GPU Computing

Network security is a critical aspect of modern enterprise information technology infrastructure. Virtual Private Network (VPN) is one of the primary solutions for securing data communications over public networks. This study aims to analyze and compare the security of three most widely used VPN protocols, namely OpenVPN, WireGuard, and IPSec, in the context of corporate network environments. The research methodology uses an experimental approach with testing on controlled network infrastructure, covering aspects of encryption, authentication, data integrity, security vulnerabilities, as well as throughput and latency performance. Testing was conducted using Cisco devices, Ubuntu Server 22.04, and Wireshark network analysis tools. The results show that WireGuard provides the highest performance with lower encryption overhead (12%) compared to OpenVPN (24%) and IPSec (18%), but OpenVPN excels in security configuration flexibility and cross-platform compatibility. IPSec proved most suitable for site-to-site VPN implementation in existing enterprise infrastructure. Research conclusions provide recommendations for VPN protocol selection based on specific company needs, considering trade-offs between security, performance, and management ease.

The development of future distributed systems faces significant challenges regarding latency, data security, and scalability due to the explosion of Internet of Things (IoT) devices. The integration of Blockchain and Edge Computing has emerged as a new paradigm to address these issues. This literature review aims to analyze how the synergy between these two technologies can optimize distributed systems. Edge Computing plays a crucial role in moving data processing closer to the source to reduce latency, while Blockchain provides a decentralized security layer, transparency, and data integrity without the need for a central authority. The results of this review indicate that this hybrid architecture enhances operational efficiency and system trust, although challenges remain regarding resource consumption on edge devices. This article provides insights for the development of more responsive and secure digital infrastructures in the future.

The development of future distributed systems faces significant challenges regarding latency, data security, and scalability due to the explosion of Internet of Things (IoT) devices. The integration of Blockchain and Edge Computing has emerged as a new paradigm to address these issues. This literature review aims to analyze how the synergy between these two technologies can optimize distributed systems. Edge Computing plays a crucial role in moving data processing closer to the source to reduce latency, while Blockchain provides a decentralized security layer, transparency, and data integrity without the need for a central authority. The results of this review indicate that this hybrid architecture enhances operational efficiency and system trust, although challenges remain regarding resource consumption on edge devices. This article provides insights for the development of more responsive and secure digital infrastructures in the future.

This research identifies two simultaneous crises in modern vision AI systems. First, Tensor Cores suffer from progressive, cumulative degradation under sustained AI workloads due to a cache inefficiency gap that, based on first-principles memory locality analysis, forces reliance on slower memory, generating thermal hotspots and theoretically reducing effective GPU lifespan from five years to approximately two years. Second, current vision models—including Vision Transformers and Mamba—suffer from geometric blindness: they flatten 2D images into 1D sequences, losing spatial relationships and causing edge distortion, dimensional incoherence, and presence-absence ambiguity.

This work proposes a theoretical alternative architecture built on the Zig programming language. The architecture operates on spatial coordinates rather than pixels, reducing data volume by an estimated 99 percent. It employs the ZR2ViM (Zigzag Recursive Vision Mamba) model to preserve spatial dimensions, and implements a stress-aware CUDA scheduler to theoretically mitigate Tensor Core decay.

Based on extrapolations from existing Zig projects (ZML, cudaz) and published results from the ZR2ViM paper, this work projects a theoretical speedup of 12.8x over Python baselines (193 vs 15 tokens per second), 98 percent edge detection accuracy, extension of effective GPU lifespan from two years to five years, and statistically significant artist attribution accuracy using geometric signature extraction. All projections are theoretical and require empirical validation.

This work challenges the peak performance paradigm, introduces geometric blindness as a fundamental limitation of current models, and demonstrates that a complete Zig-native vision stack already exists. The architecture offers a new theoretical direction for hardware-software co-design in vision AI, with implications for industry, academia, open-source development, and digital art history.

This paper presents a comprehensive analysis of the Microsoft Build 2026 opening keynote, delivered by CEO Satya Nadella and senior leadership at Fort Mason Center, San Francisco, on June 2–3, 2026. The keynote articulated a fundamental strategic pivot in the technology industry: the transition from operating-system-centric and application-centric computing paradigms toward what Microsoft terms "agentic computing" — a model in which autonomous AI agents perform complex, multi-step tasks across devices, cloud services, and enterprise systems.
Drawing on the announcements presented at the conference, this thesis examines seven critical dimensions of this transition: (1) the introduction of a proprietary MAI model family signaling reduced reliance on third-party AI providers; (2) the Surface RTX Spark Dev Box as a hardware embodiment of local AI compute ambitions; (3) Project Solara as an agent-native operating platform; (4) Microsoft Execution Containers as a governance and security framework; (5) the GitHub Copilot application as the developer interface for the agentic frontier; (6) Frontier Tuning as an enterprise personalization mechanism; and (7) the Majorana 2 quantum chip as the long-term computational foundation.
The analysis positions these announcements within the broader competitive landscape of AI development, evaluates their technical merit and strategic coherence, and identifies implications for software development practices, enterprise IT governance, and the future of human-computer interaction.

