This paper presents DriveCoach, an intelligent driving assistance and coaching system designed to strengthen safe driving skills through structured, learning-oriented intervention. The system combines risk assessment and adaptive assistance with a coaching-centered improvement cycle in which risk driving skills are diagnosed, addressed through real-time feedback, and reinforced via tailored post-drive coaching. Using the CARLA driving simulator, we conducted a mixed-method user study to evaluate DriveCoach across four representative driving skills: maintaining safe distance, responding to oncoming vehicles, handling adjacent vehicles, and negotiating intersections. Quantitative analyses demonstrated significant reductions in risk-related events when drivers received real-time assistance and notable improvements in post-coaching performance, indicating short-term skill retention and transfer. Complementary qualitative results revealed strong user acceptance and positive perceptions of the system’s usability and coaching effectiveness. These findings highlight DriveCoach as a human-centered AI system that fosters safer, more reflective driving, contributing to the design of co-adaptive driver support systems that integrate behavioral assessment with personalized coaching.

The Federated Edge Artificial Intelligence (Edge AI) deploys AI applications on Internet-of-Things (IoT) devices, addressing data privacy concerns in the real world. To achieve effective Federated Learning (FL), three challenges must be addressed: i) limited computing power on devices, ii) non-uniform impacts on devices, and iii) adaptability to changing network conditions. This study introduces a new algorithm called Adaptive Offloading Point (AOP), designed to accelerate local training on constrained devices. It decomposes deep neural network (DNN) layer blocks, enabling training on both client and server sides. The novelty of the proposed method lies in using a reinforcement learning-based Gaussian mixture model (GMM) clustering to dynamically determine DNN layer offloading, addressing nullability, computation uniformity, training time, and network bandwidth variation issues. Experimental results on real devices, using vision transformer (ViT) models with an identification image dataset, show that AOP's training time is significantly faster than that of previous baseline methods.

The study discusses challenges in deploying Artificial Intelligence (AI) on Internet-of-Things (IoT) devices and introduces a solution called Adaptive Offloading Point (AOP) for Federated Learning (FL) in Edge AI. AOP accelerates local training on resource-constrained devices and uses reinforcement learning-based clustering (RL) to determine which deep neural network (DNN) layers to offload to the server. Test results show that, in the VIT transformer model, AOP significantly reduces training time compared to classical FL and the baseline method called FedAdapt, making it a promising solution for Edge AI applications.

Split learning (SL) is a popular distributed machine learning (ML) method used to enable ML. It divides a neural network based model into subnetworks. Then, it separately trains the subnetworks on distributed parties (e.g., client and server). In distributed ML, data are generated and collected on the client-side. In contrast, the collected data are processed using an application deployed on the server side. However, when applied in practice using Internet of things systems and clients, numerous obstacles occur because of limited configuration and resources. Dividing neural networks in the SL is the biggest problem and an open question in numerous studies. This study introduces splitDyn, which is a new dynamic SL solution to solve the aforementioned problems. This method provides a solution for eliminating their inherent drawbacks. The main idea is to apply a Round-Robin schedule to select the client for the training process. Then, the next idea is to use the Hungarian optimization algorithm to assign a layer to a client and enhance the accuracy. The proposed method reasonably achieved better accuracy and reduced processing time than the other learning models. Furthermore, it applies the incident datasets to predict the incident event and in edge computing for edge artificial intelligence (AI) applications.

Federated learning (FL) has been used to protect data contributors’ privacy by allowing training at clients and then feeding back machine learning models to servers for aggregation. Conventional methods of FL aggregation methods average model weights to produce a fused global model. However, in real-world applications in cyber-space systems, which often have heterogeneous Internet of Things data configuration and collection, FL encounters obstacles with non-independent and identically distributed (Non-IID) data. The main problem is the aggregated global models deviating from the optimal model trained on centralized servers. According to recent research, most Non-IID FL aggregation methods attempt to direct the movement of gradients to the optimal one using differentiation from trained models. In this paper, we propose a framework for using feature probability distribution in aggregation calculation. The proposed aggregation algorithm shows robustness on different Non-IID datasets and outperforms state-of-the-art methods in various FL experiments.

