edge computing

MixInfoFold: A Method For Iteratively Generating Better Sequences

Abstract Current AlphaFold3-based protein structure prediction has reached remarkable levels, but sequences that can be applied in practice are mostly generated using physics-based methods. Here, we introduce a graph representation-based protein sequence generation method: MixInfoFold which incorporates new noise addition and feature extraction methods, as well as our designed encoder-decoder called MixInfo. The noise addition method adds Gaussian noise positively correlated with the sequence length at the end of the main-chain, which enhances the model’s ability to understand the dynamic changes in protein structure. The feature extraction method includes a new feature calculated by assessing the area and distance of the main-chain atoms. The MixInfo iteratively computes the relationships between edge and node features, enabling a deeper exploration of the implicit sequence information within protein structures. Compared to the best model, our model achieves an improvement of 0.80%, 0.89%, and 3.28%in sequence recovery rates on the CATH4.2, CATH4.3, and Modelfinal datasets, respectively. Additionally, our model generates sequences faster than other methods, and generates sequences closely resemble natural proteins, indicating the model’s feasibility and potential value in practical applications.

Hardware-accelerated network slicing in multi-tenant 5G networks and beyond

Fifth-Generation Mobile Networks (5G), as the most recent generation of networks, presentsdiverging requirements imposed by different vertical industries and mainly by Mobile NetworkOperator (MNO)s which aim to reduce both Capital Expenditures (CAPEX) and OperationalExpenditures (OPEX) using novel technologies based on Mobile Edge Computing(MEC), Network Functions Virtualisation (NFV) and multi-tenancy. These technologies allowto share physical infrastructures between different MNOs. However, this novel paradigmwill incur a downgrade system performance with a direct impact in the competitiveness andproductivity achieved. In order to adapt the current networking infrastructures to these heterogeneous5G scenarios, an evolution in the different components of the network is required,adapting current data planes to become more complex to cope with the multi-tenant 5G traffic.Also, the demanding requirements imposed by the generic 5G services, Enhanced MobileBroadband (eMBB), Massive Machine Type Communications (mMTC) and Ultra-Reliableand Low-Latency Communications (URLLC) associated with different use cases, imply thestudying of new network slicing solutions to provide network isolation support betweenMNOs who share the same physical infrastructure.To address the need that currently exists in network slicing implementations in the edgeto-core network segment of a 5G multi-tenant architecture, this research has designed andprototyped a hardware-based 5G data plane. This data plane provides the ability to control 5Gmulti-tenant traffic and also contains network slicing capabilities which allow the isolation ofthe traffic from different MNOs while using the same physical machine. The network slicingmechanisms achievable with this programmable data plane provide support for multiplecommunication services and also a mechanism to programme the user and the control planes. It allows the creation of 5G slices composed of a collection of Network Function (NF)s forspecific use cases. In the control plane, an Application Programming Interface (API), whichprovides full control of the network data path and the network slicing solution, has beenimplemented.This prototype has been applied to different use cases, such as security, Ultra-High-Definition (UHD) video streaming and URLLC, where the feasibility of this solution hasbeen tested. The security use case prototyped is based on a 5G firewall responsible for theprotection of the edge and core network segments of a 5G scenario, where large quantitiesof network traffic should be processed simultaneously. These considerable quantities oftraffic received by the 5G networks will incur several security problems, such as massiveDistributed Denial of Service (DDoS) attacks, which will have to be mitigated to protect thenetwork components, and therefore avoiding network disruption. The UHD video streaminguse case prototyped includes the generic eMBB service proposed by 5G communications,requiring high data rates across a wide coverage area, complemented by moderate latency.And finally, the network slicing solution has been tested with a URLLC use case, where anovel hardware-based queuing algorithm has been implemented to guarantee End-to-End(E2E) latency in critical 5G communications. These solutions have been empirically validatedin the edge-to-core network segment of a 5G multi-tenant architecture fulfilling the strict5G Key Performance Indicators (KPI)s. Furthermore, the impact of this research has beenapplied in a real testbed in the context of the H2020 5G-PPP Phase II SliceNet project.

