Resource Usage Research Articles

A boundedly rational model for category learning

The computational modeling of category learning is typically evaluated in terms of the model's accuracy. For a model to accurately infer category membership of stimuli, it has to have sufficient representational precision. Thus, many category learning models infer category representations that guide decision-making and the model's fitness is evaluated by its ability to accurately choose. Substantial decision-making research, however, indicates that noise plays an important role. Specifically, noisy representations are assumed to introduce an element of stochasticity to decision-making. Noise can be minimized at the cost of cognitive resource expenditure. Thus, a more biologically plausible model of category learning should balance representational precision with costs. Here, we tested an autoencoder model that learns categories (the six category structures introduced by Roger Shepard and colleagues) by balancing the minimization of error with minimization of resource usage. By incorporating the goal of reducing category complexity, the currently proposed model biases category decisions toward previously learned central tendencies. We show that this model is still able to account for category learning performance in a traditional category learning benchmark. The currently proposed model additionally makes some novel predictions about category learning that future studies can test empirically. The goal of this paper is to make progress toward development of an ecologically and neurobiologically plausible model of category learning that can guide future studies and theoretical frameworks.

Advanced Signal Processing Techniques for Real-Time Systems in Edge Computing

With the rapid expansion of edge computing, real-time systems are increasingly deployed in environments such as IoT, autonomous vehicles, and industrial applications, where efficient signal processing is crucial for performance and decision-making. However, real-time signal processing in edge computing environments faces significant challenges, including limited computational resources, data transmission constraints, and strict latency requirements. Objective: This paper aims to explore advanced signal processing techniques specifically tailored for real-time systems in edge computing. The goal is to propose novel algorithms and strategies to enhance processing efficiency, reduce latency, and optimize resource usage in these environments. Methods: We propose a combination of traditional signal processing methods and emerging techniques, including: Adaptive Filtering: Improving real-time performance in dynamic environments. Compressed Sensing: Leveraging sparsity to reduce data processing and transmission overhead. Deep Learning Models: Applying convolutional and recurrent neural networks for accurate signal classification and anomaly detection. Distributed Signal Processing: Offloading computation tasks to distributed edge devices to balance load and reduce latency. Results: Experimental results demonstrate that the proposed techniques outperform traditional methods in terms of signal processing speed, resource utilization, and accuracy. In particular, the integration of deep learning models significantly improves classification accuracy for real-time signal detection, while compressed sensing reduces bandwidth requirements without compromising signal quality. Conclusion: The proposed signal processing methods effectively address the unique challenges posed by real-time systems in edge computing environments. These techniques can be applied to a wide range of applications, such as smart cities, industrial IoT, and autonomous systems, providing a solid foundation for the development of efficient, low-latency real-time systems.

Librarians’ perceptions and use of open access resources for economic sustainability in selected university libraries in Kwara State, Nigeria

Purpose Open access (OA) resources are free to access and help in research and teaching/learning, allowing libraries to manage resources more sustainably and effectively. Hence, this study aims to examine use of OA resources for economic sustainability among librarians in selected university libraries in Kwara State, Nigeria. Design/methodology/approach This study adopts descriptive survey research design. Using total enumeration sampling technique, a sample of 46 professional librarians were drawn from the selected universities. Questionnaire was used for data collection, and descriptive statistics was used for data analysis. Findings The findings show that most of the librarians had high level of awareness about OA resources. Results demonstrate that majority of the librarians used PubMed Central bi-weekly. Findings show that the librarians used OA resources to support academic research, enhance library’s collection without additional cost and facilitate information access for users with limited financial resources. Results indicate that most of the librarians perceived that OA resources impact library’s operational cost and goal of economic sustainability. Practical implications This study highlights the need for targeted training and resource development to enhance the effective use of OA resources. Meanwhile, the consistent usage of OA resources suggests the importance of the resources to librarians, aligning with previous studies on frequent use of established OA repositories. Originality/value This study provides evidence on the relevance and impact of OA resources use on economic sustainability, demonstrating the importance of OA resources in knowledge dissemination and fostering innovation.