This study presents the development of a fully integrated mobile module that enables indoor pathfinding functionality from the front end to the back end. The module is implemented using the A* algorithm for route optimization and a NestJS framework with PostgreSQL and PostGIS for spatial data management. Designed as part of the University of the Fraser Valley (UFV) Campus App project, this cross-platform mobile application is built with React Native to ensure seamless usability across devices. Core functionalities include intelligent room search, interactive floor plan visualization, and real-time, turn-by-turn navigation powered by WebSocket communication. The system demonstrates the feasibility of combining spatial databases, efficient routing algorithms, and real-time communication technologies to enhance campus navigation and user experience.

The DNenc Engine is a hyper-accelerated symmetric cryptographic engine that completely offloads data obfuscation pipelines to the GPU, designed to eliminate sequential CPU core limitations in ultra-low-latency environments such as High-Frequency Trading (HFT) and live telemetry routing.

Utilizing a proprietary 5-dimensional GF(P) spatial manifold derived from the Dynamic Needle Framework, the engine operates in a massively parallel Counter (CTR) mode within isolated VRAM. Empirical benchmarks demonstrate a compute-bound peak hardware throughput of 4.52 GB/s on a consumer-grade Nvidia RTX 3070, engineered to scale linearly across enterprise clusters via GPUDirect RDMA integration. The output keystream has been verified to generate uniform, incompressible entropy via the complete NIST SP 800-22 statistical test suite, establishing a robust 160-bit post-quantum security margin.

This repository contains the foundational architectural whitepaper detailing the mathematical derivations, coordinate projection mechanics, and hardware execution parameters.

Artificial intelligence (AI) has significantly contributed to car make and model recognition in this current era of intelligent technology. By using AI, it is much easier to identify car models from any picture or video. This paper introduces a new model by fine-tuning the AlexNet architecture to determine the car model from images. First of all, our car image dataset has been created. Some of these images were taken by us, and others were taken from the website of the car connection. Then we cleaned all the unwanted images for better performance. Our dataset has ten classes containing 5,000 car images split into train and test data. After that, we augmented our data with random flip, rotation, and zoom to reduce overfitting. Finally, we used a pre-trained convolutional neural network (CNN) model AlexNet architecture. We fine tuned AlexNet (FT-AlexNet) by adding three extra layers for better classification and compared it with the original AlexNet. To measure the performance of these models, accuracy, precision, recall, and F1-score were used. The results show that fine-tune AlexNet architecture outperforms the original AlexNet architecture. The results prove that recognition accuracy has increased due to our improvement approach.

LLM serving systems increasingly move large tensors between host memory and GPU memory: model weights during cold start, KV-cache blocks during context reuse, and inactive states during multi-tenant scheduling. Today these transfers use the PCIe path attached to the destination GPU. On an eight-GPU server, this means one link is saturated while seven host-to-GPU links remain idle. VeloceIO exploits that asymmetry. Its core idea is simple: split a large host-to-device copy into chunks, send chunks through otherwise idle peer GPUs over their PCIe links, and assemble the payload on the target GPU over the faster intra-node P2P fabric. VeloceIO implements this idea below the application through CUDA runtime interposition, a bounded task scheduler, double-buffered relay copies, and a device-side stream barrier that preserves CUDA ordering without blocking the host thread. On an 8× NVIDIA H20 node, VeloceIO raises effective H2D bandwidth from 53 GB/s to 245 GB/s. In serving workloads, this reduces 128K-token context fetch latency from 5.10 s to 2.15 s and reduces 64GB model cold-start latency from 2.10 s to 0.88 s. A congestion-aware relay policy limits slowdown on co-located compute kernels to 2.7%.

The rapid growth of data-intensive applications has intensified the demand for scalable, resilient, and cost-efficient data processing pipelines capable of handling highly variable workloads. Conventional data pipeline architectures built on statically provisioned virtual machines often struggle to meet these demands due to limited elasticity, high operational overhead, and inefficient resource utilization during fluctuating load conditions. To address these challenges, this study presents the design, implementation, and empirical evaluation of a hybrid cloud-native data processing pipeline that integrates serverless computing with container-based orchestration. The proposed architecture adopts a layered design that strategically aligns workload characteristics with appropriate execution models. Event-driven serverless functions are employed for data ingestion, validation, and lightweight transformation, enabling rapid parallel execution and near-zero idle resource consumption. Compute-intensive, stateful, and long-running processing tasks are deployed as containerized microservices managed by an orchestration platform, providing execution stability, fine-grained resource control, and scalable performance. Workflow orchestration mechanisms coordinate inter-stage dependencies, retries, and failure handling, while cloud-native storage systems support both real-time analytics and batch processing. A rigorous experimental methodology was employed to evaluate the pipeline under diverse workload conditions, including bursty event streams, sustained high-throughput processing, and controlled failure scenarios. Performance was benchmarked against a traditional VM-based pipeline using metrics such as ingestion latency, end-to-end processing time, throughput scalability, cost distribution, and recovery time under failure. The results demonstrate that the hybrid pipeline achieves near-linear throughput scaling as compute resources increase, reduces ingestion latency by a significant margin, and recovers from common failure scenarios an order of magnitude faster than VM-centric architectures. Cost analysis further reveals substantial reductions in idle resource waste due to fine-grained autoscaling and pay-per-use execution. The findings confirm that integrating serverless and container-based services within a unified architectural framework delivers superior scalability, reliability, and cost efficiency for modern data processing workloads. This work provides a practical reference architecture and empirical evidence supporting the adoption of hybrid cloud-native pipelines for real-time analytics and large-scale data engineering applications.

