Serving Machine Learning Workloads in Resource Constrained Environments: a Serverless Deployment Example

We propose a set of optimization techniques for transforming a generic AI codebase so that it can be successfully deployed to a restricted serverless environment, without compromising capability or performance. These involve (1) slimming the libraries and frameworks (e.g., pytorch) used, down to pieces pertaining to the solution; (2) dynamically loading pre-trained AI/ML models into local temporary storage, during serverless function invocation; (3) using separate frameworks for training and inference, with ONNX model formatting; and, (4) performance-oriented tuning for data storage and lookup. The techniques are illustrated via worked examples that have been deployed live on geospatial data from the transportation domain. This draws upon a real-world case study in intelligent transportation looking at on-demand, realtime predictions of flows of train movements across the UK rail network. Evaluation of the proposed techniques shows the response time, for varying volumes of queries involving prediction, to remain almost constant (at 50 ms), even as the database scales up to the 250M entries. The query response time is important in this context as the target is predicting train delays. It is even more important in a serverless environment due to the stringent constraints on serverless functions’ runtime before timeout. The similarities of a serverless environment to other resource constrained environments (e.g., IoT, telecoms) means the techniques can be applied to a range of use cases.

In the evolving landscape of health information management, the application of blockchain and edge computing technologies to chronic disease management remains underexplored, despite the urgent need for scalable, secure, and real-time solutions. This systematic review examines the integration of these technologies in healthcare, with a specific focus on diabetes management. We analyzed 52 studies, categorizing findings into three key areas: enhanced data security and privacy through decentralized frameworks and tamper-proof storage; real-time data processing, enabled by edge computing for immediate analytics and alerts; and scalability, achieved via hybrid architectures that optimize data handling. Our review identifies critical gaps in the existing literature, particularly the lack of tailored approaches for chronic disease contexts like diabetes, and outlines future directions for developing robust, secure, scalable, and real-time data solutions. This study provides a foundation for innovation in healthcare technology that can significantly impact diabetes care and chronic disease management.

Natural disasters, such as earthquakes, floods, wildfires, and hurricanes, are occurring more frequently and becoming more severe, underscoring the need for advanced real-time prediction systems. Existing static monitoring infrastructures with limited coverage suffer from delayed processing and expensive maintenance costs. In comparison, mobile sensor networks (MSNs) offer cost-effective, dynamic, and scalable solutions for collecting real-time environmental data. This research focuses on the design and implementation of mobile sensor networks for real-time disaster prediction, with a particular emphasis on advanced sensing technologies and predictive algorithms. The methodology involves the deployment of mobile sensor units equipped with environmental sensors, communication modules, and data processors to monitor temperature, humidity, seismic activity, and air quality. These data are analyzed in real-time with machine learning algorithms to detect early warning signs of potential disasters and are centralized for further processing. System evaluation is performed through simulating disaster scenarios and is benchmarked against conventional fixed-network methodologies. Results indicate marked improvements in prediction accuracy, response time, and adaptability to rapidly changing conditions. This work analyzes the practical applications of MSNs in disaster-prone areas, discusses challenges such as energy consumption and data protection, and provides suggestions for future improvements. This study extends the expanding area of intelligent disaster management and mobile sensor networks to enhance early warning and response systems.

Monitoring subject recruitment is key to the success of a clinical trial. Accordingly, researchers have developed accrual-monitoring tools to support the design and conduct of trials. At an institutional level, delays in identifying studies with high risk of accrual failure can lead to inefficient and costly trials with little chances of meeting study objectives. Comprehensive accrual monitoring is necessary to the success of the research enterprise. This article describes the design and implementation of the University of Kansas Cancer Center Accrual Prediction Program, a web-based platform was developed to support comprehensive accrual monitoring and prediction for all active clinical trials. The Accrual Prediction Program provides information on accrual, including the predicted completion date, predicted number of accrued subjects during the pre-specified accrual period, and the probability of achieving accrual targets. It relies on a Bayesian accrual prediction model to combine protocol information with real-time trial enrollment data and disseminates results via web application. First released in 2016, the Accrual Prediction Program summarizes enrollment information for active studies categorized by various trial attributes. The web application supports real-time evidence-based decision making for strategic resource allocation and study management of over 120 ongoing clinical trials at the University of Kansas Cancer Center. The Accrual Prediction Program makes accessing comprehensive accrual information manageable at an institutional level. Cancer centers or even entire institutions can reproduce the Accrual Prediction Program to achieve real-time comprehensive monitoring and prediction of subject accrual to aid investigators and administrators in the design, conduct, and management of clinical trials.