Fault-tolerant quantum chemical calculations with improved machine-learning models.

Easy and effective usage of computational resources is crucial for scientific calculations. Following our recent work of machine-learning (ML) assisted scheduling optimization [J. Comput. Chem. 2023, 44, 1174], we further propose (1) the improved ML models for the better predictions of computational loads, and as such, more elaborate load-balancing calculations can be expected; (2) the idea of coded computation, that is, the integration of gradient coding, in order to introduce fault tolerance during the distributed calculations; and (3) their applications together with re-normalized exciton model with time-dependent density functional theory (REM-TDDFT) for calculating the excited states. Illustrated benchmark calculations include P38 protein, and solvent model with one or several excitable centers. The results show that the improved ML-assisted coded calculations can further improve the load-balancing and cluster utilization, owing primarily profit in fault tolerance that aims at the automated quantum chemical calculations for both ground and excited states.

Efficient File Security Scanning: Combining Hash Histories and Bloom Filters to Minimize Redundancy

As file repositories expand in size, traditional full-file security scanning becomes computationally expensive and redundant. This paper introduces a novel hybrid framework that leverages hash histories embedded in file metadata and Bloom filters for efficient security scanning. The approach ensures that only modified or newly added files are scanned, reducing overhead while maintaining robust security coverage. By augmenting file metadata with hash histories, the system provides decentralized tracking of file state changes. Bloom filters further optimize the process by efficiently determining whether a file requires scanning. The tradeoff between increased file sizes due to metadata augmentation and the improved scanning performance is thoroughly analyzed. Evaluations demonstrate significant reductions in scan time and resource usage, making the framework highly suitable for large-scale file repositories. Keywords File security, Hash histories, Bloom filters, Metadata augmentation, Incremental scanning, Security optimization

Optimized Edge-Cloud System for Activity Monitoring Using Knowledge Distillation

Driven by the increasing care needs of residents in long-term care facilities, Ambient Assisted Living paradigms have become very popular, offering new solutions to alleviate this burden. This work proposes an efficient edge-cloud system for indoor activity monitoring in long-term care institutions. Action recognition from video streams is implemented via Deep Learning networks running at edge nodes. Edge Computing stands out for its power efficiency, reduction in data transmission bandwidth, and inherent protection of residents’ sensitive data. To implement Artificial Intelligence models on these resource-limited edge nodes, complex Deep Learning networks are first distilled. Knowledge distillation allows for more accurate and efficient neural networks, boosting recognition performance of the solution by up to 8% without impacting resource usage. Finally, the central server runs a Quality and Resource Management (QRM) tool that monitors hardware qualities and recognition performance. This QRM tool performs runtime resource load balancing among the local processing devices ensuring real-time operation and optimized energy consumption. Also, the QRM module conducts runtime reconfiguration switching the running neural network to optimize the use of resources at the node and to improve the overall recognition, especially for critical situations such as falls. As part of our contributions, we also release the manually curated Indoor Action Dataset.

Epidemiology and outcomes of septic shock in Thai children: a nationwide retrospective study from 2015 to 2022