This document constitutes the first Working Paper of the RODAN AI project — the strategic successor to the Odra 1305 AI system. It describes a pivotal moment: the encounter with the hardware barrier and the resulting merger of the Odra 1305 AI project with the strategic RODAN AI project. The document formalises the decision to accelerate the project by combining the Odra system layer with a dedicated military-grade hardware platform — the RODAN computer.
The document also describes key milestones: the launch of the Odra voice layer (Whisper STT + pyttsx3 TTS) and the registration of the JSON Communication Protocol v1.1 in Zenodo (DOI: 10.5281/zenodo.20118639). Activation of letters of intent to Microsoft Corporation and Anthropic PBC is now necessary.

Modern cloud platforms host large-scale data pipelines that power analytics, machine learning, and real-time decision systems across industries ranging from healthcare to finance. As data volumes grow exponentially and workloads become increasingly heterogeneous spanning batch analytics, streaming ingestion, and GPUaccelerated ML tasks traditional rule-based scheduling and static resource provisioning mechanisms struggle to satisfy evolving performance, cost, and reliability constraints. These legacy approaches often rely on fixed heuristics, overprovisioned clusters, or reactive scaling, leading to suboptimal utilization, latency bottlenecks, and inflated operational expenses. This article explores AI-driven approaches to optimizing cloud data pipelines by embedding machine learning into workflow scheduling, predictive autoscaling, anomaly detection, and cross-layer resource orchestration. Grounded in foundational distributed systems research such as the MapReduce paradigm introduced by Jeffrey Dean and Sanjay Ghemawat, as well as the DAG-based dataflow model of Apache Spark, and extended by modern serverless execution engines and global optimization frameworks like AGORA, we propose an architectural blueprint that leverages telemetry-driven feedback loops, predictive modeling, and reinforcement learning to enable adaptive, cost-aware decision-making. By synthesizing key research developments from 2000 to 2022, this work outlines how intelligent scheduling policies, workload forecasting, and heterogeneity-aware optimization can transform static pipelines into selfoptimizing systems, while also identifying future research directions in reinforcement learning-based orchestration, multi-cloud cost modeling, and explainable AI for cloud execution environments.

Insiden hilangnya data pada GitLab.com pada 31 Januari 2017 menjadi titik balik krusial dalam pemahaman industri mengenai manajemen ketahanan sistem dan pemulihan bencana. Makalah ini menganalisis secara mendalam bagaimana satu kesalahan manusia (human error) dalam prosedur perbaikan replikasi basis data mengakibatkan penghapusan seketika terhadap 300 GB data produksi PostgreSQL. Kajian ini membedah kegagalan simultan dari lima mekanisme cadangan yang dimiliki GitLab—mulai dari cuplikan LVM, cadangan native PostgreSQL, hingga penyimpanan awan di Azure dan AWS S3—yang semuanya gagal berfungsi karena kesalahan konfigurasi, ketidaksesuaian versi, dan kegagalan autentikasi. Melalui analisis root cause, makalah ini mengevaluasi kelemahan dalam Rencana Kontinuitas Bisnis (BCP) dan Tujuan Titik Pemulihan (RPO) yang membengkak hingga lebih dari 6 jam. Hasil kajian merumuskan rekomendasi pragmatis berbasis pengembalian investasi (ROI), termasuk implementasi chaos engineering, validasi pemulihan otomatis, dan penguatan pengendalian akses melalui prinsip hak akses minimal. Kesimpulan utama menegaskan bahwa ketahanan operasional tidak hanya bergantung pada teknologi canggih, melainkan pada pengujian sistematis yang berkelanjutan dan desain sistem yang toleran terhadap kesalahan manusia.

Puji syukur senantiasa dipanjatkan atas kehadirat Tuhan Yang Maha Esa, yang telah melimpahkan rahmat dan karunia-Nya, sehingga penulisan makalah berjudul "Kajian Business Continuity Plan (BCP) & Post-Disaster Recovery Emergency Procedures terhadap Cloud Disaster Recovery Universitas Pamulang" ini dapat diselesaikan dengan baik dan tepat waktu. Makalah ini disusun untuk memenuhi tugas mata kuliah Manajemen Risiko dan Keamanan Informasi yang diampu oleh Bapak Zulhair Jidan Dj. Tamu S.kom., M.kom.. Penulis menyadari sepenuhnya bahwa penyusunan makalah ini tidak lepas dari berbagai referensi dan studi kasus mengenai risiko kegagalan sistem teknologi informasi, khususnya kajian literatur mengenai manajemen kesinambungan bisnis berbasis komputasi awan. Penulis berharap kajian ini dapat memberikan wawasan dan manfaat akademik, khususnya dalam memahami pentingnya ketersediaan sistem (availability) pada operasional institusi pendidikan tinggi. Penulis menyadari bahwa makalah ini masih jauh dari kata sempurna. Oleh karena itu, kritik dan saran yang bersifat membangun sangat diharapkan demi perbaikan di masa mendatang.

