检索结果-内蒙古大学图书馆

arXiv 2025年

作者： Ou, Yixin Yao, Yunzhi Zhang, Ningyu Jin, Hui Sun, Jiacheng Deng, Shumin Li, Zhenguo Chen, Huajun Zhejiang University China Huawei Noah’s Ark Lab Canada National University of Singapore NUS-NCS Joint Lab Singapore Zhejiang Key Laboratory of Big Data Intelligent Computing China

Despite exceptional capabilities in knowledge-intensive tasks, Large Language Models (LLMs) face a critical gap in understanding how they internalize new knowledge, particularly how to structurally embed acquired knowledge in their neural computations. We address this issue through the lens of knowledge circuit evolution, identifying computational subgraphs that facilitate knowledge storage and processing. Our systematic analysis of circuit evolution throughout continual pre-training reveals several key findings: (1) the acquisition of new knowledge is influenced by its relevance to preexisting knowledge;(2) the evolution of knowledge circuits exhibits a distinct phase shift from formation to optimization;(3) the evolution of knowledge circuits follows a deep-to-shallow pattern. These insights not only advance our theoretical understanding of the mechanisms of new knowledge acquisition in LLMs, but also provide potential implications for improving continual pre-training strategies to enhance model performance Copyright © 2025, The Authors. All rights reserved.

关键词： Knowledge acquisition

来源：评论

学校读者我要写书评

暂无评论

Any-Angle Routing Algorithm for Microfluidic Biochips Driven by Flow Path

引用

Jisuanji Yanjiu yu Fazhan/Computer Research and Development 2025年第4期62卷 978-988页

作者： Youlin, Pan Shuai, Guo Xing, Huang Genggeng, Liu College of Computer and Data Science Fuzhou University Fuzhou350116 China Engineering Research Center of Big Data Intelligence Fuzhou University Ministry of Education Fuzhou350116 China Fujian Key Laboratory of Network Computing and Intelligent Information Processing Fuzhou University Fuzhou350116 China School of Computer Science Northwestern Polytechnical University Xi’an710072 China

Continuous-flow microfluidic biochips (CFMBs) have become a hot research topic in recent years due to their ability to perform biochemical assays automatically and efficiently. For the first time, PathDriver+ takes the requirements of the actual fluid transportation into account in the design process of CFMBs and implements the actual fluid transport and removal, and plans separate flow paths for each transport task, which have been neglected in previous work. However, PathDriver+ does not take full advantage of the flexibility of CFMBs routing because it only considers the optimization of flow channel length for the global routing in the mesh model, except for the detailed routing. In addition, PathDriver+ only considers the X architecture, while the existing work shows that the any-angle routing can utilize the routing resources more efficiently and shorten the flow channel length. To address the above issues, we propose a flow path-driven arbitrary angle routing algorithm, which can improve the utilization of routing resources and reduce the flow channel length while considering the actual fluid transportation requirements. The proposed algorithm constructs a search graph based on constrained Delaunay triangulation to improve the search efficiency of routing solutions while ensuring the routing quality. Then, a Dijkstra-based flow path routing method is used on the constructed search graph to generate a routing result with a short channel length quickly. In addition, in the routing process, channel reuse strategy and intersection optimization strategy are proposed for the flow path reuse and intersection number optimization problems, respectively, to further improve the quality of routing results. The experimental results show that compared with the latest work PathDriver+, the length of channels, the number of ports used, and the number of channel intersections are significantly reduced by 33.21%, 11.04%, and 44.79%, respectively, and the channel reuse rate is i

关键词： High level synthesis

来源：评论

学校读者我要写书评

暂无评论

Control Logic Routing for Continuous-Flow Microfluidic Biochips Based on Deep Reinforcement Learning

引用

Jisuanji Yanjiu yu Fazhan/Computer Research and Development 2025年第4期62卷 950-962页

作者： Cai, Huayang Huang, Xing Liu, Genggeng College of Computer and Data Science Fuzhou University Fuzhou350116 China Engineering Research Center of Big Data Intelligence Fuzhou University Ministry of Education Fuzhou350116 China Fujian Key Laboratory of Network Computing and Intelligent Information Processing Fuzhou University Fuzhou350116 China School of Computer Science Northwestern Polytechnical University Xi’an710072 China

