With the rapid advancement of fog-cloud computing, task offloading and workflow scheduling have become pivotal in determining system performance and cost efficiency. To address the inherent complexity of this heterogeneous environment, a novel hybrid optimization strategy is introduced, integrating the Improved Particle Swarm Optimization (IPSO) algorithm, enhanced by a linearly decreasing inertia weight, with the Grey Wolf Optimization (GWO) algorithm. This hybridization is not merely a combination but a synergistic fusion, wherein the inertia weight adapts dynamically throughout the optimization process. Such adaptation ensures a balanced trade-off between exploration and exploitation, thereby mitigating the risk of premature convergence commonly observed in standard PSO. To assess the effectiveness of the proposed IPSO-GWO algorithm, extensive simulations were carried out using the FogWorkflowSim framework—an environment specifically developed to capture the complexities of workflow execution within fog-cloud architectures. Our evaluation encompasses a range of real-world scientific workflows, scaling up to 1000 tasks, and benchmarks the performance against PSO, GWO, IPSO, and the Gravitational Search Algorithm (GSA). The Analysis of Variance (ANOVA) is employed to substantiate the results. The experimental results reveal that the proposed IPSO-GWO approach consistently outperforms existing baseline methods across key performance metrics, including total cost, average energy consumption, and overall workflow execution time (makespan) in most scenarios, with average reductions of up to 26.14% in makespan, 37.73% in energy consumption, and 12.52% in total cost Beyond algorithmic innovation, this study contributes to a deeper understanding of workflow optimization dynamics in distributed fog-cloud systems, paving the way for more intelligent and adaptive task scheduling mechanisms in future computing paradigms.

Aseptic manufacturing depends on reliable equipment to maintain throughput and protect patients. This study presents a practical, reproducible maintenance engineering method for proactively replacing aging parts before failure; regulatory references are included only as bounded implementation context for execution inside governed (site validated) systems. This study presents a practical, reproducible method for proactively replacing aging parts before failure. The method combines a simple weighted Health Index (HI) that summarizes condition signals such as vibration, temperature, flow, pressure decay, and motor torque with established survival methods (Weibull and Kaplan-Meier) to estimate remaining useful life (RUL). These estimates are converted into clear Green-Yellow-Red maintenance actions and illustrated with a conceptual SAP PM execution workflow for work orders, spare reservation, and traceable recordkeeping under site governanceThe workflow is demonstrated on a vial washer, an upstream step that influences vial cleanliness and particulate control prior to depyrogenation, focusing on components including conveyors, pumps, bearings, seals, spray nozzles, and heaters. Using a simulation dataset to illustrate the full end to end analytics and decision workflow, Weibull fits with shape factor ß in the range of about 1.8 to 2 captured wears out behavior, and Kaplan-Meier pump survival at 24 months was 0.58 (95% CI 0.50 to 0.66). Applying HI together with RUL shifted interventions from unplanned breakdowns to planned stops. Relative to a reactive baseline under the stated assumptions, the scenario results showed 62.5% lower downtime, 84% lower combined maintenance and rejection costs, and 50% fewer batch rejects, with robustness demonstrated through sensitivity tests and 1,000 Monte Carlo runs.Overall, the contribution is a reproducible maintenance engineering workflow (HI + survival-based RUL + decision matrix) with a conceptual CMMS/SAP PM execution mapping for traceable work order initiation when implemented under site governance. The approach is adaptable to existing SCADA and SAP PM infrastructures following validation within the site quality system, with early benefit expected when prioritizing product quality critical components such as seals and spray nozzles.

The growing global shortage of skilled surgeons underscores the need for intelligent, assistive technologies in the operating room. To address this challenge, we introduce ImitateCholec, a publicly available dataset specifically designed to advance autonomous robotic systems during the critical clipping and cutting phase of laparoscopic cholecystectomy. The dataset comprises over 18,000 demonstrations from 34 ex vivo porcine cholecystectomies, totaling approximately 20 hours of data. Each clipping and cutting phase recorded in the dataset is segmented into 17 distinct surgical tasks. ImitateCholec uniquely integrates endoscopic videos captured from multiple camera perspectives with comprehensive kinematic data acquired through the da Vinci Research Kit. Both optimal demonstration executions and recovery maneuvers were systematically recorded, enabling the training of imitation learning models capable of robustly addressing real-world surgical variability. Primarily, ImitateCholec facilitates imitation learning for long-horizon surgical workflow execution, significantly advancing the development of autonomous robotic systems toward achieving phase-level autonomy and, ultimately, full procedural autonomy. Additional supported applications include surgical workflow modeling, error recognition, and surgical tool pose estimation.