Puji syukur dipanjatkan ke hadirat Tuhan Yang Maha Esa, karena atas rahmat dan karunia-Nya, makalah yang berjudul "Kajian Business Continuity Plan & Post-Disaster Recovery Emergency Procedures Pada Norsk Hydro Ransomware Attack (2019)" ini dapat terselesaikan tepat pada waktunya. Makalah ini disusun untuk memenuhi tugas mata kuliah Manajemen Risiko dan Keamanan Informasi. Selain itu, penulis berharap makalah ini dapat menambah wawasan bagi para pembaca mengenai Business Continuity Plan dan Disaster Recovery Plan. Penulis mengucapkan terima kasih kepada Bapak Zulhair Jidan Dj. Tamu, S.Kom., M.Kom., selaku Dosen mata kuliah ini. Ucapan terima kasih juga penulis sampaikan kepada semua pihak yang telah membantu dalam proses penyusunan makalah ini. Penulis menyadari bahwa makalah ini masih jauh dari kata sempurna. Oleh karena itu, kritik dan saran yang membangun sangat diharapkan demi kesempurnaan makalah ini di masa yang akan datang. Akhir kata, semoga makalah ini dapat memberikan manfaat bagi kita semua.

Puji syukur dipanjatkan ke hadirat Tuhan Yang Maha Esa, karena atas rahmat dan karunia-Nya, makalah yang berjudul "Kajian Business Continuity Plan & Post-Disaster Recovery Emergency Procedures Pada Norsk Hydro Ransomware Attack (2019)" ini dapat terselesaikan tepat pada waktunya. Makalah ini disusun untuk memenuhi tugas mata kuliah Manajemen Risiko dan Keamanan Informasi. Selain itu, penulis berharap makalah ini dapat menambah wawasan bagi para pembaca mengenai Business Continuity Plan dan Disaster Recovery Plan. Penulis mengucapkan terima kasih kepada Bapak Zulhair Jidan Dj. Tamu, S.Kom., M.Kom., selaku Dosen mata kuliah ini. Ucapan terima kasih juga penulis sampaikan kepada semua pihak yang telah membantu dalam proses penyusunan makalah ini. Penulis menyadari bahwa makalah ini masih jauh dari kata sempurna. Oleh karena itu, kritik dan saran yang membangun sangat diharapkan demi kesempurnaan makalah ini di masa yang akan datang. Akhir kata, semoga makalah ini dapat memberikan manfaat bagi kita semua.

Skin cancer is an example of the most dangerous disease. Early diagnosis of skin cancer can save many people’s lives. Manual classification methods are time-consuming and costly. Deep learning has been proposed for the automated classification of skin cancer. Although deep learning showed impressive performance in several medical imaging tasks, it requires a big number of images to achieve a good performance. The skin cancer classification task suffers from providing deep learning with sufficient data due to the expensive annotation process and required experts. One of the most used solutions is transfer learning of pre-trained models of the ImageNet dataset. However, the learned features of pre-trained models are different from skin cancer image features. To end this, we introduce a novel approach of transfer learning by training the pre-trained models of the ImageNet (VGG, GoogleNet, and ResNet50) on a large number of unlabelled skin cancer images, first. We then train them on a small number of labeled skin images. Our experimental results proved that the proposed method is efficient by achieving an accuracy of 84% with ResNet50 when directly trained with a small number of labeled skin and 93.7% when trained with the proposed approach.

This report analyzes the Large Processing Unit (LPU), a radically reimagined computer architecture that functions as the backbone of the Naciro Engine. It details how the LPU overcomes the von Neumann bottleneck through the Tensor Streaming Processor (TSP) paradigm, integrated SRAM, and temporal determinism to enable real-time autoregressive inference.

In densely built urban areas, explosive events produce blast waves that interact strongly with surrounding structures, resulting in amplified pressures and highly variable injury patterns. These complex interactions challenge conventional blast models and emphasize the need for improved, rapid injury estimation methods to support effective emergency response and mitigation. To address this need, the present study develops predictive equations for the rapid estimation of TNT blast-induced human injuries in urban environments, with particular emphasis on densely built areas characterized by complex street geometries. These equations were derived through regression analysis of an extensive dataset of simulated blast events with varying charge sizes, placed in three representative urban scenarios characterized by distinct street layouts and topographic conditions. The numerical simulations were performed using validated finite-volume Computational Fluid Dynamics models implemented in the GPUaccelerated commercial software Viper::Blast. The proposed equations allow the delineation of primary and secondary injury zones. Primary injuries are associated with overpressure effects on air-filled organs, while secondary injuries result from debris and projectiles and are directly correlated with predicted damage to masonry elements and glazing. Tertiary and quaternary injuries, which are respectively related to body displacement and impact, burns, inhalation hazards, and psychological effects, are not addressed in the present research. Results show that the urban fabric strongly influences both the extent and the shape of the injury zones. In contrast, terrain topography, provided that slopes remain relatively moderate, has only a minimal effect. Overall, the predictive equations offer a rapid yet reliable tool for quantifying blast-related injuries in urban environments. They support risk assessment, mitigation planning, and emergency response strategies for both accidental and deliberate explosions.

Matrix multiplication is a fundamental operation in linear algebra libraries, serving as the computational backbone for scientific computing, machine learning, and data analytics applications. This paper presents a comprehensive analysis of optimization techniques for General Matrix Multiply (GEMM) operations on modern CPU architectures. We examine six core optimization strategies: loop unrolling, vectorization using SIMD instructions, cache blocking, matrix packing, hardware-specific acceleration (including Intel AMX and IBM Power10 MMA), and prefix caching. Additionally, we discuss specialized optimizations for matrix-vector multiplication (GEMV) when matrix dimensions reduce to vectors. Our experimental results demonstrate performance improvements ranging from 2x to 24x over naive implementations, achieving 73-78% of theoretical peak performance across IBM Power architectures. The paper provides detailed implementation examples, particularly highlighting IBM Power10 Matrix-Multiply Assist (MMA) instructions based on OpenBLAS implementations.

