Human Activity Recognition (HAR) has become a cornerstone in the dynamic development of the Industrial Internet of Things (IIoT). This study introduces an extensive framework aimed at embedding HAR functionality into industrial settings to promote workplace safety, streamline operations, and facilitate predictive maintenance. Through AI techniques, HAR problems can be achieved with high accuracy. However, traditional AI models require centrally trained data on remotely powerful cloud servers. This leads to issues with privacy and security of health records and increases latency. To address this problem, the Federated Learning (FL) technique was proposed. FL allows distributed training on the patient’s IoT devices and serves as the communication mechanism between local devices and the FL aggregator. Thanks to this architecture, the health data needs only to be stored locally on its devices without being uploaded to data centers, thus ensuring security and reducing service response times and computational costs. In this study, we implement FedANN and FedConvNN independently in a federated learning setting to address human activity recognition problems toward real-time applications. Finally, we evaluate the effectiveness of the based on the variation initiation of the number of different training clients. The results show that the FedConvNN solution improves accuracy and reduces model and communication complexity compared to FedANN and centralized training models, with the potential for real-time deployment in HAR tasks. Our code is available on our GitHub repository: https://github.com/itsminhcs/Fedavg-HAR.git.

Multi-label text classification aims to assign a set of most relevant labels to a given document. To build such a classifier, apart from demanding an efficient document representation, capturing label information for classification performance improvement is still challenging. In this paper, we propose a novel model based on a graph convolutional network to model label correlation. To do that, we design a correlation matrix from labels in a data-driven way. The learned label correlations are then fused with fine-grained document information extracted by a RoBERTa-based subnet for classification. Furthermore, we introduce a simple mechanism to make the label correlation matrix more effective in propagating information among label nodes. We first normalize the correlation matrix to deal with the highly skewed problem and then filter noisy edges to alleviate the long-tailed distribution problem. Evaluation results show that our model achieves competitive results compared to existing state-of-the-art methods. Ablation studies are also conducted to explore the proposed model's behaviors.

Solving Constraint Satisfaction Problems (CSPs) through Boolean Satisfiability (SAT) requires suitable encodings for translating CSPs to equivalent SAT instances that can not only be efficiently generated, but also efficiently solved by SAT solvers. In this paper we investigate hierarchical and hybrid encodings, as proposed by Velev, namely a previously studied log-direct encoding, and a new combination, the log-order encoding. Experiments on different domain problems with these hierarchical encodings demonstrate their significant promise in practice. Our experiments show that the log-direct encoding significantly outperforms the direct encoding (typically by one or two orders of magnitude) taking advantage not only of the more concise representation, but also of the better capability of the log-direct encoding to represent interval variables. We also show that the log-order encoding is competitive with the order encoding, although more studies are required to understand the tradeoff between the fewer variables and longer clauses in the former, when expressing complex CSP constraints.