With the advancement of electronic design automation, continuous-flow microfluidic biochips have become one of the most promising platforms for biochemical experiments. This chip manipulates fluid samples in milliliters or nanoliters by utilizing internal microvalves and microchannels, and thus automatically performs basic biochemical experiments, such as mixing and detection. To achieve the correct bioassay function, the microvalves deployed inside the chip are usually managed by a multiplexer-based control logic, and valves receive control signals from a core input through the control channel for accurate switching. Since biochemical reactions typically require high sensitivity, the length of control paths connecting each valve needs to be reduced to ensure immediate signal propagation, and thus to reduce the signal propagation delay. In addition, to reduce the fabrication cost of chips, a vital issue to be addressed in the logic architecture design is how to effectively reduce the total channel length within the control logic. To address the above issues, we propose a deep reinforcement learning-based control logic routing algorithm to minimize the signal propagation delay and total control channel length, thereby automatically constructing an efficient control channel network. The algorithm employs the dueling deep Q-network architecture as the agent of the deep reinforcement learning framework to evaluate the tradeoff between signal propagation delay and total channel length. Besides, the diagonal channel routing is implemented for the first time for control logic, thus fundamentally improving the efficiency of valve switching operations and reducing the fabrication cost of the chip. The experimental results demonstrate that the proposed algorithm can effectively construct a high-performance and low-cost control logic architecture. © 2025 Science Press. All rights reserved.

关键词： Deep reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

Extracting High-order Connectivity of EEG-based Dynamic Functional Connectivity Networks for Diagnosis of Major Depressive Disorder 24

Extracting High-order Connectivity of EEG-based Dynamic Func...

引用

4th Asia-Pacific Artificial Intelligence and big data Forum, AIBDF 2024

作者： Zhao, Feng Fan, Wenxuan Han, Zhongwei Chen, Hongyu School of Computer Science and Technology Shandong Technology and Business University Shandong Yantai China Yantai Key Laboratory of Big Data Modeling and Intelligent Computing Shandong Yantai China Immersion Technology and Evaluation Shandong Engineering Research Center Shandong Yantai China Information Engineering College Yantai Institute of Technology Shandong Yantai China Shandong Technology and Business University Shandong Yantai China

ISBN: (纸本)9798400710865

Functional connectivity networks (FCNs), particularly those utilizing the phase lag index (PLI) method, have been instrumental in elucidating the pathological features of Major Depressive Disorder (MDD) by assessing the synchrony and lag of electroencephalogram (EEG) signal waveforms. Currently, the more prevalent methods are traditional FCNs that quantify phase synchronization as scalar values and dynamic FCNs (D-FCNs) that employ sliding window approaches to capture the instantaneous variations in functional connectivity. However, these two methods merely capture the low-level functional connections between paired regions of interest (ROIs) from both the static and dynamic perspectives, and fail to reflect the high-level functional connections involving multiple ROIs. Moreover, the feature dimension of the dynamic FCN is excessively high, and there exists a phase mismatch issue between different samples, which renders it impossible to be directly applied to the classification task, and no better solution is available at present. To mitigate these issues, we proposed a comprehensive framework that includes segmenting EEG sequences based on PLI to construct dynamic FCNs;employing seventh-order central moments for feature extraction to form low-order FCNs (Lo-FCNs);and establishing high-order FCNs (Ho-FCNs) based on Pearson correlation theory derived from Lo-FCNs. We conducted classification experiments across four distinct EEG frequency bands using our proposed framework while analyzing relevant features through graph theory and brain topological mapping techniques. The experiments showed that the proposed framework has a high classification accuracy rate in the classification experiments of MDD and NC, and proved the importance of high-order features in classification experiments, indicating that they may serve as physiological markers for MDD diagnosis. © 2024 Copyright held by the owner/author(s).

关键词： Electroencephalography

来源：评论

学校读者我要写书评

暂无评论

SAV-SE: Scene-aware Audio-Visual Speech Enhancement with Selective State Space Model

引用

IEEE Journal on Selected Topics in Signal Processing 2025年

作者： Qian, Xinyuan Gao, Jiaran Zhang, Yaodan Zhang, Qiquan Liu, Hexin Garcia, Leibny Paola Li, Haizhou University of Science and Technology Beijing School of Computer and Communication Engineering Beijing100083 China University of New South Wales School of Electrical Engineering and Telecommunications Sydney2052 Australia Nanyang Technological University College of Computing and Data Science Singapore Singapore Johns Hopkins University CLSP and HLT-COE United States Chinese University of HongKong Guangdong Provincial Key Laboratory of Big Data Computing Shenzhen518172 China Shenzhen Research Institute of Big data Shenzhen51872 China