Machine learning based success prediction for crowdsourcing software projects

A number of technologies enabled by Internet of Thing (IoT) have been used for the prevention of various chronic diseases, continuous and real-time tracking system is a particularly important one. Wearable medical devices with sensor, health cloud and mobile applications have continuously generating a huge amount of data which is often called as streaming big data. Due to the higher speed of the data generation, it is difficult to collect, process and analyze such massive data in real-time in order to perform real-time actions in case of emergencies and extracting hidden value. using traditional methods which are limited and time-consuming. Therefore, there is a significant need to real-time big data stream processing to ensure an effective and scalable solution. In order to overcome this issue, this work proposes a new architecture for real-time health status prediction and analytics system using big data technologies. The system focus on applying distributed machine learning model on streaming health data events ingested to Spark streaming through Kafka topics. Firstly, we transform the standard decision tree (DT) (C4.5) algorithm into a parallel, distributed, scalable and fast DT using Spark instead of Hadoop MapReduce which becomes limited for real-time computing. Secondly, this model is applied to streaming data coming from distributed sources of various diseases to predict health status. Based on several input attributes, the system predicts health status, send an alert message to care providers and store the details in a distributed database to perform health data analytics and stream reporting. We measure the performance of Spark DT against traditional machine learning tools including Weka. Finally, performance evaluation parameters such as throughput and execution time are calculated to show the effectiveness of the proposed architecture. The experimental results show that the proposed system is able to effectively process and predict real-time and massive amount of medical data enabled by IoT from distributed and various diseases.

In the evolving landscape of intelligent transportation systems, ensuring robust cybersecurity for in-vehicle communication is critical. This paper proposes a hybrid learning framework for intrusion detection in Controller Area Network (CAN)-based autonomous vehicles by integrating a Decision Transformer with a Light Gradient Boosting Machine (LightGBM) classifier. The system incorporates contextual information from raw CAN frames and multimodal sensor data, enabling the construction of temporal trajectories that reflect historical driving behavior. These trajectories are encoded using a transformer-based temporal model to capture dynamic patterns, followed by lightweight boosted tree classification for efficient and interpretable decision-making. To enable real-time inference, the model is implemented on a Zynq UltraScale+ FPGA using hardware modules such as FIFO buffers, digital filters, tanh scalers, and sliding window controllers. The implementation was trained using a Python environment and evaluated on the OCSLab CAN-intrusion dataset comprising over 100,000 frames across multiple attack types, including DoS, Fuzzy, and Impersonation. The proposed system achieves a True Positive Rate of 71.50%, an Attack Detection Rate of 72.00%, and an overall accuracy of 80.05%, while maintaining a low False Positive Rate of 3.50%. Compared to conventional CNN and Rainbow DQN-based models, it demonstrates superior performance in detection, decision latency (under 10 ms), and computational efficiency. This hybrid model presents a scalable, cost-effective, and real-time capable solution for enhancing vehicular cybersecurity, making it highly suitable for deployment in safety-critical autonomous driving environments.

Introduction: Cardiovascular disease (CVD) has become the leading cause of death worldwide, accounting for 32% of global deaths in 2016. Hypertension, which can be mitigated through diet, exercise and adherence to medication, is the greatest risk factor for CVD. However, public spending on CVD prevention remains low in many low- and middle-income countries partly due to a lack of evidence of low-cost, scalable, cost-effective and evidence-based CVD prevention programs. The goal of this study was to quantify the scalability and cost effectiveness of an evidence-based community-based hypertension management study (COBIN) in Nepal. Hypothesis: A community health worker-led BP monitoring and lifestyle counseling intervention is a cost-effective solution to achieving greater BP control in Nepal. Methods: We conducted a cost-effectiveness (CE) and budget impact (BI) analysis of a community health worker-led hypertension prevention and management intervention in Nepal that has previously been shown to reduce BP. Costs were retrospectively collected to estimate per capita and total costs of a national scale-up focusing on three scenarios: (A) hypertensives only; (B) prehypertensives and hypertensives; and (C) all adults aged 25-65 years. The primary CE measure was incremental cost per CVD disability-adjusted life year (DALY) averted. Both CE and BI analyses were conducted for each scenario. One-way sensitivity analyses were conducted to assess the impact of uncertainty in key parameter values on the primary CE measure. Results: The first-year BI was estimated to be an average of US$0.88 per participant: a total of US$2,698,181, US$5,667,503, and US$10,832,403 in scenarios A, B, and C respectively. In subsequent years costs are roughly half as much. In the base-case CE analysis, scenarios A, B, and C resulted in an incremental cost-effectiveness ratio (ICER) of US$185, US$340, and US$303/DALY respectively. One-way sensitivity analyses around the base-case analysis for scenario C show that the ICER was most sensitive to uncertainty in the estimate of SBP reduction among normotensives, varying the ICER between US$225 and US$465/DALY. Conclusions: The program is highly cost effective in all three scenarios compared to the WHO threshold of US$835/DALY for Nepal. This work presents the first evidence from Nepal that a community-based hypertension prevention and management program can be a cost-effective, low-cost, and scalable solution to control blood pressure nationwide.