This study aims to analyze risk treatment strategies in campus information systems, including e-learning, academic information systems (SIAKAD), and campus WiFi networks. The study is based on real issues such as data breaches and system vulnerabilities, as seen in cases at Universitas Udayana.

The research identifies several major risks, including data leakage, account misuse, lack of monitoring, insecure server access, and website attacks. Each risk is analyzed using four risk treatment approaches: risk avoidance, risk reduction, risk transfer, and risk acceptance. The results show that risk reduction is the most appropriate strategy for most risks due to the necessity of maintaining system functionality.

Furthermore, the study prioritizes risks based on their impact and likelihood, with data breaches and account misuse identified as the most critical. The findings provide a foundation for implementing appropriate security controls in future system development. This research contributes to improving information security management in higher education institutions.

Perkembangan teknologi informasi di lingkungan perguruan tinggi telah mendorong penggunaan berbagai sistem berbasis digital seperti e-learning, Sistem Informasi Akademik (SIAKAD), serta jaringan WiFi kampus. Sistem-sistem tersebut berperan penting dalam mendukung proses pembelajaran, pengelolaan data akademik, serta aktivitas administrasi di kampus. Namun, penggunaan teknologi informasi juga diiringi dengan meningkatnya risiko keamanan yang dapat mengganggu operasional sistem. Risiko tersebut meliputi kebocoran data, penyalahgunaan akun, serangan terhadap sistem, hingga gangguan jaringan. Kasus kebocoran data mahasiswa di Universitas Udayana menjadi contoh nyata bahwa sistem kampus masih memiliki kelemahan dalam pengelolaan keamanan informasi. Dalam manajemen risiko, tidak semua risiko dapat dihilangkan sepenuhnya, sehingga diperlukan strategi perlakuan risiko (risk treatment) yang tepat. Risk treatment merupakan proses menentukan bagaimana organisasi akan menangani risiko, apakah dengan menghindari, mengurangi, mentransfer, atau menerima risiko tersebut. Oleh karena itu, diperlukan analisis yang mendalam untuk menentukan strategi perlakuan risiko yang tepat pada sistem e-learning, SIAKAD, dan jaringan WiFi kampus, sehingga dapat menjadi dasar dalam penerapan kontrol keamanan pada tahap selanjutnya. Tujuan dari penelitian ini adalah sebagai berikut: 1. Mengidentifikasi risiko yang terdapat pada sistem kampus 2. Menentukan strategi perlakuan risiko (risk treatment) 3. Menentukan prioritas risiko berdasarkan tingkat dampak dan kemungkinan 4. Menyusun rencana mitigasi awal sebagai dasar implementasi kontrol keamanan

Transformasi digital di institusi pendidikan tinggi telah mengubah cara kegiatan akademik dan administrasi dijalankan. Informasi kini menjadi aset strategis yang menentukan kelangsungan operasional dan reputasi organisasi. Sistem E-learning, Sistem Informasi Akademik (SIAKAD), dan jaringan WiFi kampus bukan lagi sekadar fasilitas tambahan, melainkan tulang punggung dari Triad Keamanan Informasi: Confidentiality (Kerahasiaan), Integrity (Integritas), dan Availability (Ketersediaan). Meskipun sistem ini memberikan kemudahan, ancaman siber terus berkembang mulai dari phishing hingga ransomware yang menargetkan data kritis. Berdasarkan hasil audit dan identifikasi masalah pada sistem kampus saat ini, ditemukan kerentanan yang serius: akses akun sering disalahgunakan, ketiadaan aturan keamanan yang jelas, ruang server yang mudah diakses tanpa otorisasi, serta tidak adanya monitoring aktivitas pengguna. Mengibaratkan keamanan informasi seperti menjaga sebuah rumah, saat ini kampus memiliki "rumah" dengan banyak aset berharga di dalamnya, namun pintu gerbangnya dibiarkan terbuka tanpa penjagaan. Oleh karena itu, diperlukan perancangan kontrol keamanan berlapis yang mencakup Kontrol Teknis, Administratif, dan Fisik. Makalah ini akan membahas perancangan strategi kontrol tersebut secara mendalam untuk memitigasi risiko operasional dan siber di lingkungan kampus. 1. Mengidentifikasi dan menganalisis 5 risiko keamanan utama yang mengancam Elearning, SIAKAD, dan jaringan WiFi kampus. 2. Mengklasifikasikan solusi keamanan ke dalam tiga pilar utama: Kontrol Teknis, Kontrol Administratif, dan Kontrol Fisik. 3. Merancang strategi implementasi kontrol beserta justifikasi analisis pemilihannya. 4. Mengevaluasi kelemahan sistem apabila kontrol tidak diterapkan dan mengidentifikasi risiko sisa (residual risk).