BackgroundPaediatric septic shock is a formidable challenge worldwide that significantly impacts health care systems. This nationwide retrospective study analyses the prevalence and mortality rates of paediatric septic shock across Thailand from 2015 to 2022, focusing on hospital burdens, including mechanical ventilation and renal replacement therapy.MethodsThe study included paediatric patients ranging from infants to individuals under 18 years of age who were admitted to hospitals due to septic shock across Thailand. Data were retrospectively gathered from the Thai National Health Security Office for the years 2015–2022. The data included demographic data, clinical outcomes, and hospital burden indicators such as mechanical ventilation and renal replacement therapy.ResultsFrom 2015 to 2022, there were 30,718 paediatric admissions for septic shock, with a peak in 2019. The highest incidence was observed in infants, accounting for 44.7% of the cases. The prevalence rate increased from 190 per 100,000 population in 2015 to a peak of 280 per 100,000 population in 2020. Mortality rates decreased from 30.7% in 2015 to 20.2% in 2022, with a peak of 40.5% observed in the central region in 2015. The study highlighted a substantial health care burden, with 34.9% of patients requiring prolonged mechanical ventilation and 23.4% needing renal replacement therapy.ConclusionsDespite a decrease in mortality, paediatric septic shock remains a significant burden on the health care system in Thailand. Urgent improvements in resources and adherence to clinical guidelines, especially in under-resourced areas, are necessary. Addressing disparities in mortality and resource usage across hospital levels is vital for improving outcomes and reducing the health care burden of paediatric patients with septic shock.

Navigating the path to environmental sustainability: Insights from CIVETS on the intersection of ICT diffusion, natural resources, and green technological innovation.

The rapid technological and economic growth in CIVETS countries poses challenges to environmental sustainability. This study explores the intricate relationships between Information and Communication Technology (ICT), natural resource usage, and green innovation in these economies. Employing advanced statistical models uncovers long-term and short-term patterns, providing valuable insights for policymakers, stakeholders, and scholars. Key findings reveal that ICT diffusion can potentially reduce carbon emissions, while natural resource rent is linked to increased ecological footprints. Conversely, green technological innovation is promising in alleviating environmental degradation. The study underscores the importance of comprehensive policies integrating ICT diffusion, sustainable resource management, and green innovation. To achieve environmental sustainability, the study recommends responsible natural resource extraction, fiscal incentives for green innovation, digital tools for environmental monitoring, and international collaboration. Public awareness and sustainable urbanization practices guided by ICT are also crucial. These insights help policymakers balance economic development with ecological preservation in the CIVETS nations.

Green and efficient fine control of regional irrigation water use coupled with crop growth-carbon emission processes

The effective, environmentally friendly, and optimally deployed management of regional agricultural water resources is crucial to ensuring food security and the sustainable use of water resources in the face of challenging agricultural issues, such as increasing greenhouse gas (GHG) emissions, water scarcity, and rapid population and economic growth. In this study, a multiobjective model for the optimal distribution of regional agricultural water resources was built, using the GIS-DSSAT and GIS-DNDC models, to simulate the regional spatial raster crop growth process and carbon emission process. The model allowed for the creation of synergies in both crop production and emission reduction. The model was solved using a fuzzy planning algorithm, and applied to the main Sanjiang Plain grain production area in Heilongjiang Province, yielding the best water allocation scheme and a set of planting structure adjustment schemes for the main grain crops of rice, maize, and soybeans in 1074 Sanjiang Plain response units. In contrast to the current method, which relies heavily on soil and water resources, the model developed in this paper reduced greenhouse gas (GHG) emissions by 10 %, energy consumption by 14.4 %, and regional irrigation water use by 34.48 %. It achieved the dual goal of reducing GHG emissions and conserving water, while increasing the synergy between increased regional food production and decreased emissions by 21 % compared to the status quo. Climate change will pose series of challenges to agricultural production, particularly the arable land and water resource usage. By optimizing cropping structures and irrigation systems, climate-adaptive management can not only reduce the area of arable land and water consumption, but also decrease carbon emissions; for the SSP2–4.5 and SSP3–7.0 scenarios, this trend resulted in increases in economic benefits of 2.4 % and 3.3 %, respectively, and decreases in carbon emission scenarios of 14.6 % and 20.7 %, respectively. The yields and GHG emissions of the response units were sensitive to these changes. This study offers decision-making support for the intense, effective, and low-carbon management of water resources.