The Meeting Request Station for College HOD/Principal Meeting Management is a sophisticated automated system designed to streamline the process of scheduling meetings in academic institutions using embedded technology, IoT, and GSM communication. The system addresses the inefficiencies of traditional manual methods by providing a self-service terminal where students and faculty can independently submit meeting requests. Users interact with the station by selecting the purpose of their meeting, scanning a QR code displayed on the screen, and having the details automatically sent via SMS to a central management database. This real-time data transmission allows the HOD/Principal to access and manage meeting requests through a dedicated interface, where they can view requests, prioritize based on queue, and acknowledge or schedule meetings as per their availability. The system ensures that users receive automated SMS notifications about their meeting status, including updates on queue position and acceptance, thus eliminating the uncertainty and delays common in manual scheduling. The integration of an Arduino-based embedded platform with a color display and GSM module provides a cost-effective, scalable, and flexible solution suitable for various institutional requirements. Additionally, the automated nature of the system reduces the administrative burden on staff, who would otherwise need to handle requests manually. By enabling real-time communication and efficient queue management, the Meeting Request Station significantly enhances user experience, providing a structured approach to meeting scheduling. It addresses key challenges such as high administrative workload, lack of real-time updates, and ineffective communication. This system represents a practical and innovative approach to improving meeting management in academic settings, ensuring that scheduling is not only faster and more accurate but also user-friendly and resource-efficient.

Underwater acoustic monitoring plays a vital role in marine research, environmental conservation, and vessel detection. However, the rapid growth of sonar data generated by modern sensing technologies has created a strong demand for automated, accurate, and scalable analysis systems. Conventional sonar analysis relies heavily on manual interpretation and isolated machine learning models, leading to inefficiencies, inconsistent results across varying environments, and limited real-time applicability. These challenges highlight the need for a unified framework capable of handling the complete analytical pipeline in a streamlined and user-friendly manner. This study introduces a Unified MultiModel Learning Platform designed for intelligent classification and regression of SONAR signals in complex marine acoustic environments. The proposed system integrates end-to-end functionalities, including data preprocessing, feature scaling, dataset partitioning, model training, evaluation, comparison, visualization, and real-time prediction. Central to the framework is a novel hybrid ensemble model based on Classification and Regression Tree (CART) principles, combining Decision Tree and Extra Trees algorithms through a voting mechanism, referred to as the Ensemble EDT model. To ensure robust benchmarking, additional models such as Gradient Boosting and Support Vector Machine are incorporated for comparative performance evaluation. Experimental findings indicate that the Ensemble EDT model achieves superior classification accuracy with reduced misclassification rates and demonstrates strong regression performance with high predictive alignment. The integrated CART-based ensemble approach enhances model stability and generalization in noisy underwater environments. Furthermore, the platform provides a web-based interface supporting both individual and batch predictions, offering an efficient, scalable, and practical solution for advanced underwater acoustic signal analysis.

A hybrid soft sensor framework for real-time biodiesel yield prediction: Integrating mechanistic models and machine learning algorithms

Surveillance in wild birds is essential for the timely detection of high pathogenicity avian influenza (HPAI) strains. As flocks congregate in large numbers in wetlands and may potentially contaminate the environment with pathogens, the monitoring of such water bodies represents an attractive opportunity to complement animal testing and to improve surveillance for avian influenza. To increase sensitivity, water concentration is often required but available methods based on (ultra)filtration and precipitation are mostly limited by the use of pumping equipment and by the need to identify the representative sample volumes. In contrast, passive samplers (PS) offer a cost-effective and scalable solution that requires basic devices for the deployment of adsorbent materials and minimal training for their installation in the field. This study evaluated nine materials for their virus adsorption efficiency in brackish and freshwater. Cotton gauze, nitrocellulose, and nylon showed the best performance across different deployment times, with the highest recovery after 24 h. Shorter (3 h) and longer (7 days) deployments also proved effective, accommodating different sampling regimens according to the logistical needs. Importantly, PS revealed their efficacy in adsorbing also deteriorated virions or in dynamic ecosystems subjected to changes in water volumes. Field trials in wetlands corroborated laboratory findings and demonstrated that PS allowed detecting avian influenza virus (AIV, including HPAI strains) genome in water bodies, yielding consistent results with active surveillance in wild birds. By offering a simple, cost-effective, and versatile solution, PS represent a promising tool for environmental AI monitoring and can successfully complement existing avian influenza surveillance activities.

