In the context of urban informatization, meeting the stringent requirements of emergency communication presents a significant challenge for Urban Emergency Communication Networks (UECNs). Mobile ad hoc networks deploy...
详细信息
In the context of urban informatization, meeting the stringent requirements of emergency communication presents a significant challenge for Urban Emergency Communication Networks (UECNs). Mobile ad hoc networks deployed in these environments often experience node degradation and link disruptions due to the complex urban landscape, leading to frequent communication failures. This paper introduces a novel resilient routing strategy, termed Deep Reinforcement Learning-based Resilient Routing (DRLRR). The proposed routing strategy first utilizes node and link state information to accurately characterize dynamic changes in network topology. The routing decision-making process is then formalized as a Markov decision process, integrating multiple performance metrics into a reward function tailored for the specific demands of urban emergency communications. By leveraging deep reinforcement learning, DRLRR effectively adapts to the complexities of urban environment, enabling intelligent and optimal route selection during network topology fluctuations to ensure seamless data transmission during emergencies. Comparative simulations conducted using NS3(Network simulator 3) demonstrate that DRLRR significantly outperforms three other routing protocols, achieving notable improvements in packet delivery rate, average end-to-end delay, and throughput, thus fulfilling the requirements for reliable and consistent communication in urban emergency scenarios.
Identifying schizophrenia (SZ) using brain networks has attracted increasing attention since features extracted from these networks can provide potential biomarkers. However, most existing feature extraction methods d...
详细信息
Identifying schizophrenia (SZ) using brain networks has attracted increasing attention since features extracted from these networks can provide potential biomarkers. However, most existing feature extraction methods designed for brain networks are based on traditional learning frameworks without considering anatomical information. In this paper, we propose a novel anatomy-guided brain network transformer (AGBN-Transformer) to automatically learn features of brain networks for SZ diagnosis. Our AGBN-Transformer embeds two types of anatomical information (i.e., functional hemispheric asymmetries and the spatial location of brain regions) in the process of feature learning. Specifically, we first construct a functional connectivity matrix for each subject and divide this matrix into two parts based on the anatomical information that the brain can be divided into the right and the left hemispheres. Then, we design a Siamese neural network (SNN) consisting of Transformer encoders to extract features of brain networks. This SNN contains two structurally consistent subnetworks with different inputs, which can model the functional hemispheric asymmetries and learn the features of two hemispheres, respectively. Also, we add the spatial location of brain regions into Transformer encoders by using anew positional embedding operation, which can further improve the learning capability of our model. Finally, we fuse features from two hemispheres by using a bilinear pooling layer, and fed these fused features into a fully connected layer for SZ diagnosis. Our experimental results on five multi-centre cohorts with 773 subjects demonstrated that our AGBN-Transformer is superior to several state-of-the-art methods, and the anatomy-guided information can improve the accuracy of identifying brain diseases.
A sentiment analysis scheme for image and text comments based on multimodal deep learning and spatiotemporal attention is proposed to address the issues of incomplete spatiotemporal considerations, incomplete implemen...
详细信息
ISBN:
(数字)9798331522667
ISBN:
(纸本)9798331522674
A sentiment analysis scheme for image and text comments based on multimodal deep learning and spatiotemporal attention is proposed to address the issues of incomplete spatiotemporal considerations, incomplete implementation details, and cutting-edge theoretical algorithms in graphic and textual sentiment analysis schemes. The proposed model has clear layering including data preprocessing layer, modal encoding layer, modal fusion layer, sentiment classification layer, loss function and optimizer, evaluation and feedback. The implementation details of each layer are introduced. The entire scheme model incorporates Multimodal Fusion Neural Network (MFNN) deep learning and spatiotemporal attention mechanism, which makes the scheme perform well in terms of security, robustness and performance, making up for the shortcomings of existing research schemes.
This study identifies key factors affecting dust susceptibility in Gavkhouni Basin, central Iran, using three feature selection algorithms and a perceptual neural network model. Accuracy assessment statistics were use...
详细信息
This study identifies key factors affecting dust susceptibility in Gavkhouni Basin, central Iran, using three feature selection algorithms and a perceptual neural network model. Accuracy assessment statistics were used to evaluate the prediction capabilities of the models. The aerosol optical depth dataset validated the dust-generating area map, with the permutation feature importance method prioritizing factors controlling dust events. Using the variables selected by the genetic algorithm improved the coefficient of explanation by 31% compared to relief, and 19% compared to ElasticNet algorithm. The genetic algorithm proved effective in identifying variables that significantly enhanced model accuracy in high-risk zones (precision = 0.75, recall = 0.71, and F1 = 0.73). The study found that topographic diversity, geology, soil sand content, precipitation, wind speed, soil salinity, soil subsidence, vegetation cover, slope, and soil moisture were key environmental factors. These findings are very important for the formulation of specific measures for improving air quality and limiting dust-related effects as a key factor in the sustainable management of vulnerable ecosystems.
Data anonymization is one of the common techniques for ensuring data security and privacy. However, the existing anonymization techniques often suffer lower execution efficiency and unnecessary information loss when d...
详细信息
Data anonymization is one of the common techniques for ensuring data security and privacy. However, the existing anonymization techniques often suffer lower execution efficiency and unnecessary information loss when dealing with complex data. Therefore, we propose a dynamic anonymity privacy-preserving model based on hierarchical sequential three-way decisions. Specifically, we first divide the data into multiple granularity spaces by attributes and dynamically process the data in the granularity spaces. Then, in a single granularity space, we construct a generalization hierarchy for the data based on the attributes generalization trees and divide it into the positive, negative and boundary regions based on anonymous parameter. Next, we can acquire the positive and boundary regions by generalization and dynamically update the processed data at the next granularity. After that, we suppress the data in the final negative and boundary regions while releasing the positive region. To further improve data availability, we combine the idea of differential privacy by adding noise data to the final boundary region enabling its release and propose an enhanced anonymity model. Finally, we compare our proposed algorithms with other methods on six datasets. Experimental results show that our method effectively reduces processing costs, improves data usability and protects data privacy.