Sistem Akademik (SIAKAD) merupakan tulang punggung operasional perguruan tinggi modern. Sistem ini mengelola data kritis seperti nilai akademik mahasiswa, Kartu Rencana Studi (KRS), transkrip, jadwal kuliah, dan data personal ribuan pengguna aktif setiap semester. Namun, berbagai temuan di lapangan menunjukkan bahwa implementasi SIAKAD di banyak kampus masih rentan terhadap berbagai ancaman keamanan informasi. Lemahnya autentikasi, absennya monitoring, serta akses fisik server yang tidak terkontrol menjadi kombinasi berbahaya yang dapat dieksploitasi oleh pihak-pihak tidak bertanggung jawab. Laporan ini menyajikan analisis mendalam berbasis skenario nyata kampus, mencakup identifikasi risiko, perancangan kontrol keamanan tiga lapis (teknis, administratif, fisik), serta analisis konsekuensi apabila kontrol tidak diterapkan.

Penggunaan teknologi informasi dilingkungan universitass saat ini lagi sekedar sebagai alat bantu, tetapi telah bertansformasi menjadi elemen esensial dalam pelaksanaan kegiatan akademik. Beragam sistem seperti E-learning, Sistem Akademik(SIAKAD), dan Jaringan WiFi di kampus dimanfaatkan untuk mendukung proses pendidikan, pengelolaan data, serta komunikasi diantara anggota komunitas akademis. Namun, dibalik manfaat tersebut, penerapan sistem informasi juga membawa sejumlah potensi ancaman yang dapat mempengaruhi kelancaran operasional lembaga. Ancaman tersebut bisa berupa masalah teknis, permasalahan keamanan data, sampai dengan kerusakan sistem yang bisa mengganggu jalannya proses pendidikan. Tanpa adanya manajemen risiko yang efektif, keberlanjutan layanan yang bergantung pada teknologi bisa terpengaruh dan bahkan menyebabkan kerugian yang besar bagi lembaga tersebut. Dalam hal ini, Manajemen Risiko menjadi elemen krusial yang perlu diperhatikan oleh administrator sistem informasi di lingkungan akademik. Salah satu langkah utama dalam manajemen risiko adalah penanganan risiko, yang merupakan proses untuk menentukkan strategi yang efeketis dalam mengatasi resiko yang diidentifikasi. Strategi ini mencakup penghindaran resiko, pengurangan risiko, transfer risiko, maupun penerimaan risiko, tergantu pada Tingkat dampak dan probabilitas terjadinya. Dengan demikian, dibutuhkan analisis menyeluruh terhadap risiko yang ada pada sistem Elearning, SIAKAD, dan Jaringan WiFi di kampuus, serta penetapan strategi penanganan risiko yang tepat sebagai landasan dalam penerapan kontrol keamanan di tahap berikutnya.

The current hospitals require quick and dependable computing platforms to facilitate real-time clinical artificial intelligence. The paper is quantitative research on designing GPU accelerated supercomputing system in large hospitals. The suggested platform is based on multi-GPU and multi-node architecture that implies serving medical imaging, patient monitoring, and clinical AI workloads. The experimental performance indicates that end-to-end latency that was 210ms on a single GPU dropped to 61ms on 8GPUs and throughput was increased to 238 samples per second compared to 48 samples per second. There was increased efficiency in the use of resources with the use of GPUs increasing to 88 percent. To a scale of 4 GPUs and 8 GPUs efficiency was above 0.89 and 0.78 respectively. The reliability tests had 99.8 system availability, mean recovery time of 18 seconds with no failed clinical activities recorded. These findings indicate that supercomputing architecture with a graphics card can address the next-generation clinical AI demands of real-time, scalable, and reliable computing needs of the large hospital setting.

Graphics Processing Units (GPUs) are pivotal in modern computational tasks, ranging from real-time rendering to artificial intelligence. However, the diversity of hardware capabilities often leads to suboptimal performance on low-to mid-range systems. This paper proposes a dynamic, middleware-based adaptive framework designed to optimize GPU utilization by monitoring real-time metrics and adjusting workloads programmatically. By implementing a multi-layered optimization logic, the framework bridges the performance gap across heterogeneous hardware. Experimental results show a 35% increase in frame rate stability and a 20% reduction in thermal spikes.

Modern radio telescopes like the Square Kilometer Array (SKA) will need to process in real-time exabytes of radio-astronomical signals to construct a high-resolution map of the sky. Near-Memory Computing (NMC) could alleviate the performance bottlenecks due to frequent memory accesses in a state-of-the-art radio-astronomy imaging algorithm. In this paper, we show that a sub-module performing a two-dimensional fast Fourier transform (2D FFT) is memory bound using CPI breakdown analysis on IBM Power9. Then, we present an NMC approach on FPGA for 2D FFT that outperforms a CPU by up to a factor of 120x and performs comparably to a high-end GPU, while using less bandwidth and memory.

Modern cloud computing environments demand dynamic, intelligent resource allocation strategies capable of adapting to fluctuating workloads while minimizing operational expenditure. This paper presents a comprehensive framework for intelligent cloud resource management by integrating machine learning (ML) algorithms with advanced observability tools to achieve simultaneous cost and performance optimization. The proposed system leverages real-time telemetry dataencompassing metrics, logs, and distributed traces-collected through observability platforms such as Prometheus, Grafana, and OpenTelemetry, which are subsequently processed by predictive ML models including Long Short-Term Memory (LSTM) networks and reinforcement learning agents. These models enable proactive auto-scaling, anomaly detection, and workload forecasting, significantly reducing over-provisioning and under-utilization of cloud resources. Experimental evaluations conducted across multi-cloud and hybrid environments demonstrate that the integrated framework achieves up to 35% reduction in infrastructure costs while maintaining service-level agreement (SLA) compliance exceeding 99.5%. Furthermore, the system exhibits adaptive behavior under sudden traffic spikes, outperforming conventional threshold-based autoscaling mechanisms. The findings underscore the transformative potential of combining ML-driven intelligence with full-stack observability, establishing a scalable and robust foundation for next-generation cloud resource governance in enterprise-grade deployments.