An experimental study of edge deployment strategies for cryptocurrency analytics workloads

DDMS: Dynamic Duty Management via Offline-First Geo-Spatial Constraint Orchestration

Civil security and law enforcement agencies regularly manage field deployments at scales exceeding tens of thousands of uniformed personnel. In conventional field operations, duty scheduling and redeployment remain heavily manual; static paper rosters are drafted days in advance and patched reactively across ad-hoc voice or radio channels. When operational disruptions occur (such as severe weather halts, route obstructions, spontaneous crowd surges, or emergent personnel unavailability), the latency between the operational plan and physical ground truth causes unverified attendance, multi-hour redeployment lags, and irrecoverable after-action data. The Dynamic Duty Management System (DDMS) addresses these operational challenges by modeling the operational theater as a continuously updated multipartite graph. Personnel, posts, route corridors, incidents, and command structures form vertices, while assignments, spatial adjacencies, capabilities, and hierarchical command authorities form edges. Upon an operational disturbance, an incremental solver isolates the affected local subgraph and computes optimal duty adjustments under a formal hierarchical relaxation mechanism when joint feasibility is initially empty. Duty deltas synchronize with edge devices through an offline-first causal protocol that handles extended disconnections and forks concurrent writes for supervisor reconciliation. We evaluate DDMS using operational telemetry from a 45-day multi-agency high-altitude deployment coordinating 13,000 to 21,000 personnel daily across complex terrain, alongside Kanwar Yatra pilgrimage deployments across Bareilly Range and Muzaffarnagar managing over 10,000 to 14,000 personnel across hundreds of sub-sectors and corridors. Reported empirical outcomes demonstrate a reduction in sector roster preparation from approximately 6 h to 25 min, median incident response latency from 29 min to 8 min, and unverified attendance from approximately 14% to 1.2%, while preserving the constitutional chain of command.

PMDS: Arquitectura Middleware en el Borde para la Sincronización Neuro-Semántica y Mitigación de Degradación Sintética en Modelos de Frontera

El Protocolo de Mitigación de Degradación Sintética (PMDS), desarrollado por A.T. Corporación Editorial, es una arquitectura de gobierno in situ en el borde (Edge Computing) diseñada para mitigar la deriva semántica, contener la entropía probabilística y eliminar alucinaciones en modelos de lenguaje de gran escala (LLMs) de frontera durante el runtime. Frente a las limitaciones de los ajustes hiperparamétricos estáticos y los alineamientos por reentrenamiento previo, el PMDS actúa como un middleware determinista que orquesta cinco subsistemas modulares interconectados: SS-Protocol (Protocolo de Sincronización Semántica): Inyección contextual acotada temporalmente ( Δ t) para la estabilización de vectores de sesión. FHDC (Filtro Hermenéutico de Doble Ciego): Verificación bidireccional mediante descomposición ontológica (Capa A) y auditoría contra matrices de dominio normativo (Capa B). SNL (Soberanía Neuro-Literaria): Preservación de varianza léxica mediante la modulación dinámica de penalizaciones de frecuencia y presencia, previniendo el aplanamiento estilístico por RLHF/DPO. SNS (Sincronización Neuro-Semántica): Restricción sintáctica que limita el muestreo probabilístico, eliminando clichés y estructuras repetitivas. Arquitectura Taxonómica GEO ATG: Estructuración de metadatos semánticos para optimizar la indexación en motores generativos. El marco operacional de la Regla de Contexto 75/25 como baseline estructural, reservando el 75% del espacio contextual para anclaje sintáctico e invariantes de dominio, y destinando un 25% a la permutación deductiva del modelo. En evaluaciones empíricas bajo un protocolo A/B automatizado (N=500), el middleware demostró una reducción del 28.5% en la latencia de inferencia y una supresión del 91.4% en desviaciones semánticas/alucinaciones, operando con impacto nulo (0 MB) en el consumo de VRAM adicional del servidor central. Metadatos de Publicación y Licenciamiento: Autor: Alexander Torres (CEO & Director Editorial, A.T. Corporación Editorial) ORCID: 0009-0008-6832-3814 Licencia de Documentación y Preprint: Creative Commons Attribution 4.0 International (CC BY 4.0) Licencia de Código e Infraestructura: Apache License 2.0 Palabras Clave (Keywords): PMDS, Middleware, Edge Computing, Sincronización Neuro-Semántica, Generative Engine Optimization (GEO), Model Collapse, Entropía Léxica, LLM Governance, Science Open Data.

Serverless edge intelligence for SLA-aware function placement and resource slicing in wireless networks

Wireless edge networks must support delay-sensitive services under fluctuating traffic, interference, heterogeneous SLA priorities, and limited radio, computing, and memory resources. This paper proposes Serverless-PRONTO, an SLA-aware orchestration framework that jointly controls serverless function placement, task offloading, bandwidth allocation, CPU slicing, and warm-instance management. The system model captures uplink and downlink transmission, queueing, execution, cold-start initialization, and memory occupied by retained function instances. To solve the resulting dynamic, partially observable, mixed discrete-continuous problem, Serverless-PRONTO combines a physics-aware sparse graph encoder with multi-agent TD3 under centralized training and decentralized execution. The encoder represents inter-node coupling through channel, interference, distance, queue, resource, function-demand, and cold-start features, while top- \(\:K\) attention limits signaling. An SLA-risk mechanism prioritizes requests according to urgency, queue state, service class, and cold-start probability. A sequential feasibility projection converts raw actor outputs into valid placement, bandwidth, CPU, and memory decisions. Simulations against four serverless and edge-orchestration baselines under varying traffic loads and network sizes show higher SLA satisfaction, lower end-to-end delay and cold-start ratio, and more stable scalability, demonstrating the benefit of jointly coordinating wireless resources and serverless runtime states.

