The concept of cloud computing has revolutionized information processing across the world due to the services that are elastic, scalable and cost effective. Nevertheless, multi-tenancy brings in critical risks, such as, unauthorized lateral movement, data leakage by tenants, insider threats, privileged escalation and advanced cyberattacks. Existing perimeter security models fail to perform properly in dynamic, distributed and cloud environments, in which there is no clearly defined trust boundaries. In this paper, I suggest a Multi-Tenant Cloud Security Framework (MTCSF), which integrates Zero-Trust Architecture (ZTA) and AI-Based Anomaly Detection to develop an active and adaptable defense system. The framework provides the continuity of authentication, the regulation of least-privilege access, micro-segmentation, and encryption of the lifecycle data. An anomaly detection engine is a hybrid AI-based system of supervised deep learning and unsupervised clustering that determines the anomalies (insider abuses or anomalous patterns of resource access) in real time. The system will also utilize a Security Orchestration, Automation, and Response (SOAR) layer that will automatically start mitigation actions. The simulated multi-tenant cloud testbed with benchmark datasets (NSL-KDD and CIC-IDS 2021) of simulated threat scenarios such as DDoS, privilege abuse, and malicious file transfer were experimented. Findings indicate better detection and were lower than false positives with the existing cloud security systems. The most important key performance indicators are a detection rate of 98.21 percent, a false-positive rate of 1.79 percent, a shorter switching time to threat responses, and low latency overhead (less than 3 ms). The suggested MTCSF fulfills architectural and operational security vulnerabilities. The contributions involve: (1) a Multi-tenant access control Zero-trust enforcement model, (2) a hybrid AI abnormality detector to be incorporated into security processes, Detection and isolation of threats tailored to the homes occupied by tenants. This study has shown that trust-agnostic and intelligent control are beneficial to improve confidentiality, integrity, and availability of multi-tenant clouds and to satisfy regulatory compliance and tenant isolation needs. Optimized model scalability and increased cross-cloud sharing of threat intelligence will be optimized in future work

Cloud computing has reached the core of the digital ecosystems and provides high-availability and scale-able access to the services of computational services. Nevertheless, the active resource requirements, enforcement of service-level agreement (SLA), energy usage restrictions, and a variation in the cost-performance represent significant challenges to the cloud resource management. The conventional heuristic-based schedulers in clouds orchestrators like Kube and OpenStack fail to adjust to real-time workload variability, fractional resources, violation of SLA and inflated operational costs frequently occur. The adaptive, reward-driven decision-making properties of reinforcement learning (RL) have been discovered to be an interesting alternative to a heuristic policy. However, centralized RL has scalability and state observability scale in distributed clouds. As a way of overcoming these limitations, this study hypothesizes a Multi-Agent Reinforcement Learning (MARL) model of autonomous resource scheduling and optimization in cloud infrastructures. The offered structure suggests the use of decentralized intelligent agents executing a coordinated management of computational, storage, and network allocations. The agents are optimized to adapt local behaviors under environment feedback with the aid of cooperative communication to enhance global optimization. The system uses hybrid in reinforcement system consisting of Deep Q-Learning (DQN) and Proximal Policy Optimization (PPO) to trade between exploration, stability, and on-policy improvement. The goal of an adaptive reward model is to maximize important performance indicators (KPIs): SLA adherence, task completion latency, energy efficiency and balance resource utilization. The Cloud sim Plus experimental measurements were performed in a simulated multi-cluster environment and supported by extra features of container virtualization. The proposed version of the MARL scheduler showed a better throughput (18.7%), SLA violations reduction (27.4%) and energy consumption (14.3) than the baseline strategies like Round-Robin, Min-Min, and centralized DQN models. Convergence time was reduced and decision correlation under the workload bursts was improved due to the incorporation of the inter-agent communication. The research paper has a tremendous impact in the field by proposing a foresight, smart cloud orchestration framework that has the capability to adapt itself in the distributed large scale systems. Additionally, we offer an improvement in algorithms, performance analysis, system workflow optimization, and theoretical support of a cooperative reinforcement learning in clouds. The implication of these findings in the future is that the applicability of the research to edge and fog computing settings can be applied to Industry 5.0 automation