Personalized federated learning (PFL) has garnered attention due to its capability to address statistical heterogeneity among clients. Typically, prevailing PFL methods aggregate a single global model for personalizat...
详细信息
Personalized federated learning (PFL) has garnered attention due to its capability to address statistical heterogeneity among clients. Typically, prevailing PFL methods aggregate a single global model for personalization, which maybe inadequate for clients with diverse data distributions. Furthermore, in the local update, each private dataset is used to optimize the model independently, which increases the risk of overfitting the current data distribution and losing previously acquired knowledge, resulting in knowledge forgetting. In this study, a personalized federated learning with multiple classifier aggregation (FedMCA) method is proposed. FedMCA splits the client model into its head and base, optimizing them respectively using an alternating strategy that sequentially targets the head and base. Initially, to address the suboptimal model problem, the proposed method aggregates multiple classifiers using data distribution and employs knowledge distillation to impart positive and negative classifier knowledge for learning the most suitable personalized model head. Additionally, to mitigate knowledge forgetting, a learnable personalization layer is introduced, and hidden loss is utilized to learn the knowledge of the global base and prevent overfitting of the model base. The experimental results demonstrate that the proposed method achieves competitive performance across various benchmarks, outperforming most state-of-the-art PFL algorithms. The source code is publicly available at https://***/xiaye-maker/FedMCA.
As the society industrialized, mathematical modeling and simulation become increasingly important in the product design. At present, the multidomain unified modeling with Modelica is a mainstream technology in the fie...
详细信息
ISBN:
(纸本)9783642343803
As the society industrialized, mathematical modeling and simulation become increasingly important in the product design. At present, the multidomain unified modeling with Modelica is a mainstream technology in the field of complex systems. Modeling of complex physical systems with Modelica often produces a high-index differential algebraic equation (DAE) system. It needs to be transformed to low-index DAE before solving it. The structure index reduction algorithm is one of the popular index reduction methods. But in some special circumstances, its solution may be incorrect. At present, combinatorial relaxation algorithm is a widely used method for solving the problem. Solving maximum weighted matching is one of important problems of the combinatorial relaxation algorithm. This paper describes the combinatorial relaxation algorithm and proposes three different implementations of Hungarian algorithm for the maximum weighted matching problem. The theory results are consistent with the experiment results,
This study aims to develop a safe and effective multi-parameter MRI-based molecular subtype prediction model for breast cancer, emphasizing the advantages of this multi-parameter approach over single-parameter models....
详细信息
This study aims to develop a safe and effective multi-parameter MRI-based molecular subtype prediction model for breast cancer, emphasizing the advantages of this multi-parameter approach over single-parameter models. This study retrospectively collected and organized MRI data from 318 breast cancer patients at Liaoning Provincial Cancer Hospital, including dynamic contrast-enhanced MRI (DCE-MRI, abbreviated as DCE), diffusion weighted MRI (DWI-MRI, abbreviated as DWI), T1-weighted MRI (T1WI-MRI, abbreviated as T1WI), and T2-weighted MRI (T2WI-MRI, abbreviated as T2WI). The dataset includes 57 cases of Luminal A type, 162 cases of Luminal B type, 46 cases of human epidermal growth factor receptor-2 (HER-2) overexpression type, and 53 cases of triple-negative type. Predictive models were established using four single-parameter MRI methods and seven multi-parametric MRI methods, employing quantitative feature extraction. Model performance was evaluated through the area under the curve (AUC) and balanced accuracy (BA). In the single-parameter MRI models, the T2WI-MRI model demonstrated the best predictive performance for four-class classification, with average AUC and BA values of 0.794 and 0.518, respectively. In contrast, the multi-parameter model combining DWI+T2WI exhibited even better performance, with these metrics reaching 0.823 and 0.565, respectively. The multi-parameter feature fusion model for breast cancer molecular subtypes prediction, utilizing DWI+T2WI, exhibited superior BA and AUC values compared to models based solely on single-parameter MRI. It showed enhanced predictive capabilities for Luminal A, Luminal B, HER-2 overexpression, and triple-negative subtypes. Therefore, the multi-parameter MRI-based model offers improved predictive performance over single-parameter models.
Shot peening (SP) is a widely used surface treatment technique that enhances the mechanical performance of materials, notably improving fatigue resistance by inducing compressive residual stresses (RS) and modifying s...
详细信息
Two personification strategies are presented, which yield a highly efficient and practical algorithm for solving one of the NP hard problems——circles packing problem on the basis of the quasi-physical algorithm. A v...
详细信息
Two personification strategies are presented, which yield a highly efficient and practical algorithm for solving one of the NP hard problems——circles packing problem on the basis of the quasi-physical algorithm. A very clever polynomial time complexity degree approximate algorithm for solving this problem has been reported by Dorit *** and Wolfgang Maass in J. ACM. Their algorithm is extremely thorough-going and of great theoretical significance. But, just as they pointed out, their algorithm is feasible only in conception and even for examples frequently encountered in everyday life and of small scale, it is the case more often than not that up to a million years would be needed to perform calculations with this algorithm. It is suggested toward the end of their paper that a heuristic algorithm of higher practical effectiveness should be sought out. A direct response to their suggestion is intented to provide.
暂无评论