Biological experiments often require a series of precisely timed operations, and small variations in treatment can result in inconsistent or biased results. To handle multiple samples in parallel with precise temporal resolution, experimentalists may stagger treatments by initiating the workflow of one sample during the wait or incubation time of another. However, as the number of samples processed in parallel and the number of operations increase, it becomes increasingly difficult to identify and execute valid treatment regimens that permit the handling of each sample. To address this, I developed StaggR, an interactive web application that calculates and visualizes compatible staggering intervals for parallelized execution of identical processing workflows. This tool provides a user-friendly interface for defining protocol operations, durations, and wait times. It can automatically calculate the shortest possible conflict-free interval for initiating sample treatments, or allow users to simulate specific intervals to explore potential treatment regimens or bottlenecks. Using StaggR, users of any experience level can rapidly generate complete, color-coded experimental schedules, visualize these workflows in an easy-to-read chart, and execute them using a built-in timer displaying a treatment schedule with live updates. The experimental designs can be saved, shared, and re-imported, ensuring full reproducibility and user control. The application of StaggR is expected to expedite the design and throughput of complex experimental workflows while maximizing reproducibility.

Whole-brain radiotherapy (WBRT) is a common treatment due to its simplicity and effectiveness. While automated Field-in-Field (Auto-FiF) functions assist WBRT planning in modern treatment planning systems, it still requires manual approaches for optimal plan generation including patient-specific hyperparameters definition and plan refinement based on quality feedback. This study introduces an automated WBRT planning pipeline that integrates a deep learning (DL) Hyperparameter Prediction model for patient-specific parameter generation and a large-language model (LLM)-based conversational interface for interactive plan refinement. The Hyperparameter Prediction module was trained on 55 WBRT cases using geometric features of clinical target volume (CTV) and organs at risk (OARs) to determine optimal Auto-FiF settings in RayStation treatment planning system. Plans were generated under predicted hyperparameters. For cases in which the generated plan was suboptimal, quality feedback via voice input was captured by a Conversation module, transcribed using Whisper, and interpreted by GPT-4o to adjust planning settings. Plan quality was evaluated in 15 independent cases using clinical metrics and expert review, and model explainability was supported through analysis of feature importance. Fourteen of 15 DL-generated plans were clinically acceptable. Normalized to identical CTV D95% as the clinical plans, the DL-generated and clinical plans showed no statistically significant differences in doses to the eyes, lenses, or CTV dose metrics D1% and D99%. The DL-based planning required under 1 minute of computation and achieved total workflow execution in approximately 7 minutes with a single mouse click, compared to 15 minutes for manual planning. In cases requiring adjustment, the Conversational module successfully improved dose conformity and hotspot reduction.

The proliferation of heterogeneous Graphics Processing Unit (GPU) clusters has introduced unprecedented computational capabilities for workflow execution across diverse scientific and industrial domains. However, the inherent heterogeneity of GPU resources, coupled with dynamic workload characteristics and complex workflow dependencies, presents substantial challenges for efficient scheduling. Traditional heuristic-based scheduling algorithms such as Heterogeneous Earliest Finish Time (HEFT) and First-In-First-Out with Duplication and Earliest Finish Time (FIFO-DEFT) often fail to adapt to rapidly changing cluster states and evolving workload patterns. This paper proposes an adaptive workflow scheduling framework leveraging Deep Reinforcement Learning (DRL) to intelligently allocate workflow tasks to heterogeneous GPU resources. The proposed approach employs a Deep Q-Network (DQN) architecture integrated with prioritized experience replay to learn optimal scheduling policies through continuous interaction with the cluster environment. The framework models workflow scheduling as a Markov Decision Process (MDP) where the agent learns to minimize makespan, maximize resource utilization, and maintain quality-of-service guarantees. Extensive experimental evaluations demonstrate that the DRL-based scheduler achieves significant performance improvements compared to baseline algorithms including HEFT, FIFO-DEFT, and other state-of-the-art schedulers. The proposed method exhibits superior adaptability to varying cluster configurations and workflow characteristics, maintaining robust performance across diverse execution scenarios while reducing average makespan and improving scheduling length ratio metrics.