Speech enhancement plays an essential role in various applications, and the integration of visual information has been demonstrated to bring substantial advantages. However, the majority of current research concentrates on the examination of facial and lip movements, which can be compromised or entirely inaccessible in scenarios where occlusions occur or when the camera view is distant. Whereas contextual visual cues from the surrounding environment have been overlooked: for example, when we see a dog bark, our brain has the innate ability to discern and filter out the barking noise. To this end, in this paper, we introduce a novel task, i.e. Scene-aware Audio-Visual Speech Enhancement SAV-SE. To our best knowledge, this is the first proposal to use rich contextual information from synchronized video as auxiliary cues to indicate the type of noise, which eventually improves the speech enhancement performance. Specifically, we propose the VC-S $^{2}$ E method, which incorporates the Conformer and Mamba modules for their complementary strengths. Extensive experiments are conducted on public MUSIC, AVSpeech and AudioSet datasets, where the results demonstrate the superiority of VC-S $^{2}$ E over other competitive methods. We will make the source code publicly available. Project demo page: https://***/ © 2007-2012 IEEE.

关键词： Speech enhancement

来源：评论

学校读者我要写书评

暂无评论

Autonomous Multi-Objective Optimization Using Large Language Model

引用

IEEE Transactions on Evolutionary Computation 2025年

作者： Huang, Yuxiao Wu, Shenghao Zhang, Wenjie Wu, Jibin Feng, Liang Tan, Kay Chen Department of Data Science and Artificial Intelligence Hong Kong Polytechnic University Hong Kong Department of Electrical and Elctronics Engineering Hong Kong Polytechnic University Hong Kong Chongqing University College of Computer Science Chongqing Key Laboratory of Big Data Intelligence and Privacy Computing Chongqing400044 China

Multi-objective optimization problems (MOPs) are ubiquitous in real-world applications, presenting a complex challenge of balancing multiple conflicting objectives. Traditional multi-objective evolutionary algorithms (MOEAs), though effective, often rely on domain-specific expertise for improved optimization performance, hindering adaptability to unseen MOPs. In recent years, the Large Language Models (LLMs) has revolutionized software engineering by enabling the autonomous generation and refinement of programs. Leveraging this breakthrough, we propose a new LLM-based framework that autonomously designs MOEAs for solving MOPs. The proposed framework includes a robust testing module to refine the generated MOEA through error-driven dialogue with LLMs, a dynamic selection strategy along with informative prompting-based crossover and mutation to fit textual optimization pipeline. Our approach facilitates the design of MOEA without the extensive demands for expert intervention, thereby speeding up the innovation of MOEA. Empirical studies across various MOP categories validate the robustness and superior performance of our proposed framework. © 1997-2012 IEEE.

关键词： Multiobjective optimization

来源：评论

学校读者我要写书评

暂无评论

Fetal Cerebellum Landmark Detection Based on 3D MRI: Method and Benchmark

引用

IEEE Journal of Biomedical and Health Informatics 2025年

作者： Gong, Haifan Liu, Huixian Wang, Yitao Liu, Xiaoling Wan, Xiang Shi, Qiao Li, Haofeng Guangdong Provincial Key Laboratory of Big Data Computing Shenzhen Research Institute of Big Data Shenzhen518172 China The Chinese University of HongKong School of Science and Engineering Shenzhen518172 China Shenzhen Baoan Women's and Children's Hospital Department of Radiology Shenzhen518000 China The Chinese University of HongKong School of Data Science Shenzhen518172 China

Fetal cerebellum landmark detection is crucial for assessing fetal brain development. Although deep learning has become the standard for automatic landmark detection, most previous methods have focused on using 2D ultrasound or thick Magnetic Resonance Imaging (MRI). To improve accuracy, landmarks should be located on thin 3D MRIs. However, abnormal development, high noise, and fuzzy boundaries in 3D fetal brain images make traditional methods less effective for cerebellum landmark detection. To address this, we introduce the Anatomical Pseudo-label Guided Attention (APGA) network alongside a 3D MRI-based benchmark for fetal cerebellum landmark detection. During training, we use a shared encoder to extract image features and two decoders for landmark regression and anatomical pseudo-label segmentation. We design a Feature Decoupling Transformer (FDT) and embed it into the encoder to better calibrate the features for the two tasks. We only need the encoder, the FDT, and the landmark decoder during the inference phase. Extensive experiments on our proposed benchmark and out-of-domain test set have shown the effectiveness of our method. Our simulations also demonstrated that 3D biometrics are better than 2D biometrics. Code is available at https://***/lhaof/LFC. © 2013 IEEE.