The rise of cloud-based microservices means that how we design and deploy large-scale distributed applications is fundamentally changing. The benefits of microservices include scalability, more flexible and faster development. However, the highly flexible and interconnected nature of microservices presents many challenges to maintaining system reliability. An example of such a roadblock is predicting and preventing errors before they can manifest into service degradations or outages. Traditional rule-based monitoring systems can only detect faults after performance has already degraded, reducing the capacity to actively prevent slowdowns. This research output enables real-time observability data and provides a machine learningbased approach to fault prediction for cloud-based microservices. The framework utilizes the Google 2019 Cluster Trace dataset - one of the largest operational trace datasets - containing pervasive application and system-level metrics such as CPU utilization, page cache memory, cycles per instruction (CPI), memory accesses per instruction (MAI), and scheduling class. These are combined with feature engineering, solid preprocessing steps, and multiple classification algorithms. The ability of six machine learning models (Logistic Regression, Random Forest, Extra Trees, Histogram Gradient Boosting, Linear Support Vector Classifier, and XGBoost) to classify cases into normal or fault-prone was assessed. Ensemble techniques were also applied, and for this task, both interpretable and complex models provided high-performance solutions. The system architecture is designed to be flexible, scalable, and capable of providing insights into faults to prevent them from occurring in real time. From the results, it is evident that real-time failure prediction solutions require good telemetry, robust preprocessing, and suitable algorithms. The proposed approach challenges the Interpretability vs Accuracy trade-off, which must coexist with operational feasibility, reflecting the ongoing evolution of predictive reliability engineering within microservices.

The integration of Internet of Things (IoT) devices and artificial intelligence has revolutionized smart healthcare, enabling continuous monitoring, personalized treatment, and proactive intervention. Cardiovascular disease (CVD) is still one of the main causes of death worldwide, yet we don’t have a strong, unified way to monitor it in real time using deep learning. Traditional machine learning methods often fall short; they strive to combine different kinds of medical data, deal with delays in processing, and work reliably in low-resource settings where technology and infrastructure are limited. This review focuses on closing that gap by examining how transformer-based deep learning models can be applied to multimodal IoT data in smart healthcare, with a special emphasis on predicting CVD. We classify and assess the latest transformer architectures based on how they fuse data, their areas of application, and their readiness for real-time use. Our analysis shows that transformer models, with their attention mechanisms and ability to handle information across multiple formats, perform much better than traditional approaches when it comes to combining data from sources like physiological signals, medical imaging, and clinical records. However, we also identify several challenges: high computational demands for edge devices, limited interpretability, a limited multimodal dataset, and infrastructure barriers in under-resourced regions. To address these challenges, we highlight future directions such as creating lightweight transformer models, using privacy-preserving federated learning, and developing unified multimodal pretraining strategies. This review aims to provide a roadmap for building fair, scalable, and low-latency AI solutions for real-time cardiovascular prediction, offering valuable insights for both researchers and healthcare system developers.

In this chapter we have tried to explore how IOT, connected devices and automation shall bring about a revolution in the field of agriculture and tremendously improve nearly every facet of it. IOT in agriculture is the amalgamation of Information technology, telecommunications and sensor technology. Agriculture has been a neglected field in India as far as automation and technological use and applicability are concerned due to lack of funds for technological expansion and limited technical expertise for implementation of the available technologies. Agro based economy like India being perfect for implementation of smart agriculture; Organizations are focusing on improving accessibility of low cost IOT sensors and providing more scalable and cost effective solutions. Smart farming techniques encompass a number of applications such as Precision Farming, Smart Greenhouses, Livestock Management, and Agriculture Drones and Farm Management Systems. The use of smart & connected devices in agriculture have helped farmers gain better control over their farm produce, help make climatic predictions and raise their livestock’s more efficiently. The use of IOT for digitization of farms has caught the attention of Government of India and has been included in its policy for Digital India. A recent study by Statistic shows that smart agriculture shall take up $26.76 billion of global market size by 2020. According to a report by NASSCOM, India has around 40 startups dealing in smart agriculture.