Authors - Bhavya Balakrishnan, Srinivasa HP Abstract - The massive deployment of heterogeneous, resource- con-strained and always-on devices underlying the Internet of Things (IoT) has introduced complex cybersecurity challenges. The rapid growth of the Internet of Things (IoT) due to the large-scale deployment of heterogeneous, resource-constrained and always-on devices has resulted in complex cybersecurity challenges. The physical and digital components in the IoT systems are tightly bound which increases the attack sur-face and makes them highly prone to threats of malware infections, data theft, unauthorized access and distributed denial of service. Traditional security mechanisms and rule-based intrusion detection systems cannot manage the dynamic, large-volume and evolving IoT traffic. The solutions provided by machine learning have been widely concerned due to its capability of learning data patterns and finding abnormal and malicious activities. However, existing machine learning models have serious constraints such as lack of labelled information, extreme class imbalance, and inability to generalize to new and never-seen attacks. In recent years, Generative Adversarial Networks (GANs) have emerged as a promising paradigm to improve the cybersecurity of IoT through artificial generation of realistic synthetic data, adversarial sample enhancement, alleviating data imbalance and modelling adversarial attack-defense dynamics. GAN based models have showed great gains in intrusion detection, anomaly detection and malware analysis in the IoT networks . However, modern studies are still divided on this issue due to variations in GAN architectures, datasets, evaluation procedures, and experimental procedures. In addition, most of the researches have been more concentrated on offline benchmark databases, with less focus on checking through realistic IoT testbeds, which could be more precise in capturing the actual deployment conditions.
Authors - Ferry Setyadi Atmadja, Sabo Hermawan, Eka Dewi Utari, Suciati Putri Nurjanah, Siti Dwi Hastuti Abstract - The exorbitant costs associated with professional Content Management Systems (CMS) have precipitated a severe theory to practice gap in digital archive education. This infrastructural barrier disproportionately disadvantages institutions with constrained budgets, fundamentally threatening the inclusive education mandates of Sustainable Development Goal (SDG) 4. To bridge this ped-agogical divide, this study developed and validated a zero-license educational framework utilizing Microsoft Excel's Visual Basic for Applications (VBA) to simulate a professional electronic records environment. Employing an R&D methodology (ADDIE model) with a cohort of 40 undergraduate students, the proposed framework circumvented hardware and financial constraints by operating offline on low-specification devices. Results indicated high expert validation (4.35/5.0) and a statistically significant enhancement in students' practical archival skills, evidenced by a moderate to high Normalized Gain (N-Gain) of 0.61. Furthermore, the system demonstrated exceptional usability with a System Usability Scale (SUS) score of 76.5. These findings provide empirical evidence that strategic, low-cost technological interventions can effectively democratize digital archive learning, offering a highly scalable solution for marginalized educational ecosystems in developing regions.
Authors - Rayyan Naufal Anandito, Muhammad Fedylopa Ginting, Trias Septyoari Putranto Abstract - The rise of Automated Biometric Boarding Systems (ABBS) for public transportation, driven by the potential to enrich convenience while integrating artificial intelligence into their activities has not been without the desire among policymakers and business leaders to get a better grasp on how biometry could be integrated in mandatory adoption contexts. Abstract This study aims to investigate passenger acceptance and continuance intention of AI-based face recognition boarding system in PT Kereta Api Indonesia (KAI) Gambir Railway Station 2023. Based on an integrated framework of Technology Acceptance Model (TAM) and Expectation-Confirmation Model (ECM), complemented with Trust and Perceived Privacy Risk, this study explores the pathways through which affective factors and institutional factors influence long-term behavioral intentions in a compulsory acceptance context. Data from cross-sectional, quantitative. 150 purposively sampled passengers were analyzed by PLS-SEM using SmartPLS 4.0. This is the first time that these findings challenge many of the assumptions about technology adoption and provide relevant policy recommendations for transport authorities based on a framework for AI governance aligned with Indonesia's Personal Data Protection Law (UU No. 27/2022).
Authors - Ridhi Sharma, Ashok Kumar Abstract - This manuscript discovers the role of information theoretic measures for feature selection while dealing with high dimensional data sets. The study uses entropy, mutual information and divergence measures to address the issues of classification and high computational complexity of real data set which is affect by redundant and irrelevant features, by analyzing the dependency patterns and feature relevance in complex data set. Under different data conditions, the proposed approach for feature selection, in comparison to traditional methods, handles the non-linear relationships and noisy attributes effectively in terms of relevance, classification and interpretation. In-formation theoretic methods provide more precise feature selection and pattern identification results in the data sets. Despite the challenges of computational cost and scalability, the study shows that information theoretic measures can perform better in feature selection and decision making of the data mining.