Hydrological models play a critical role in advancing environmental modeling. They are particularly significant in contexts requiring short-term decision-making, where real-time simulation capabilities support timely and informed actions. The advancement of Internet of Things (IoT) technology has provided new opportunities for enhancing real-time hydrological modeling. However, most widely used hydrological models were originally designed as desktop applications with process-oriented execution workflows, which hinder fine-grained state access and standardized integration with IoT systems, thereby limiting their suitability for real-time, observation-driven modeling scenarios. This paper proposes a method for describing hydrological model components and data using a standard IoT conceptual model. By establishing a generic object-oriented framework, we integrate hydrological models with IoT systems, systematically representing model elements and data while mapping them to the Open Geospatial Consortium (OGC) SensorThings API conceptual model. This approach enables real-time, observation-driven hydrological modeling and facilitates fine-grained state acquisition. Finally, we developed a prototype system based on the Storm Water Management Model (SWMM) and validated the feasibility of our methodology through case studies.

Abstract The exponential growth of bioinformatics and drug discovery data necessitates computational frameworks capable of efficiently managing large-scale, heterogeneous, and continuously evolving datasets. This study presents a hybrid deep learning framework built upon a microservices-based architecture to enhance scalability, modularity, and efficiency in bioinformatics workflows. The proposed system integrates generative deep learning—specifically, a variational autoencoder (VAE)—to tackle challenges in bioinformatics. Through a microservices-driven design, the framework enables modular deployment, semantic interoperability, and scalable integration of generative models into bioinformatics pipelines. The architecture is model-agnostic and supports flexible validation by encapsulating cheminformatics evaluations, as independent microservices. This allows automated updates and seamless incorporation of new evaluation metrics without disrupting the overall pipeline. Rather than replicating existing cheminformatics validation efforts, the framework provides an extensible foundation for integrating, reusing, and scaling such studies across large datasets in bioinformatics and drug discovery. In contrast to traditional monolithic architectures, the microservices paradigm supports independent deployment, optimization, and scaling of deep learning components—such as molecule generation, toxicity prediction, and pharmacokinetics analysis—while maintaining uninterrupted workflow execution. This design ensures real-time data processing, interoperability across distributed systems, and reduced computational overhead. Experiments conducted on benchmark datasets from DrugBank, ChEMBL, and TCGA demonstrate superior performance compared to monolithic baselines in predictive accuracy, biomarker identification, and computational efficiency. These results underscore the potential of microservices as a software engineering paradigm for advancing bioinformatics and accelerating the drug discovery process.

Artificial intelligence has become foundational to enterprise operations across sectors—from customer engagement and decision support to risk management and real-time automation. While adoption delivers measurable operational gains, it also introduces new governance challenges stemming from opacity, bias, and unclear accountability. This article introduces the Ethical Lifecycle Governance Framework (ELGF), a unified implementation model for ethical, transparent, and explainable AI across organizational environments. ELGF incorporates five execution pillars: transparency, human oversight, validation, continuous monitoring, and accountability, mapped directly onto the AI system lifecycle from data intake through retirement. The model provides organizations with actionable mechanisms rather than abstract policy recommendations, ensuring measurable accountability across program stakeholders. Because the framework is industry-agnostic, organizations can adopt it without changing their existing architecture or operational models. When implemented, ELGF reduces regulatory exposure, enhances stakeholder confidence, and accelerates innovation by enabling responsible deployment of high-impact AI systems. The combined outcome is operational growth aligned with ethical principles rather than in conflict with them. Executive Summary AI now influences access to financial services, healthcare eligibility, employment decisions, and essential digital services, making governance no longer optional. The proposed Ethical Lifecycle Governance Framework operationalizes responsible AI deployment by linking policy to execution workflows. Rather than treating ethics as a compliance checkpoint, ELGF embeds governance into every phase of the AI lifecycle. This approach ensures that fairness, transparency, interpretability, and accountability are continuously tested, validated, and traceable. By institutionalizing these controls, organizations not only reduce risk but also scale AI initiatives confidently and sustainably.

Artificial Intelligence Integration in Financial Systems and Enterprise Automation: Technical Architecture and Implementation Frameworks

