ABSTRACT Entity relationship extraction (ERE) plays a crucial role in natural language processing and has great practical significance, especially when applied to electronic medical records (EMRs). Traditional methods usually use hierarchical structures or local decision‐making processes and often perform poorly in complex situations such as Single Entity Overlap (SEO) and Entity Pair Overlap (EPO), resulting in inaccuracies and loss of information. To overcome these limitations, this study introduced URRel, a combination model of entity and relationship extraction that combines unified coding and reinforcement learning (RL) techniques. URRel uses a semantic fusion approach inspired by UniRel to merge sentences and relational descriptions into a single representation, allowing the ability to simultaneously capture entities and their relationships. An interaction matrix was also built to model multiple dependencies between entity pairs and between entities and relationships to improve the model's ability to understand complex relational patterns. URRel views relationship extraction as a sequential decision‐making process and uses an Actor‐Critic RL algorithm to dynamically optimise the extraction strategy and reduce the possibility of decoding errors accumulating over time. Experimental results on the China coronary EMR private dataset (CACMeD) and the publicly available DuIE relational extraction dataset show that URRel significantly outperforms several leading baseline models, yielding F1 scores of 80.35% and 77.87% on DuIE and private dataset CACMeD. This research emphasises the effectiveness of combining unified semantic representation with RL, providing a novel framework for solving complex information extraction challenges.

Transformer has widely been applied in various low-vision tasks, achieving significant strides in single-image super-resolution (SISR). However, its low-pass characteristic still limits the ability of Transformer-based models to represent rich texture details in images. Although some recent works try to combine the Transformer with CNN, the high-frequency details are still easily ignored in deep layers. Additionally, the quadratic computational complexity of the attention mechanism also restricts its application in many low-level tasks. In this paper, we propose HyLiteNet, a lightweight hybrid network for SISR, combining CNN and Mamba to efficiently capture high and low-frequency information. HyLiteNet integrates the Non-causal Selective Scan (NSS) module and CNN to model global relationships and local textures. Meanwhile, a High-freq Flow component ensures deep propagation of high-frequency details. Extensive experiments demonstrate that HyLiteNet achieves state-of-the-art performance with low computational cost, outperforming existing baselines in SISR.

In wireless sensor networks, network security against virus propagation is one of the challenges with the applications. In severe cases, the network system may become paralyzed. In order to study the process of virus propagation in wireless sensor networks with the media access control mechanism, this paper uses the susceptible-infectious-removed (SIR) model to analyze the spreading process. It provides a theoretical basis for the development of virus immune mechanisms to solve network virus attack hidden dangers. The research shows that the media access control (MAC) mechanism in the wireless sensor network can inhibit the process of virus propagation, reduce the network virus propagating speed, and decrease the scale of infected nodes. The listen/sleep duty cycle of this mechanism will affect the suppression effect of virus propagation. The smaller the listen/sleep duty cycle, the stronger the suppression effect. Energy consumption has a peak value under specific infection probability. Meanwhile, it is also found that the spreading scale of the virus in wireless sensor networks can be effectively inhibited by the MAC mechanism.

The segmentation of brain tumor plays an important role in diagnosis, treatment planning, and surgical simulation. The precise segmentation of brain tumor can help clinicians obtain its location, size, and shape information. We propose a fully automatic brain tumor segmentation method based on kernel sparse coding. It is validated with 3D multiple-modality magnetic resonance imaging (MRI). In this method, MRI images are pre-processed first to reduce the noise, and then kernel dictionary learning is used to extract the nonlinear features to construct five adaptive dictionaries for healthy tissues, necrosis, edema, non-enhancing tumor, and enhancing tumor tissues. Sparse coding is performed on the feature vectors extracted from the original MRI images, which are a patch of m×m×m around the voxel. A kernel-clustering algorithm based on dictionary learning is developed to code the voxels. In the end, morphological filtering is used to fill in the area among multiple connected components to improve the segmentation quality. To assess the segmentation performance, the segmentation results are uploaded to the online evaluation system where the evaluation metrics dice score, positive predictive value (PPV), sensitivity, and kappa are used. The results demonstrate that the proposed method has good performance on the complete tumor region (dice: 0.83; PPV: 0.84; sensitivity: 0.82), while slightly worse performance on the tumor core (dice: 0.69; PPV: 0.76; sensitivity: 0.80) and enhancing tumor (dice: 0.58; PPV: 0.60; sensitivity: 0.65). It is competitive to the other groups in the brain tumor segmentation challenge. Therefore, it is a potential method in differentiation of healthy and pathological tissues.