Wastewater treatment plants circular performance models evaluation: Portugal case-study

Population growth, economic growth, and changes in societal habits have led to significant changes in resource consumption. Therefore, it's crucial to accelerate the “reduce, reuse, recycle, and recover” of resources to ensure the balance of ecosystems, and water is surely one of the most fundamental resources. The acceleration of this approach in the water cycle makes sense only if we combine a circular economy (CE) transition with a sustainable perspective. In this context, more rational usage of water resources (which are under pressure) and more sustainable wastewater practices are expected to be a way towards the CE in the water and wastewater sector. This study provides a description and evaluation of existing frameworks that can be used to measure and assess the level of circularity of the wastewater treatment plant (WWTP). The treatment of urban wastewater requires new concepts of management and operation for the adaptation of existing plants, which lack robustness and flexibility, to face these new challenges and requirements because we can no longer continue to look at the WWTP only as treatment units, but as wastewater resource recovery facilities. This transformation must be transposed according to a matrix that allows the assessment to describe the current situation, analyse the problem, identify vulnerabilities and opportunities, identify, and evaluate measures, and identify and evaluate strategies. Considering that decision-makers face profound uncertainties such as climate change, population growth, population needs, innovative technologies, economic developments, ecosystem preservation and the impacts of human and natural activities.

Generative artificial intelligence, present and perspectives in public administration

Artificial Intelligence (AI) is playing an increasingly important role in public administration, offering innovative solutions for efficiency and better public services. The application of AI in public administration aims to modernise and streamline the services offered to citizens. AI contains advanced algorithms, neural networks, and machine learning techniques to improve performance and efficiency over time. Smart algorithms, responsibly implemented, could help governments enhance service delivery and strengthen citizen engagement in local, city, regional, and national administrations around the world. Governments in some states have discreetly launched pilot programs to test developing AI applications, demonstrating an enthusiasm for technology at least as great as that in the private sector. By utilising AI, public administrations can automate repetitive processes, optimize resource usage, and improve interaction with the public. The purpose of the research is to show the extent and scope that AI tools integrated into organisations have in almost all areas of the public and private sectors in general, and state administrations in particular. The research objectives were: to analyse the dynamics of investments in AI tools; to discover the main services in state administrations that can be better performed using AI; to analyse the risks; and to highlight the perspective of development and integration of AI in public administrations of a few European states. The research is based on a comparative analysis of the implementation of AI tools across fields, components, and regions at the European level mainly, but also for Romania. The analysis period is 2021-2030. The results show a continuously increasing trend and an ever-faster rate of assimilation of AI tools in the field of public administration in an increasing number of states from most developed regions worldwide.

ZeROf-Offload: forward-gradient scheme for efficient full parameter fine-tuning of billion-scale language models

Abstract In large language models (LLMs), full-parameter fine-tuning is crucial for task-specific adaptation. Traditionally, this relies on deep learning training frameworks utilizing the back-propagation scheme. However, this scheme presents inherent issues, e.g. activation memory bottlenecks and backward locking, which limit the efficient computational resource usage. In this work, we propose the design and analysis of ZeROf-Offload, an innovative fine-tuning framework that adapts the forward-gradient scheme. This framework adopts a unique forward-gradient-oriented CPU offload strategy, enabling fine-tuning of billion-scale LLMs solely in the forward phase and enhancing computational efficiency. Empirical evaluations reveal the advantage of eliminating the backward phase in fine-tuning. ZeROf-Offload achieves134 TFlops/GPU for models with over 130 billion parameters on a single DGX-A100 node, outperforming DeepSpeed’s ZeRO-Offload, which achieves 102 TFlops/GPU for models with up to 53.7 billion parameters, the largest size manageable within GPU memory limitations. Furthermore, we have expanded ZeROf-Offload for multi-DGX-A100 environments with integrated 3D parallelism, achieving near-linear speedup across up to 128 GPUs and the token throughput by 1.4x and 1.5x, respectively. The experimental results demonstrate that the proposed ZeROf-Offload has achieved the highest throughput performance compared to all examined state-of-the-art frameworks.