Due to the rapid recent development of cloud-edge networks, smart devices can facilitate rapid access to patients’ health information. Success has been achieved in the healthcare sector with the training of a federated learning (FL) model on large amounts of the personal data of users. However, some challenges remain that other FL models have not yet addressed. Firstly, FL models with computational parameters are very complex, which results in a high communication cost in the cloud-edge network. Furthermore, trained models in the cloud are not personalized. If personalization is present, the models do not provide practical solutions to fine-tune parameters in order to accurately predict performance in health monitoring. To address the above challenges, this paper presents the Fed xData framework for contextual health monitoring in cloud-edge networks. The Fed xData framework introduces a continuous data balancing supplemented structure using the RandomOverSample method, which solves all data classes. The FL model is an encode depth convolutional network (EDCN) model designed for both server and client. It solves various problems, for instance by using the fine-tuning model to increase personalization and solving not independent and identically (Non-IID) distribution problems regarding user health. Test results based on human activity recognition indicate that Fed xData is far superior to others for use in general centralized learning models and FL models.

Air pollution prediction for smart city applications has been attracted in artificial intelligence research to overcome problems to the health of citizen. Conventionally, environmental IoT data is gathered from monitoring station sensors then is sent to servers for centralized predictive model training at a whole region. That causes latency issues in data transmission from IoT devices to cloud servers. This paper describes federated learning paradigm approach for air pollution prediction model training on environmental monitoring sensor data. In the research, we design distributed learning framework that assists cooperative training among participants from different spatial areas such as cities and prefectures. At each area, Convolutional Recurrent Neural Networks (CRNN) are trained locally aiming to predict local Oxidant warning level while aggregated global model enhances distilled knowledge from all areas of a region. The research illustrates that designed common parts of CRNN can be fused globally meanwhile adaptive structure at predictive part of the deep neural network model can capture different environmental monitoring stations configuration at local areas. Some experiment results also hint methods to keep balance between federated learning synchronous training rounds and local deep neural network training epochs to maximize accuracy of the whole federated learning system. The results also prove that new participating areas can train and quickly obtain optimized local models by using transferred common global model.

In multitask reinforcement learning, tasks often have sub-tasks that share the same solution, even though the overall tasks are different. If the shared-portions could be effectively identified, then the learning process could be improved since all the samples between tasks in the shared space could be used. In this paper, we propose a Sharing Experience Framework (SEF) for simultaneously training of multiple tasks. In SEF, a confidence sharing agent uses task-specific rewards from the environment to identify similar parts that should be shared across tasks and defines those parts as shared-regions between tasks. The shared-regions are expected to guide task-policies sharing their experience during the learning process. The experiments highlight that our framework improves the performance and the stability of learning task-policies, and is possible to help task-policies avoid local optimums.

Recently, a number of methods have been conducted by combining deep neural network and reinforcement learning to solve problems. Neural networks have strong ability to deal with high dimensional data and a good mean to learn features while reinforcement learning allows a system to learn by experience optimal action control and adjust its behavior to new environments. It thus may be used in mobile robot navigation tasks with perceptron as deep neural network and learning to control by reinforcement learning. We first investigate some issues in agents learning to navigate in indoor environments such as how robot memorize vision information and how it discovers the environments. We show that agents can rely on some position with vision information in navigation task such as reduce training time and locating object efficiently. Some approaches based on deep reinforcement learning are step by step discussed and proposed to deal with each problem in target object navigation task. The key ideas include adding checking point as vision memory and using auxiliary learning task to enhance agents to discover the environment.

We have developed a rough set model for analyzing an information system in which some conditions as well as decision values, are missing. Current studies have focused mainly on the missing of condition data but seem to ignore the missing of decision data. The common approach is to remove objects with no decision values because such objects are apparently considered fruitless from the decision-making standpoint. However, this deletion may lead to the risk of information loss. We observe that such a situation is somewhat similar to the semi-supervised situation in the sense that some objects are characterized by complete decision data while some are not. Considering both kinds of objects from a probabilistic view, we predict potential candidates for missing values by comparing measurements of two factors, local decision belief and universal decision belief, with a parameter threshold α. These possible decision candidates help to form a relative dissimilarity relation, which measures the unlikeness of pairs of objects rather than their likeness. Contrasting with the other approaches, rough set definitions based on this relation do not approximate the target set but its complement instead. The knowledge acquisition induced by the common approach and the proposed approach is compared, and the result shows that the latter can overcome some limitations of the former. This approach is new and flexible to deal with missing decision information.

The original rough set theory deals with precise and complete data, while real applications frequently contain imperfect information. A typical imperfect data studied in rough set research is the missing values. Though there are many ideas proposed to solve the issue in the literature, the paper adopts a probabilistic approach, because it can incorporate other types of imperfect data including imprecise and uncertain values in a single approach. The paper first discusses probabilities of attribute values assuming different type of attributes in real applications, and proposes a generalized method of probability of matching. It also discusses the case of continuous data as well as discrete one. The proposed probability of matching could be used for defining valued tolerance/similarity relations in rough set approaches.