Authors - Nishi Doshi, Shrey Shah Abstract - Hospitals run more machine learning on GPUs while the carbon footprint of grid electricity rises and falls through the day. Using a computer simulation, we compare 13 scheduling rules on mixed GPU hardware, with synthetic patient-style jobs, urgency tiers, and time-ofday carbon traces. We do not study patient outcomes; every percentage we report is a simulator queue number, not a clinical finding. We ask whether running non-urgent jobs overnight is almost as good as a richer rule that mixes urgency and carbon (CUCA at weight 0.45, written CUCA 0.45). The comparison keeps carbon reduction secondary to clinical priority and deadline compliance, so each policy is judged on both average kg CO2e and missed-deadline behavior. CarbonGreedy and CarbonShift are carbon-first stress tests that demonstrate how poorly wrong vendor presets can disrupt clinical priorities, and are not meant for production. Numbers are averages over many test settings, with wide run-to-run spread and no statistical adjustment, so headline ratios are exploratory. On an eight-GPU baseline, the overnight rule closes about 78% of the carbon gap between urgency-only and CUCA 0.45 while missing fewer urgent deadlines than either. CarbonShift lets about 46% of the most urgent jobs miss their deadline; this is simulated queueing, not bedside harm. At 48 jobs per hour, the carbon footprints almost tie, yet the overnight rule still misses fewer urgent deadlines. A geography test, where regions share one daily carbon shape with only timezone shifts, trims under one percentage point of average carbon; a twelve-hour routine window saves a little carbon for CUCA 0.45 but raises overall missed deadlines. Overnight batching stays competitive on average modelled carbon; carbon-only rules belong only in stress tests.
Authors - Abhijit Dnyaneshwar Jadhav, Prashant G. Ahire, Madhuri Hiwale Abstract - In vitro fertilization (IVF) is currently one of the most powerful assisted reproductive technologies for infertility treatment. However, the embryo selection process still represents a bottleneck that greatly influences the rates of implantation and live birth. Traditional methods of embryo evaluation involve embryo morphology grading. But this approach suffers from subjectivity, variability, and heavily depends on the skill and experience of the embryologist. To go beyond the limitations of human assessment, the latest improvements in artificial intelligence (AI), machine learning (ML), and deep learning (DL) have made possible the automated embryo evaluation using pictures, time-lapse morphokinetics, and clinical data. This paper reviews comprehensively the currently available AI-enabled IVF systems while also first introducing the conventional embryo assessment and later presenting the most sophisticated multimodal deep learning frameworks. The paper also discusses some of the major outstanding issues such as the poor performance of models on new datasets, the lack of the shared and agreed upon benchmarks, and the limited explainability of the models. We have also developed a Multimodal Explainable Artificial Intelligence Frame-work for IVF (MEAIF-IVF) to fill in these gaps in which image of the embryo, time-lapse video of the embryo, and clinical patient information are all combined into one deep learning model. This system uses convolutional neural networks and vision transformers for spatial feature extraction, recurrent neural networks for temporal modeling, and attention-based fusion for multimodal integration.
Authors - Peruru Gayathri, Rohini M, Anand R Nair Abstract - Cyber threats are getting more sophisticated and conventional security solutions are not keeping up with detecting cyber-attack. In this research, a hybrid detection and prediction system for TTP (Tactics, Techniques and Procedures) based on deep learning and graph-based is presented. The planned study is based on an analysis of data originating from cyber security systems at large scale, which can be used to detect attack patterns and correlations of attacks. Host logs and threat intelligence data are trained using deep learning models to detect discriminative features, while graph-based models are used to model the structural relationships between users, systems, and attack patterns. Combined these techniques will result in more complex attacks and lateral movement being easier to detect. It also assumes probable attack methods to move to the next level, so that it can predict the attacks and take proactive actions to mitigate attacks in the future. The entire predictive and graph based solution enhances threat visibility and threat response speed, while boosting threat detection accuracy. The system enables the detection of the APTs and real time monitoring them by the Cyber Security analysts. The experimental results show that the highest accurate transformer is able to achieve 95% classification accuracy, and the graph neural network is demonstrated to achieve 78.26% accuracy for predicting next technique. The framework has been shown end-to-end, with the intent of showing it can be utilized as an extra layer of Intelligence on the enterprise security side, with Splunk.
