Sema Bayraktar

Next-generation Intelligent Transportation Systems (ITSs) require ultra-reliable, low-latency Vehicle-to-Everything (V2X) communication frameworks that support safety-critical vehicular services. Conventional centralized, monolithic architectures suffer from excessive transmission latency, limited scalability, and authentication overheads that are ill-suited to the highly dynamic and dense vehicular environment. This paper presents a secure and low-latency edge–cloud architecture for intelligent V2X communications based on a lightweight microservice-driven design paradigm. A formal latency-constrained model is presented to ensure that the end-to-end delay satisfies tight real-time constraints. The proposed framework is lightweight and includes HMAC-based authentication, nonce-based replay protection, timestamp validation, and short-lived encrypted session tokens in a stateless architecture using the Laravel framework deployed at the edge layer. Security validation is performed at edge gateways, and asynchronous SQLite-backed job queues support non-blocking telemetry processing and scalable service orchestration. Experimental evaluation shows that the edge-based deployment achieves a mean response time of 2.58 ms with small variance under repeated request conditions, while centralized processing exhibits significantly higher latency. The results demonstrate that secure authentication and telemetry exchange can be achieved without breaching strict latency requirements. The proposed solution creates a deployable, scalable, and security-aware foundation for next-generation V2X ecosystems and Intelligent Transportation Systems (ITSs) in real time.

This study introduces a hybrid framework that combines Multi-Agent Reinforcement Learning (MARL) based on Mean Field Theory (CSMF) with Large Language Models (LLMs) to provide optimized and explainable charging station recommendations for EV users in Turkey. By integrating real-time data from 11,702 stations and 30,941 sockets via the Energy Market Regulatory Authority of Türkiye (EMRA) API, the system captures dynamic information on station availability, socket types, charging power, and pricing, enabling context-aware, user-centric guidance. The proposed architecture follows a Centralized Training and Decentralized Execution (CTDE) framework, in which multiple agents make local decisions while a centralized critic evaluates global states to improve policy learning and reduce congestion. CSMF reduces computational complexity and lowers the average waiting time by 71.8%, while the LLM component translates technical optimization outcomes into human-readable explanations, enhancing trust and transparency. A React Native mobile application demonstrates end-to-end usability, allowing users to filter stations by distance, cost, socket type, and green energy usage, with real-time availability updates. Experimental results show that the hybrid CSMF + LLM approach achieves a 100% Available Facility Rate (AFR) while maintaining low response time (180 ms), outperforming conventional nearest-station, random, and basic MARL methods. By combining technical efficiency with explainable recommendations and real-world deployment, this study presents a scalable, practical, and trustworthy solution for EV charging infrastructure management, offering both academic contributions and a ready-to-use application for drivers.

Background/Objectives: Accurate estimation of meal composition from food images can support safer and more reliable insulin bolus decision-making for individuals with Type 1 diabetes. Existing food recognition and nutrition estimation systems are often designed for general dietary logging and do not directly integrate food analysis with personalized insulin therapy parameters. Methods: This study presents an image-based nutrition estimation and insulin decision-support module developed within the AI-assisted Diabetes Care (AIDCARE) platform. The proposed system uses a convolutional neural network (CNN) to classify food items from a single meal image, and retrieves reference nutritional values from a food composition database. A separate multimodal large language model (MLLM)-based estimation component is then used to estimate portion size, allowing carbohydrate and nutrient values to be scaled according to the observed serving. Results: A curated food image dataset containing 40 food categories was used to evaluate three CNN architectures: ResNet50, Inception V3, and EfficientNet-B0. EfficientNet-B0 achieved the best classification performance, with 94.91% validation accuracy, 95.55% precision, 94.87% recall, and 94.90% F1-score. The portion-estimation component achieved an MAE of 12.27 g and an RMSE of 15.11 g. The estimated carbohydrate value is combined with user-specific clinical parameters, including the insulin-to-carbohydrate ratio and insulin sensitivity factor, to generate advisory bolus guidance. To support safety, the system requires user confirmation or correction of the recognized food category and estimated portion before insulin guidance is displayed. Conclusions: The proposed system is intended for advisory decision support only and is not designed to replace clinical judgment or autonomous insulin delivery systems.

Accurate estimation of meal composition from food images can support safer and more reliable insulin bolus decision-making for individuals with Type 1 diabetes. Existing food recognition and nutrition estimation systems are often designed for general dietary logging and do not directly integrate food analysis with personalized insulin therapy parameters. This study presents an image-based nutrition estimation and insulin decision-support module developed within the AI-assisted Diabetes Care (AIDCARE) platform. The proposed system uses a convolutional neural network (CNN) to classify food items from a single meal image and retrieves reference nutritional values from a food composition database. A separate multimodal large language model (MLLM)-based estimation component is then used to estimate portion size, allowing carbohydrate and nutrient values to be scaled according to the observed serving. A curated food image dataset containing 40 food categories was used to evaluate three CNN architectures: ResNet50, Inception V3, and EfficientNet-B0. EfficientNet-B0 achieved the best classification performance, with 94.91% validation accuracy, 95.55% precision, 94.87% recall, and 94.90% F1-score. The portion-estimation component achieved an MAE of 12.27 g and an RMSE of 15.11 g. The estimated carbohydrate value is combined with user-specific clinical parameters, including the insulin-to-carbohydrate ratio and insulin sensitivity factor, to generate advisory bolus guidance. To support safety, the system requires user confirmation or correction of the recognized food category and estimated portion before insulin guidance is displayed. The proposed system is intended for advisory decision support only and is not designed to replace clinical judgment or autonomous insulin delivery systems.

We consider an initial-boundary value problem for a fourth-order nonlinear parabolic equation with constant coefficients. Our primary focus lies in establishing a priori estimates for the solution to this equation, with a particular emphasis on its continuous dependence on both the initial data and parameters. Using energy estimates, we establish the continuous dependency for both the solution and its gradient concerning the fourth-order nonlinear parabolic equation.