The dramatic increase in the number of heterogeneous computing architectures (HCAs) which include CPUs, GPUs, TPUs, FPGAs, and new neuromorphic processors has transformed the high-performance computing (HPC) and artificial intelligence (AI) workloads. As the complexity of architecture increases there is corresponding uncertainty on reliability and performance checks, particularly when systems are incorporated with different processing elements, interconnections, and memory levels. Modern non-deterministic or non-stochastic heterogeneous environments are associated with traditional reliability models that feel great necessity to model nonlinear interdependencies between components. The current paper proposes an innovative AI-based modeling framework that would be capable of measuring system reliability and performance indicators in the context of HCAs. The suggested solution merges Deep Neural Networks (DNNs), Bayesian Inference Models, and the strategies of Reinforcement Learning (RL) to forecast the increase of the system reliability, the performance bottlenecks, and the mean time between failures (MTBF) at the system level. To train AI models that can capture the multi-domain interactions of computation, communication, and thermal dynamics, a large-scale simulation dataset was created with synthetic benchmarks, actual workloads (SPEC ACCEL, MLPerf, HPCG), unlikely fault profiles, and injected fault profiles to have AI models that are trained to simulate interactions in large-scale systems. The approach puts the feature engineering of runtime telemetry (e.g., power, temperature, utilization) and hardware counters (e.g., instruction-level parallelism, cache miss rates) to construct predictive and adaptive reliability estimators. As opposed to the traditional models, our AI-based system advances its state dynamically with online learning methods of training, allowing its internal state to identify fault and self-optimize it. Findings suggest that the suggested AI system can make accurate predictions in both system performance degradation and reliability estimates with a success level of 96.8 and 94.2 per cent respectively when operating in a heterogeneous environment. Comparative research with classical models of reliability, i.e. Markov and Weibull-based models reveal high gains on adaptability, precision and generalization. Moreover, the optimization using reinforcement learning provided a 1525 percentage ratio in task scheduling performance expressed in a limited thermal and power budget. This article adds to the increasing overlap between system modeling based on AI and the optimization of heterogeneous computing, sets the stage of the introduction of new intelligent reliability management of data centers, autonomous computing systems, and HPC infrastructure

Real-time integration middleware sits at the core of how modern enterprises move data, trigger actions, and keep distributed systems in sync. Platforms like Apache Kafka, RabbitMQ, IBM MQ, and Apache Pulsar carry the weight of mission-critical operations—from payment processing and supply chain coordination to IoT data collection and distributed analytics. Yet, despite how much it depends on them, the methods used to benchmark their performance are still surprisingly immature. Most current ways of testing these systems—like checking the infrastructure, running application load tests, or assessments done by vendors—fail to capture the important behaviors that really affect their performance in real situations: how well they handle sudden spikes in traffic, how they deal with pressure, and if keeping messages safe slows things down too much. This article covers the theoretical basis for middleware performance evaluation, maps out where current practice falls short, and offers a detailed look at the Middleware-Aware Performance Benchmarking Framework (MAP-Bench). MAP-Bench combines a set of standard measurements, models for business workloads, tests for handling faults, and a more advanced way to evaluate architecture into one complete method. The practical value extends to platform selection, capacity planning, SLA design, and performance tuning. The goal is to move middleware benchmarking away from narrow load testing and toward something that actually reflects how these systems perform when it counts.

Customary methods of managing the workforce are often inefficient, relying on simple forecasting methods, manual scheduling, and reactive actions to manage the expected workload․ Overstaffing, understaffing, and inefficient resource allocation are common issues․ This article explores the opportunities created by artificial intelligence and machine learning for workforce optimization in service-oriented industries. The three technical sub-areas are predictive demand forecasting approaches, clever skill-based scheduling, and the related performance, service quality and employee well-being analysis. Demand forecasting includes the modeling of seasonality using time series methods such as Long Short-Term Memory networks (LSTM) and Autoregressive Integrated Moving Average (ARIMA) models as well as the prediction of surges when unexpected demand spikes occur․ The skill-based scheduling problem gives an overview of multi-dimensional human competency management frameworks and mixed integer linear programming optimization approaches that focus on productivity, compliance and ergonomics. The recent peer-reviewed literature supports that the optimization of the workforce in a sustainable manner depends on a combination of technological capabilities and organizational culture, leadership commitment, and digital preparedness. For organizations to maximize productivity and ensure the health of their employees, human-centered design must be at the crux of AI predictions. The article concludes that AI-driven workforce optimization represents a model shift from administrative labor management to a performance-oriented labor management capability․

Progressive delivery, a technique that has been in use for a long time in software engineering, gradually gets essential applications in ML and AI model deployment scenarios, where the risks of performance drift, ethical lapses, or bias amplification require cautious and measurable release strategies. This article discusses the use of gradual delivery concepts in ML models, which are mainly controlled, incremental rollouts regulated by well-defined Quality Key Performance Indicators (KPIs). Progressive delivery allows for staged exposure thus, a model is not immediately deployed to full production, but it is exposed only to limited user segments or traffic percentages from the very beginning and the real-time validation metrics such as precision, latency, fairness, and stability are used to make the decisions on promotion or rollback. Before they may get a chance to interact with the model, progressive delivery principles ensure that ML-based changes are tested, verified, and monitored for quality on a smaller scale and with the possibility of rapid rollback. Accordingly, the paper offers a well-thought-out approach that mixes CI/CD principles with ML observability frameworks to build a smart, feedback-driven delivery pipeline. The paper presents a case study that shows how companies can put this framework into practice, managing the trade-off between innovation speed and performance assurance. The experimental evidence conveys that the suggested model deployment pipeline not only encourages implementation trust but also leads to the performance anomalies' early and quick recovery, eventually raising model robustness and business impact to higher levels. Ultimately, the paper conceptualizes progressive delivery as a core capability for the deployment of responsible AI through this synthesis of engineering rigor and statistical governance, thus enabling a smooth transition from experimentation to production at scale.