Open-source software underpins much of the modern digital economy. The prevailing belief within software engineering communities is that transparency improves reliability and security because publicly accessible code allows for broader scrutiny and faster detection of errors. However, the rapid emergence of generative artificial intelligence capable of analyzing large volumes of code may significantly alter the dynamics of vulnerability discovery. This paper proposes Manikutty's Audit Shock Effect, a conceptual framework suggesting that technological advances that dramatically reduce the cost of system inspection can trigger a sudden surge in defect discovery within transparent systems. Because open-source projects expose their codebases publicly, they are particularly susceptible to large-scale automated auditing. The short-term consequence may be a surge in discovered vulnerabilities that overwhelms the maintenance capacity of smaller projects and legacy infrastructure libraries. In the longer term, however, such transparency combined with automated analysis may strengthen the resilience of open-source ecosystems. The paper discusses implications for organizations dependent on open-source infrastructure and outlines directions for future research.

Modern cloud computing environments demand dynamic, intelligent resource allocation strategies capable of adapting to fluctuating workloads while minimizing operational expenditure. This paper presents a comprehensive framework for intelligent cloud resource management by integrating machine learning (ML) algorithms with advanced observability tools to achieve simultaneous cost and performance optimization. The proposed system leverages real-time telemetry dataencompassing metrics, logs, and distributed traces-collected through observability platforms such as Prometheus, Grafana, and OpenTelemetry, which are subsequently processed by predictive ML models including Long Short-Term Memory (LSTM) networks and reinforcement learning agents. These models enable proactive auto-scaling, anomaly detection, and workload forecasting, significantly reducing over-provisioning and under-utilization of cloud resources. Experimental evaluations conducted across multi-cloud and hybrid environments demonstrate that the integrated framework achieves up to 35% reduction in infrastructure costs while maintaining service-level agreement (SLA) compliance exceeding 99.5%. Furthermore, the system exhibits adaptive behavior under sudden traffic spikes, outperforming conventional threshold-based autoscaling mechanisms. The findings underscore the transformative potential of combining ML-driven intelligence with full-stack observability, establishing a scalable and robust foundation for next-generation cloud resource governance in enterprise-grade deployments.

The dominant conversation about AI focuses almost entirely on the model layer — parameters, benchmarks, compute. This paper argues the real bottleneck for most users is not the model but the interaction layer: how humans structure, frame, and extract value from AI systems. Drawing on the Mirror Effect — the observation that AI amplifies user signal quality rather than supplying signal independently — the paper introduces Interaction Layer Intelligence (ILI) as a distinct and developable domain. The central thesis is that model capability has substantially outpaced interaction usability, and that the next significant frontier in AI is not solely smarter models but better interaction architecture.

7/04/2026

Artificial Intelligence (AI) has rapidly reshaped the cybersecurity landscape, introducing both unprecedented defensive capabilities and a new generation of sophisticated threats. As AI systems become deeply embedded in digital infrastructures, adversaries increasingly exploit machine learning models, automated decision engines, and largescale data pipelines to launch faster, more adaptive, and more evasive cyberattacks. This paper examines the emerging risks associated with AIdriven cyber threats, including automated exploitation, adversarial machine learning, model poisoning, and largescale inference attacks. It further analyzes advanced defensive strategies such as AIenhanced threat detection, Zero Trust architectures, securebydesign engineering, and resilient model governance. Finally, the study evaluates the regulatory imperatives required to ensure safe and trustworthy AI deployment, emphasizing the need for comprehensive policy frameworks, transparency mandates, and coordinated national cybersecurity standards. The findings highlight those securing digital ecosystems in the age of AI demands a combined approach of technical innovation, organizational readiness, and proactive regulatory oversight.

Current developments in computing have shown the advantage of using one or more Graphic Processing Units (GPU) to boost the performance of many computationally intensive applications but there are still limits to these GPU-enhanced systems. The major factors that contribute to the limitations of GPU(s) for High Performance Computing (HPC) can be categorized as hardware and software oriented in nature. Understanding how these factors affect performance is essential to develop efficient and robust applications codes that employ one or more GPU devices as powerful co-processors for HPC computational modeling. The present work analyzes and understands the intrinsic interrelationship of both hardware and software categories on computational performance for single and multiple GPU-enhanced systems using a computationally intensive application that is representative of a large portion of challenges confronting modern HPC. The representative application uses unstructured finite element computations for transient composite resin infusion process flow modeling as the computational core, characteristics and results of which reflect many other HPC applications via the sparse matrix system used for the solution of linear system of equations. This work describes these various software and hardware factors and how they interact to affect performance of computationally intensive applications enabling more efficient development and porting of High Performance Computing applications that includes current, legacy, and future large scale computational modeling applications in various engineering and scientific disciplines.