Scientific progress relies on the generation, validation, and reuse of research data, yet standard practices and cultural, legal, and technological challenges have long limited data sharing. In the 21st century, growing volumes of data, higher transparency requirements, and concerns about reproducibility have pushed research data management to the forefront. This manuscript brings together three perspectives to provide an extensive overview of data sharing: theoretical foundations, ethical and normative frameworks, and practical implementation. First, it discusses theway research data differs across fields and formats, the distinction between primary and secondary data, and how metadata helps ensuredata can be reused. It emphasizes how open data fosters transparency, reproducibility, accountability, and innovation, while also acknowledging that research data has historically been viewed as private intellectual property. Second, it explores the emergence of principles and ethical standards designed to enhance data quality and promoteresponsible use. Documentation standards, data management plans, and sharing of code and workflows have helped the FAIR (Findability, Accessibility, Interoperability, and Reusability) principles become a cornerstone for data sharing. Regulatory frameworks, such as the General Data Protection Regulation (GDPR) and California Consumer Privacy Act (CCPA), as well as mechanismssuch as de-identification and Data Trusts, address legal and ethical issues, including privacy protection, licensing, and data governance. Finally, the third major topic discusseshow theseprinciples are implemented through infrastructure, incentives, and new technologies. It addresses the significance of cultural change and recognition systems, the impact of policies by journals and funders, and the role of repositories in preservation and interoperability. It also emphasizes the emergence of novel trends, such as artificial intelligence-driven metadata generation, blockchain-based provenance, executable workflows, and privacy-preserving computation, all of which are redefining the concept of responsible and scalable data sharing. By connecting conceptual, ethical, and practical dimensions, the manuscript outlines both current challenges and realistic pathways toward transparent, collaborative, and future-oriented research. © 2025 Wiley Periodicals LLC.

Architectures of emerging biofoundry platforms for synthetic biology.

The proliferation of Internet of Things devices and edge computing infrastructure has created unprecedented opportunities for distributed workflow execution across heterogeneous edge-cloud environments. However, optimal workflow partitioning in such dynamic systems remains a significant challenge due to the complexity of resource heterogeneity, network variability, and diverse application requirements. This paper proposes a novel Dynamic Workflow Partitioning framework leveraging Deep Reinforcement Learning to intelligently distribute workflow tasks between edge nodes and cloud data centers. The framework employs a Deep Q-Network architecture enhanced with a Graph Neural Network encoder to capture workflow dependencies and system state representations. Through comprehensive evaluation using real-world workflow applications including CyberShake, Epigenomics, Inspiral, Montage, and Sipht, our approach demonstrates superior performance in minimizing execution time, reducing network overhead, and maintaining quality of service guarantees compared to traditional heuristic-based methods. The experimental results show that the proposed approach achieves up to 32% reduction in average workflow completion time and 41% improvement in resource utilization efficiency across various heterogeneous edge-cloud configurations.

Modern digital enterprises execute workflows across heterogeneous platforms (APIs, headless interfaces, messaging systems) lacking unified orchestration frameworks. While existing automation systems (IFTTT, Zapier) process millions of workflows daily using static rule- based models, and recent multi-agent frameworks (AutoGPT, LangChain) address singleplatform coordination, they fail to handle cross-platform heterogeneity with context-aware prioritization [1][3]. This gap motivates a unified orchestration architecture combining (1) context-aware Bayesian priority scoring that dynamically re-ranks tasks based on urgency, dependencies, resource cost, and user impact; (2) deterministic workflow execution integrating LangGraph state machines with Temporal's exactly-once semantics [4]; and (3) hybrid adapter architecture supporting both API-first integration and headless browser automation for platforms lacking official APIs. Our system employs Redis Streams event bus achieving sub-200ms synchronization latency. Evaluation over six months with 500 synthetic workflows spanning ecommerce, social media, productivity, and communication platforms demonstrates 65% reduction in orchestration latency (445ms vs. 760ms for commercial baseline), 99.7% system uptime (vs. 97.8% Zapier), 98% reduction in rate-limit violations, and 89% deadline adherence (vs. 63% baseline). Statistical analysis (ANOVA, p < 0.001) confirms significance with large effect sizes (Cohen's d > 1.2). Key contributions include the Bayesian priority algorithm, hybrid integration architecture, and empirical validation demonstrating enterprise-grade reliability for cross-platform autonomous agent orchestration [4][5][6].

