning by unifying crop suitability analysis with future price trends. The system offers a pragmatic and user-friendly solution for smart farming and agricultural decision-making.
Crop Recommendation, Machine Learning, Price Forecasting, Agricultural Decision Support, Smart Farming, Prophet.