This article systematically reviews the academic literature on the applications of artificial intelligence (AI) in cybersecurity, with a specific focus on generative techniques (generative AI or GenAI), such as large language models (LLMs). The analysis covers publications from 1993 to 2025 extracted from the IEEE Xplore Digital Library (1,647 publications) and SpringerLink (1,742 publications) databases, resulting in a collection of 3,389 documents.The litstudy tool was utilised for thematic mapping, automatic topic modelling, and n-gram analysis. The analysis shows an exponentially upward interest since 2022, particularly between 2023 and 2024, indicating rapidly growing interest in GenAI methods. Through topic modelling, the study identified the following key thematic areas: LLMs (734 publications; 21.88%), identified as the most dominant topic, blockchain and cyber attacks (394 publications; 11.74% each topic), generative coding for software (296 publications; 8.82%), smart energy and internet of things (278 publications; 8.29%), and malware (70 publications; 2.09%). The study revealed that GenAI and LLMs present a significant dual-use dilemma. Malicious actors increasingly leverage these methods for offensive purposes. Conversely, the same methods are actively developed to enhance cybersecurity defences. GenAI and LLMs are fundamentally reshaping the cybersecurity landscape, as evidenced by the visible growth in research interest. In light of the dual-use dilemma posed by GenAI, organisations should urgently consider investing in securing their sensitive data, and enhance staff capabilities in threat detection and response using GenAI-/LLM-driven methods.

Rational tensor product surfaces, (B~zier, NURBS, Hermite, polynomial, etc.) are rendered in real-time by uniform faceting. The described methods are modular and can be balanced for optimal implementation on different hardware plaOrorms. Discretization anomalies such as angularities, Math banding, cracking etc. are avoided by tessellating the surface patches and segmenting the trimming curves based on the view.

The paper addresses the design issue of fault-tolerant high-availability HPC clusters of missioncritical systems in the public-sector, such as energy grid operators and government data centers. The paper focuses on such architectural designs as redundant pathing, multi-layer failure, homogeneity in the firmware and risk managed cutovers to assist in achieving zero-downtime operations. The system performance was quantitatively evaluated by running a simulation-based experiment which incorporated Monte Carlo fault injection. Multi-layer failover with firmware homogeneity availability, MTTR, and recovery success was 99.97, 38 and 99.9 seconds respectively, whereas controlled cutovers also increased availability to 99.98. The workload was completed over 99.5 percent even with several faults. These findings point to the fact that properly designed HPC systems may be maintained, and used in handling large volumes of data and be cyberresilient in serving millions of people. The paper provides a methodical method of designing and assessing the high availability HPC systems that can be applied in the national interest activities.

This paper is a technical proposal to deploy the secure hyperconverged infrastructure (HCI) to transform the way the statewide government operates as well as safeguard sensitive information against cyberattacks. The case study concentrates on storage with PowerFlex, virtual machines that deal with compliance, network isolation, NTP / DNS integrity, and migration of a multi-petabyte of data. The City of Oklahoma 4 million resident in-service digital-first modernization demonstrated low latency (2.4 ms), high throughput (92 Gbps), 1,250,000 IOPS and almost constant uptime (99.98%). Migration of multipetabytes data at a rate of 92-96 TB/hour with complete data integrity. The security testing was blocking 93-100% of the simulated attacks such as ransomware, DoS, and VM escapes. The results suggest that a well-developed HCI can guarantee high performance, scale, and super security, and lower the management overhead and operational expenses. The article underscores the fact that the author developed and proved these systems, offering a nationwide applicable architecture of federal and state-based digital resiliency in the government IT infrastructure.

In this research study, the authors explore strategic processes that can be used to maintain security and compliance in contemporary data pipelines. It explores construction concerns, most important security practices, and obligatory structures that can alleviate the vulnerabilities across ingestion, transformation, and storage levels. The research interprets encryption, access-control and secret-management technologies as the ways of enhancing protection of data. It also highlights the fact that continuous monitoring and data lineage tracking along with regulatory compliance must be tracked according to GDPR and CCPA. The results highlight the need to establish effective security measures and end with effective recommendations that widen pipeline security, operational effectiveness, and compliance with regulations.

The present study investigates the GEKO turbulence model trained by a Feed Forward Neural Network for the improvement of gas turbine film cooling simulations. The key flow features from the GEKO simulation are used as the inputs for the neural network, and the turbulence kinetic energy source coefficient of the GEKO model serves as the output parameter. Both the inputs and the output are being updated by the adjoint optimization during each training cycle, known as a "design iteration". The training of GEKO model is successful on a very coarse mesh of the single-hole, jet-in-crossflow MIT Liner film cooling case. The trained GEKO solutions better match the LES and the experiments than the untrained GEKO on the wall heat fluxes. Plus, the trained GEKO model can be extended to another flow blowing ratio and cooling material of the MIT Liner, without any retraining. Furthermore, the same trained GEKO can even be applied to a multi-hole, film-cooled Perforated Plate case with complex geometry, where the trained GEKO considerably improves the plate's wall temperature predictions, better matching the LES and the measurements than the untrained GEKO model. The present automated GEKO training workflow, and the good generalizability of the trained GEKO model across different film cooling conditions and geometries, may help the gas turbine industry speed-up the film cooling design process for components such as combustor liners and turbine blades, saving a significant amount of computational time and resources.