Revisiting workflow execution in HPC: a data-flow approach

This hybrid study assessed the implementation and clinical effectiveness of a structured mental health care workflow for epilepsy. Eligible inpatients were screened systematically. Patients with scores above cutoff scores underwent structured diagnostic interviews followed by a multi-component psychotherapeutic intervention (one or two sessions) aiming to develop a personalized treatment plan. Follow-up at 1, 3, 6, and 12 months assessed treatment plan adherence and reliable change indices (RCIs) of outcomes (self-reported depressive and anxiety symptoms, health-related quality of life, work and social adjustment). Implementation was assessed through initial step penetration, fidelity of workflow execution, and diagnostic/therapeutic yields (appropriateness). Of 345 inpatients with epilepsy, 210 were eligible and 202 entered screening. Neurocognitive and linguistic deficits were the most important reasons that only 59% of all inpatients completed the screening procedure. The workflow was implemented with high fidelity (96% across all steps) and proved clinically appropriate for the population, with one in five screened patients with epilepsy receiving a psychiatric diagnosis and a personalized treatment plan based on the brief, tailored psychotherapeutic intervention (n = 41). Fifteen of these patients (37%) had not been diagnosed previously. After 12 months, 17 patients (41%) were lost to follow-up; this group showed significantly higher baseline depression scores. Of the 24 patients with complete follow-up data, 17 (71%) had initiated the recommended treatment. Eleven of those who had started treatment (65%) showed reliable improvements in at least one outcome, whereas no improvements were observed in non-adherent patients. The integrated workflow was implemented with high fidelity and was associated with promising outcomes. However, the findings highlight the need for structural reforms to improve access and effectiveness for patients with cognitive impairment, language barriers, and severe depressive symptoms.

Cloud-native data environments running on distributed architectures are severely challenged when classic Extract-Transform-Load orchestration patterns depend on static Directed Acyclic Graph structures, which do not support dynamic data dependencies, schema change, and heterogeneous source system integration. Contemporary data platforms handling data from hundreds of heterogeneous sources are burdened with increasing operational complexity as pipeline logic hard-coded in applications forms maintenance bottlenecks and governance hurdles. The metadata-driven orchestration pattern overcomes these limitations by decoupling control logic from application code into versioned metadata stores that act as centralized sources of truth for pipeline specifications. Everything configurable, such as source connections, transformation rules, data quality constraints, dependency relationships, and lineage mappings, gets declaratively defined through structured metadata schemas independent of the execution fabric. Orchestration engines query metadata repositories at runtime to build dynamic execution plans sensitive to real-time system conditions and upstream data availability trends. Technology deployments use Apache Airflow as a task orchestrator, dbt framework as an SQL-based transformer, and OpenLineage standards for end-to-end lineage tracking across distributed processing environments. The metadata layer also serves as an observability and governance platform that supports end-to-end traceability, reproducibility, and impact analysis during workflow execution. Empirical implementations in multi-tenant data platforms illustrate dramatic decreases in pipeline maintenance overhead and faster recovery from schema drift events. Cross-functional coordination is greatly enhanced as abstraction of metadata separates transformation logic from infrastructure code, allowing business rules to be defined by data analysts without requiring proficiency in intricately complex orchestration frameworks. Metadata-based orchestration lays grounding capabilities towards self-adaptive data pipelines, combining data engineering, governance, and observability under concerted architectural frameworks.

An efficient framework for scheduling security-critical tasks in resource-limited mobile edge computing using hybridized gold rush with golden jackal optimization strategy

Serialization and deserialization dominate the state transfer time of serverless workflows, leading to substantial performance penalties when executing various serverless workflow applications. We identify the key reason for serialization and deserialization as a lack of ability to efficiently access the (remote) memory of another function. To this end, we propose RMMap , an OS primitive for remote memory map, which allows a serverless function to directly access the memory of another function, even if it is located remotely. RMMap is the first to completely eliminate serialization and deserialization overhead when transferring states between any pairs of functions in (unmodified) serverless workflows. To make remote memory map efficient and feasible, we co-design it with modern networking (RDMA), OS, language runtime, and serverless platform. Evaluations using real-world serverless workloads show that integrating RMMap with Knative reduces the serverless workflow execution time on Knative by up to 2.6× and improves resource utilizations by 86.3%.

Event-Driven Architecture (EDA) has emerged as a critical paradigm for modern enterprise integration, enabling organizations to transition from traditional synchronous communication models to more responsive, decoupled systems. This comprehensive exploration begins by establishing the foundational elements of EDA—events, producers, consumers, and brokers—and contrasts these with conventional REST-based and batch processing approaches. The article examines how publish/subscribe messaging patterns serve as the backbone for scalable event distribution, while event sourcing and CQRS patterns provide powerful mechanisms for state management and specialized data access. Through real-world implementations in order processing, inventory management, and contract execution workflows, the article demonstrates the tangible benefits of event-driven systems. Technical considerations, including message broker selection, data integrity mechanisms, and schema evolution strategies, are explored in depth. The discussion culminates with resilience engineering practices for enterprise-scale event streams, covering observability, retry strategies, and scaling considerations for complex business processes like quote-to-cash workflows.