Cumulative Intraoperative Hypothermic Burden, Transfusion, and Estimated Blood Loss: A Retrospective Cohort Study

Analysis pipeline for a retrospective cohort study of cumulative intraoperative hypothermic burden, intraoperative red cell transfusion, and estimated blood loss in 2,567 adults undergoing non-cardiac surgery, using the open VitalDB perioperative database. Scripts run in numeric order: 01-14 reproduce the originally submitted analysis, and 15-23 produce the first revision, including the edge-trimmed thermal exposure metric, threshold and dose analyses, and the table and figure builders. The repository carries code only; the manuscript, cover letters and peer-review correspondence are not distributed. VitalDB source data are openly available at https://vitaldb.net and are not redistributed here. This record accompanies the first revision submitted to BMC Anesthesiology. The archive contains the repository in full: scripts/, README.md, requirements.txt, .zenodo.json and .gitignore. Method note. Core temperature was previously integrated across the whole anaesthesia window, which counts probe equilibration at insertion and probe withdrawal at emergence as hypothermia, implying an implausible cohort nadir of 33.30 °C. Exposures are now computed after excluding the first and last 5 minutes of each case's monitored window. Cumulative burden is essentially unchanged (Pearson r=1.00, Spearman rho=0.99); the nadir corrects to 35.04 °C. Scripts run in numeric order. 01-14 reproduce the originally submitted analysis; 15-22 produce the revision. The manuscript, cover letters and peer-review correspondence are not part of this distribution. VitalDB source data are openly available at https://vitaldb.net and are not redistributed here.

EcoFair: Energy-efficient inference routing for edge AI under data degradation

علم الرسم الذكي

العنوان التقني: منظومة المعالجة الطرفية البصرية والتوليد التفاعلي للأنماط الرسومية اللامركزية 1. النطاق التقني ومجال الاختراع (Technical Field): تتعلق المنظومة الحالية بأنظمة الحوسبة الطرفية المستقلة (Edge Computing Frameworks)، وتحديداً المعماريات المدمجة للذكاء الاصطناعي المحلي المخصص لتوليد، وتحليل، ومحاكاة البيانات البصرية والهندسية ثنائية الأبعاد، دون الاعتماد على شبكات الاتصال السحابية الخارجية، مع آليات تكامل مادية لتحويل المخرجات الرقمية إلى وسائط ملموسة. 2. الخلفية التقنية والهدف الابتكاري (Background & Objective): تعاني المنظومة التقليدية لمعالجة الرسوميات الرقمية من الارتباط الدائم بالبنى التحتية السحابية لتعويض النقص في قواعد البيانات التوليدية، فضلاً عن غياب آليات التدريب التراكمي الموجه محلياً. يهدف الاختراع الحالي إلى تجاوز ذلك عبر دمج وحدة معالجة محلية مدعومة بمكتبات بيانات معرفية شاملة، لتوفير بيئة تفاعلية لا مركزية تقوم بتحليل مدخلات المستخدم الحركية، واستقراء أنماط الأداء عبر محرك استدلالي محلي، وتقديم توجيهات بصرية هندسية متقطعة لتعزيز كفاءة التفاعل البشري الرقمي، مع توفير قناة إخراج مادية متزامنة. 3. التوصيف الهيكلي والوظيفي للوحدات (Architectural Embodiments): وحدة الاستشعار والمعالجة الطرفية (Edge Processing & Input Subsystem): تتألف من منصة عرض تفاعلية مدعومة ببطارية داخلية قابلة لإعادة الشحن، مرتبطة بأداة إدخال دقيقة (Digital Stylus) مزودة بمصفوفة استشعار مزدوجة لقياس المتغيرات الفيزيائية (الضغط الديناميكي وزاوية الميل الحركي)، مما يضمن محاكاة دقيقة للاحتكاك السطحي والملمس الرقمي. الذاكرة المحلية ونماذج التعلم الذاتي (Local Neural Engine & Database): تحتوي على مستودع بيانات ضخم ومحلي بالكامل مخزن مسبقاً يضم الأنماط البصرية وقواعد المعرفة الفنية والهندسية. تعتمد الوحدة على خوارزميات التعلم التراكمي (Cumulative Edge Learning) لتطوير خيارات النظام بناءً على الأنماط السلوكية للمستخدم وسجل محاولاته السابقة محلياً. محرك الإرشاد التفاعلي والمعرفي (Cognitive Guidance Engine): نظام برمجي استدلالي مبرمج لتفسير الاستعلامات النصية أو الصوتية المتنوعة، وتوليد مسارات هندسية وبصرية إرشادية (خطوط إرشادية مفرقة ديناميكية) على الشاشة، تتيح للمستخدم تتبعها وإتمامها، يليه نظام مقارنة وتحليل آلي للمخرجات لتقديم تقييم فوري وتصحيح ذاتي. وحدة الإخراج والمزامنة المادية (Hardware-to-Print Synchronization Unit): وحدة ربط مباشر متكاملة مع طابعة دقيقة، تترجم البيانات اللونية والرقمية المعالجة على الشاشة إلى مخرجات ورقية ملموسة مطابقة للأصل الرقمي بدقة متناهية.