关键词： Magnetic resonance imaging

来源：评论

学校读者我要写书评

暂无评论

Community-Aware Heterogeneous Graph Contrastive Learning 19th

Community-Aware Heterogeneous Graph Contrastive Learning

引用

19th CCF Conference on Computer Supported Cooperative Work and Social computing, ChineseCSCW 2024

作者： Li, Xinying Wu, Ling Guo, Kun College of Computer and Data Science Fuzhou University Fuzhou350108 China Engineering Research Center of Big Data Intelligence Ministry of Education Fuzhou350108 China Fujian Key Laboratory of Network Computing and Intelligent Information Processing Fuzhou University Fuzhou350108 China

ISBN: (纸本)9789819623723

Recently, heterogeneous graph contrastive learning, which can mine supervision signals from the data, has attracted widespread attention. However, most existing methods employ random data augmentation strategies to construct contrastive views, which may destroy the semantic information in heterogeneous graphs. Moreover, they often select positive and negative samples based solely on node-level proximity and overlook hard samples that are difficult to distinguish from anchors. To solve the above problems, we propose a Community-Aware Heterogeneous Graph Contrastive Learning model called CAHGCL. In particular, we design an adaptive data augmentation strategy to construct views, including feature augmentation and topology augmentation. To improve the quality of samples, we propose a dynamic sample weighting strategy based on node similarity and community information, capable of identifying both hard positive samples and hard negative samples. Finally, we introduce community-level contrast to improve community cohesion. Extensive experiments and analyses demonstrate that CAHGCL consistently outperforms state-of-the-art baselines on three datasets. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Community Evolution Tracking Based on High-Order Neighbor Consideration and Node Change Identification 19th

Community Evolution Tracking Based on High-Order Neighbor C...

引用

19th CCF Conference on Computer Supported Cooperative Work and Social computing, ChineseCSCW 2024

作者： Zhang, Yunan Wang, Chaohui Wu, Ling Guo, Kun College of Computer and Data Science Fuzhou University Fuzhou350108 China Engineering Research Center of Big Data Intelligence Ministry of Education Fuzhou350108 China Fujian Key Laboratory of Network Computing and Intelligent Information Processing Fuzhou University Fuzhou350108 China

ISBN: (纸本)9789819623723

Community evolution tracking is widely used in complex network analysis, which analyzes and identifies how communities evolve over time based on dynamic community detection. However, the current incremental dynamic community detection method has the phenomenon of ’concept drift’, which leads to inaccurate tracking of community evolution. Although some studies on dynamic community detection have proposed to perform global community detection on the network by timing or when the error information accumulates to a certain extent, this method will lead to a sudden change in community division, which will interrupt the evolution sequence of some communities and bring great difficulties to the continuous tracking of community evolution. In addition, the current community evolution tracking methods match the communities of two adjacent snapshots, and it is difficult to capture the fine-grained evolution process within the community. This paper proposes a community evolution tracking algorithm based on high-order neighbor consideration and node change identification. Firstly, based on an incremental update strategy that considers high-order neighbors, we adaptively expand the scope of incremental updates to reduce the accumulation of error information. Secondly, through the node change identification strategy to mine fine-grained community evolution events. Experiments show that our method can effectively track the evolution of communities in dynamic networks. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Community evolution Complex network Dynamic network Evolution tracking

来源：评论

学校读者我要写书评

暂无评论

D-FGNAE: Decentralized Federated Graph Normalized AutoEncoder 19th

D-FGNAE: Decentralized Federated Graph Normalized AutoEncode...

引用

19th CCF Conference on Computer Supported Cooperative Work and Social computing, ChineseCSCW 2024

作者： Liang, Yuting Cai, Weixin Guo, Kun College of Computer and Data Science Fuzhou University Fuzhou350108 China Engineering Research Center of Big Data Intelligence Ministry of Education Fuzhou350108 China Fujian Key Laboratory of Network Computing and Intelligent Information Processing Fuzhou University Fuzhou350108 China

ISBN: (纸本)9789819623723

Graphs widely exist in real-world, and Graph Neural Networks (GNNs) have exhibited exceptional efficacy in graph learning in diverse fields. With the strengthening of data privacy protection worldwide in recent years, Federated graph neural networks (FedGNNs) have gained increasing attention in academia and industry owing to their ability to train the model in a collaborative manner while complying with the privacy protection regulations. However, in federated learning, the non-independent and identically distributed (non-IID) problem of local data possessed by multiple participants can significantly undermine model accuracy. We propose a new Decentralized Federated Graph Normalized AutoEncoder (D-FGNAE). First, the model is designed as a decentralized federated learning framework with dynamically assigned tripartite roles. This design eliminates the fixed server role found in traditional federated learning, enhances system fault tolerance, avoids single points of failure, and protects model privacy. Second, the splitting and correcting of calculation by layer in the model, along with the special design of the normalization layer, effectively tackle the non-IID problem in both the structural and attribute aspects. Experimental results on real-world networks demonstrate the effectiveness of D-FGNAE, which can achieve nearly the same accuracy as the centralized model. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Federated learning

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：