Usability and Performance Comparison: Implementation of Tibero and Oracle Databases in the Context of CAMS Software Development

In the world of software development, the role of database systems is very vital. Enterprise software, designed to handle the complex challenges of large organizations, requires reliable and efficient databases. Oracle, one of the top choices in the industry, stands out with its performance and flexibility. On the other hand, Tibero, a relational DBMS from TmaxSoft, offers the high performance, reliability and scalability required in business environments that require big data management. This research was conducted to analyze the technical side of the Oracle and Tibero databases in the context of the CAMS (Customer Asset Management System) application, with a focus on usability and performance aspects. This research uses the Performance Testing method to evaluate CPU, Memory, Storage resource usage and TPS (Transaction Per Second) of the two databases as well as the System Usability Scale (SUS) to measure user experience. The results provide information to software developers in selecting databases that suit business needs, while contributing to the development of the information technology industry

Ultrasonic-assisted membrane processes for the systematic purification of glycyrrhiza wastewater

A significant amount of glycyrrhiza wastewater is generated in the cleaning process of glycyrrhiza. The wastewater contains polysaccharide, glycyrrhizic acid liquiritin, and other polyphenols, which is expensive for cleanup and wastes medical resources. To reduce environmental pollution from glycyrrhiza wastewater and increase the resource usage efficiency of glycyrrhiza components. According to the physicochemical properties of the component in glycyrrhiza wastewater, the ultrasonic assisted membrane separation mode was adopted to regulate the micellar state of glycyrrhizic acid and enhance the differences in membrane separation of polysaccharides, saponins, and flavones, in order to achieve the classification and separation of polysaccharides, saponins, and flavones while removing organic matter in glycyrrhiza wastewater. However, the efficiency, application, and mechanism of ultrasonic-assisted membrane technology for the separation of polysaccharides, saponins, and flavonoids from glycyrrhiza wastewater remain unclear. This study presents a green and feasible technical strategy for glycyrrhiza wastewater treatment that was developed by adjusting the parameters of ultrasonic assisted membrane separation. In this study, the systematic separation mode of ultrasonic enhanced ultrafiltration combined with nanofiltration is provided. The SCQ-9200E ultrasonic system was provided for the study with adjustable ultrasonic power, and the ultrasonic frequency was 40 kHz. The glycyrrhizic acid micelle was changed using ultrasonic power, pH, and molecular weight cut off (MWCO), and the separation differences among polysaccharide, glycyrrhizic acid, and liquiritin were enhanced. The optimal polysaccharide separation parameters used in the first step: MWCO 30 kDa, ultrasonic power 500 W and pH 5.00, and the rejections of polysaccharide, glycyrrhizic acid, and liquiritin were 87.72 %, 8.01 %, and 6.57 %, respectively. The second step included the following parameters for the separation of liquiritin and glycyrrhizic acid: MWCO 10 kDa, ultrasonic power 100 W and pH 8.00, the rejections of liquiritin and glycyrrhizic acid were 9.22 % and 40.65 %, respectively. The third step is to remove the low molecular sugar in liquiritin by nanofiltration: MWCO 800 Da, pH 8.00, retention solution diluted and separated twice, the rejection of liquiritin and total sugar were 95.72 % and 3.70 %, respectively. Ultrasonic may regulate the microtopography of glycyrrhiza wastewater with the power intensity of 50 W/L, improving the mass transfer efficiency of glycyrrhizic acid and liquiritin in the ultrafiltration separation process. As the separation volume of wastewater increased from 2.00 L to 20.00 L, the concentrations of polysaccharide, glycyrrhizic acid, and liquiritin increased by 2.5–35.4 times, 0.6–15.2 times, and 2.4–32.8 times, respectively, significantly increasing the content of index components in wastewater and solving the problem of recycling and resource utilization in glycyrrhiza wastewater.