Based on Cognitive Maps: A Distributed Knowledge Reasoning System

This paper proposes a novel approach to distributed knowledge reasoning by leveraging cognitive maps. The core idea is to represent knowledge as a structured graph, where nodes represent concepts and edges represent relationships. This graph representation facilitates efficient and reliable reasoning through the application of graph algorithms. Furthermore, we integrate distributed computing techniques to enable knowledge sharing and updates across a distributed system. The system's architecture is designed to enhance both the speed and interpretability of knowledge inference. We demonstrate the potential of this approach in a distributed environment, aiming for a more robust and scalable solution for knowledge reasoning tasks. The key contributions lie in the structured representation of knowledge using cognitive maps and the utilization of graph algorithms for reasoning, leading to improved efficiency and explainability compared to traditional methods.

PlanarBench: Evaluating LLM Spatial Reasoning via Planar Graph Drawing

Existing LLM graph benchmarks typically ask models to answer graph-theoreticquestions or compute symbolic solutions rather than construct spatial layouts.Within-task difficulty is also primarily stratified by vertex count. However, existingresearch also suggests that task difficulty is more closely related to the number ofconstraints imposed by the edges than to the number of vertices being arranged.We introduce PlanarBench, a benchmark that asks models to produce crossing-free ASCII drawings of planar graphs given only an edge list. Across 91 modelconfigurations and 199 non-isomorphic connected planar graphs with 2–7 vertices,edge count is more strongly associated with mean task score than vertex count(r = −0.85 versus r = −0.47) and remains strongly associated after controllingfor vertex count (rpartial = −0.80). PlanarBench provides a controlled settingfor separating these two difficulty axes. In addition, neither drawing area nortotal response length demonstrated a meaningful correlation with score, which isevidence against a simple output-size explanation. Performance varies widely: thebest model scores 159.5 out of 199, most models below 30B parameters scoreunder 25, and substantial failures remain among frontier systems.

Artificial Intelligence and Edge Computing for Sustainable Smart Water-Safety Monitoring in Low-Resource Communities: A Critical Review

Poor water quality monitoring and delayed responses to pollution remain major challenges in low-resource areas. Traditional methods of monitoring water and wastewater resources are ineffective because they take a long time to report contamination. Therefore, this critical review aims to examine how a combination of artificial intelligence (AI) and edge computing can comprehend decentralised, real-time water quality monitoring, even in areas with limited infrastructure and internet, constrained maintenance capacity, and shortages of skilled personnel. To our knowledge, this study is a first-of-its-kind integrated framework that showcases edge AI architectures and refers to specific operational, societal, and infrastructural limitations in the water, sanitation, and hygiene (WASH) sector. The synthesis clearly shows that the implementation of edge AI techniques has the capability to improve global water quality through immediate pollution detection, disaster forecasting, and automatic filter or alarm response without the need for cloud infrastructure. The examples given from developing countries support this statement by demonstrating that the technologies are low-cost and implementable in the long term. The problem of sensor calibration, data quality, and energy efficiency was identified as the most important implementation challenge. There is enough evidence of pilot-scale tests, but long-term validation of the technology in field trials is needed. The authors also present future research directions, such as the integration of AI, edge computing, machine learning, and IoT, and open-source edge frameworks. Edge AI systems provide promising avenues for decentralised water safety surveillance in low-resource communities in real time and can support the sustainable development goals of the United Nations through the realisation of clean water and sanitation for all.