Digital infrastructures are growing much faster, and as such, have increased the complexity and volumes of cyber threats. Conventional cyber security programs tend to fail to protect and counter complex attacks on the fly. This essay discusses how to adopt the use of graph-based Artificial Intelligence (AI) frameworks in detecting and responding to cyber threats on a grand scale. The graph-based AI relies on the existing relationships between the entities of a network, which results in the ability to have a holistic view of the pattern of attack. Our plan is to introduce a multiple-layer system that comprises graph neural networks (GNNs), anomaly detecting algorithms, and autonomous response templates that would help to optimize cybersecurity systems. We have a scalable, real-time detection, and proactive threat mitigation approach to things. The experimental findings reveal that they have a great deal better detection accuracy and response latency than traditional signature-based systems. The results highlight the promise of graph-based AI in enhancing cyber defenses and it offers a roadmap on how to introduce autonomous security systems in a dynamic network environment

Modern multi-core and many-core processors expose massive parallelism, deep cache hierarchies, and non-uniform memory effects that make manual performance prediction increasingly fragile. This paper proposes an AI-driven analytics framework for predictive performance modeling that unifies workload characterization, hardware telemetry ingestion, and hybrid learning. We construct multi-scale feature views from hardware performance counters, memory/IO traces, and compiler IR to capture compute intensity, synchronization pressure, NUMA locality, and cache/branch behavior. A two-stage learner couples fast analytical baselines (e.g., roofline and queueing approximations) with machine-learned residuals, where gradient boosting and graph neural networks model code data topology interactions across CPU/GPU tiles. To generalize across architectures, we employ transfer learning and meta-features describing microarchitectural knobs (core count, cache geometry, interconnect, DVFS states) and use domain adaptation to reduce drift when porting workloads. Online updates with lightweight Bayesian last-layer adaptation provide calibrated uncertainty for “what-if” queries core pinning, memory placement, and DVFS policies while SHAP-style attributions expose bottlenecks for developer feedback and autotuners. The framework outputs latency/throughput predictions, energy performance trade-offs, and scaling curves under contention, enabling proactive scheduling and configuration search in CI and runtime orchestration. Experiments span stencil codes, graph analytics, and ML inference pipelines, demonstrating robust accuracy under workload and platform shifts. By blending interpretable theory with data-centric learning, the approach delivers portable, explainable performance predictions suitable for cloud, edge, and HPC settings

Hybrid clouds spanning on-premises; edge; and multiple public providers demand orchestration that balances conflicting goals: low latency and high throughput; energy and carbon reduction; and rigorous security. This work frames resource provisioning; workload placement; and autoscaling as a multi-objective optimization problem under uncertainty. We model application components (VMs/containers/functions) with performance SLOs; security postures; and energy carbon profiles tied to time-varying grid intensity and datacenter PUE. Objectives minimize end-to-end latency and SLO violations; total energy and carbon cost; and security risk (e.g.; exposure time; vulnerability risk scores; data-in-motion footprint); while respecting budget; data-sovereignty; and trust-zone constraints. Solution strategies combine exact formulations (MILP for small horizons) with scalable heuristics and metaheuristics (e.g.; NSGA-II/MOEA-D) to approximate the Pareto frontier; and a model-predictive layer that adapts to demand; spot price volatility; and carbon signals. Robust and risk-aware variants incorporate chance constraints and CVaR to hedge against workload and failure uncertainty. We integrate zero-trust placement rules; encryption overhead; and privacy controls (federated learning/differential privacy) as first-class constraints; and co-optimize network paths to limit cross-cloud data exfiltration risk. A learning-augmented scheduler uses surrogate models for queueing delays and power draw to accelerate search at runtime. The resulting policies enable operators to trade milliseconds of tail latency for double-digit energy/carbon savings or reduced attack surface; making decisions transparent via Pareto sets and what-if analysis. The framework generalizes across microservices; data pipelines; and AI inference; and can plug into Kubernetes-centric control planes

The growing use of artificial intelligence in healthcare, employment, and business has raised significant concerns regarding safety, ethics, and societal impact. While prior work offers ethical guidelines and technical solutions independently, a gap remains between ethical principles and practical system design. This study presents a unified framework that integrates ethical values, technical safeguards, and governance mechanisms to support safe and human-centered artificial intelligence. The framework maps principles such as fairness, accountability, transparency, and non-maleficence to engineering practices including safe reward design, explainable models, robustness testing, and privacy-preserving techniques. Governance and regulatory alignment are incorporated as continuous components of the AI lifecycle. Use-case analyses demonstrate how ethical objectives can be operationalized across real-world domains. The results emphasize that AI safety must be treated as a socio-technical process combining technical alignment with institutional oversight. This work contributes a practical approach for translating ethical commitments into trustworthy and resilient AI systems.