Dynamic Matrix Multiplication Using Reinforcement Learning

Efficient matrix multiplication is a core computational task in machine learning, deep learning, and scientific computing, as it significantly impacts the performance of model training and inference. This project explores the use of reinforcement learning (RL) to dynamically select the optimal matrix multiplication algorithm based on input matrix characteristics. We compare traditional matrix multiplication methods, including Strassen’s algorithm, Winograd’s algorithm, and tiled matrix multiplication, and implement a Q- learning-based agent to choose the most efficient algorithm adaptively. Our RL-driven approach enables real-time algorithm selection, balancing computational speed and resource usage across different matrix types and sizes. Experimental results demonstrate the system's ability to reduce execution time by selecting algorithms suited to specific scenarios, providing insights into algorithm efficiency and adaptability for various data sizes. Furthermore, this research suggests practical optimization strategies for handling large-scale computations in machine learning applications, including using parallel processing, GPU acceleration, and parameter tuning to minimize computational bottlenecks. This project underscores the potential of integrating RL for adaptive computation within AI workflows, offering a scalable solution for enhancing matrix operation efficiency in data-intensive applications

Efficient resource allocation for generative AI workloads in cloud-native infrastructures: A multi-tiered approach

Resource management becomes essential in ensuring that generative AI workloads in cloud-native infrastructures deliver the best results. The architecture described in this article targets such workloads due to their inherent fluctuations in resource usage and the difficulties in scaling them. The proposed framework divides resources into groups to guarantee that applications are given support based on difficulty level. The features of the proposed methodology are the performance assessment of resource distribution effectiveness, taking into account metrics, including latency, throughput, and utilization rates. Furthermore, examples have been provided to support the use of this approach and its efficiency in real-life situations. Based on these, applying the multi-tiered approach to resource management improves the organization's operations performance and minimizes expenses connected with resource provisioning. Such a study also emphasizes the importance of developing flexible and effective resource management tools that can be especially useful in modern generative AI development environments.

An Approach to Synergise the Management of Lean, Green, Quality and Waste for Achieving ZED Emphasising SDG12

The advent of the Zero Defect and Zero Effect (ZED) concept has thrown a significant challenge in integrating quality management, lean philosophy, waste management, and green management within industrial operations. Achieving quality outcomes requires fulfilling both external and internal customer expectations through effective requirement analysis and meeting specific quality dimensions. At the same time, organizations must minimize resource usage— such as time, energy, manpower, and materials— while maintaining product quality. This is a core principle of lean management, which aims to improve margins and ensure organizational growth and survival. Simultaneously, environmental sustainability must be prioritized by minimizing air, water, and soil pollution, in line with green management principles. This study employs the Fuzzy Analytical Hierarchy Process (FAHP) to prioritize key parameters for assessing ZED performance across three industrial sectors in India. Additionally, it proposes practical methods for improving vital components using Six Sigma metrics to enhance operational excellence and sustainability. The novelty of this research lies in the integration of a ZED Maturity Model with the Business Excellence Model, along with the analysis of vital components through Six Sigma metrics to develop a comprehensive framework for improving operational performance and sustainability. This paper provides a theoretical framework and actionable strategies for researchers and practitioners to effectively integrate ZED principles, fostering sustainability and operational performance in manufacturing industries while supporting Sustainable Development Goal 12.

EARLY DETECTION OF BRAIN TUMOR USING MRI AND TRANSFER LEARNING

Brain tumors pose significant risks to cognitive functions, making early detection crucial for improving patient survival rates. Accurate tumor detection aids neuro-oncologists in diagnosing tumor types and recommending appropriate treatments. However, manual detection is challenging, time-consuming, and prone to human error. Recently, Deep Learning (DL) models have demonstrated substantial potential in efficiently classifying large image datasets. This paper presents three novel and efficient multi-class DL architectures leveraging transfer learning to classify different brain tumors. Our primary contribution is to enhance the predictive accuracy while minimizing the usage of computational resource compared to other state-of-art models in the literature by forgoing preprocessing, segmentation, hybrid models, and augmentation techniques. Additionally, we reduce the number of FC layers to streamline computation. We conduct extensive experiments to evaluate the performance of our models using the brain tumor Figshare T1-weighted contrast-enhanced MRI dataset, comprising 3064 images from three distinct tumor types. The InceptionV3 model records a 98.36% accuracy level with five-fold cross-validation. By incorporating batch normalization and optimizing the learning rate for the Adam optimizer, the Xception model reached 99.19% accuracy. Finally, we utilize Particle Swarm Optimization (PSO) to fine-tune the learning rate of the Stochastic Gradient Descent (SGD) optimizer. The Xception model attained 98.85% accuracy. These results highlight the novelty of our approach, offering a practical solution for neuro-oncologists, particularly through the fine-tuned Xception model, which delivers early and accurate tumor detection with minimal computational resources.

Application of Stable Diffusion and LoRA Models in AI Drawing

Abstract. This paper explores the application of Stable Diffusion model and LoRA (Low-Rank Adaptation) model in AI-generated artwork. The authors introduce the foundational principles of Stable Diffusion model and LoRA, as well as their application in high-quality image generation. Using three popular datasets ImageNet, COCO, and CelebA we apply various image quality assessment metrics, including PSNR (Peak Signal-to-Noise Ratio), IS (Inception Score), and FID (Frchet Inception Distance), and further validate their potential in artistic creation through subjective evaluations. By comparing the performance of these two models across different datasets, we examine their strengths, weaknesses, areas for improvement in image generation tasks, along with user experience considerations. The experimental results show that the Stable Diffusion model excels in terms of image quality and diversity, while the LoRA model offers significant benefits in computational efficiency and resource usage. Through comprehensive experimental evaluations, this paper provides a scientific basis for model selection in AI art creation and offers insights for the future development of hybrid models.

Gwydion: Efficient auto-scaling for complex containerized applications in Kubernetes through Reinforcement Learning

Containers have reshaped application deployment and life-cycle management in recent cloud platforms. The paradigm shift from large monolithic applications to complex graphs of loosely-coupled microservices aims to increase deployment flexibility and operational efficiency. However, efficient allocation and scaling of microservice applications is challenging due to their intricate inter-dependencies. Existing works do not consider microservice dependencies, which could lead to the application’s performance degradation when service demand increases. As dependencies increase, communication between microservices becomes more complex and frequent, leading to slower response times and higher resource consumption, especially during high demand. In addition, performance issues in one microservice can also trigger a ripple effect across dependent services, exacerbating the performance degradation across the entire application. This paper studies the impact of microservice inter-dependencies in auto-scaling by proposing Gwydion, a novel framework that enables different auto-scaling goals through Reinforcement Learning (RL) algorithms. Gwydion has been developed based on the OpenAI Gym library and customized for the popular Kubernetes (K8s) platform to bridge the gap between RL and auto-scaling research by training RL algorithms on real cloud environments for two opposing reward strategies: cost-aware and latency-aware. Gwydion focuses on improving resource usage and reducing the application’s response time by considering microservice inter-dependencies when scaling horizontally. Experiments with microservice benchmark applications, such as Redis Cluster (RC) and Online Boutique (OB), show that RL agents can reduce deployment costs and the application’s response time compared to default scaling mechanisms, achieving up to 50% lower latency while avoiding performance degradation. For RC, cost-aware algorithms can reduce the number of deployed pods (2 to 4), resulting in slightly higher latency (300μs to 6 ms) but lower resource consumption. For OB, all RL algorithms exhibit a notable response time improvement by considering all microservices in the observation space, enabling the sequential triggering of actions across different deployments. This leads to nearly 30% cost savings while maintaining consistently lower latency throughout the experiment. Gwydion aims to advance auto-scaling research in a rapidly evolving dynamic cloud environment.