检索结果-内蒙古大学图书馆

Diffusion-Based Radiotherapy Dose Prediction Guided by Inter-Slice Aware Structure Encoding

IEEE TRANSACTIONS ON EMERGING TOPICS IN computational INTELLIGENCE 2025年第2期9卷 1119-1129页

作者： Feng, Zhenghao Wen, Lu Xiao, Jianghong Xu, Yuanyuan Wu, Xi Zhou, Jiliu Peng, Xingchen Wang, Yan Sichuan Univ Sch Comp Sci Chengdu 610017 Peoples R China Sichuan Univ West China Hosp Canc Ctr Dept Radiat Oncol Chengdu 610017 Peoples R China Chengdu Univ Informat Technol Sch Comp Sci Chengdu 610103 Peoples R China Sichuan Univ West China Hosp Canc Ctr Dept Biotherapy Chengdu 610017 Peoples R China

Deep learning (DL) has successfully automated dose distribution prediction for radiotherapy planning, increasing both efficiency and quality. However, existing methods commonly utilize ${{\mathbf{L}}_1}$ or ${{\mathbf{L}}_2}$ loss to calculate the posterior average, thus heavily suffering from the over-smoothing problem. To address this, we propose a diffusion model-based method, named DiffDose, to automatically predict radiotherapy dose distribution for cancer patients. Specifically, our DiffDose model contains a forward process and a reverse process. In the forward process, DiffDose gradually adds small noise to dose distribution maps via multiple steps until converting them to pure Gaussian noise, and a noise predictor is simultaneously trained to estimate the noise added in each step. In the reverse process, DiffDose iteratively removes the noise from a pure Gaussian noise leveraging the well-trained noise predictor and finally outputs the predicted dose distribution maps. Concretely, to provide the model with essential structure information, we design a structure encoder to learn the anatomical knowledge from patients' anatomy images, guiding the noise predictor to generate dose distribution maps that are aware of personalized structures. Considering the latent continuity and similarity among sliced anatomy images, an inter-slice interaction transformer (I2T) block is embedded in the structure encoder to capture such long-range dependency. Extensive experiments on an in-house dataset involving 130 rectum cancer cases validate the superiority of our method.

关键词： Predictive models Diffusion models Radiation therapy Gaussian noise computational modeling Accuracy Computed tomography Radiotherapy treatment dose prediction diffusion model transformer

来源：评论

学校读者我要写书评

暂无评论

Network Monitoring Data Recovery Based on Flexible Bi-Directional Model

引用

IEEE TRANSACTIONS ON NETWORK SCIENCE AND ENGINEERING 2025年第2期12卷 623-635页

作者： Lin, Qixue Li, Xiaocan Xie, Kun Wen, Jigang He, Shiming Xie, Gaogang Fan, Xiaopeng Feng, Quan Hunan Univ Coll Comp Sci & Elect Engn Changsha 410082 Peoples R China Hunan Univ Key Lab Supercomp & Artiffcial Intelligence Fus Co Minist Educ Changsha 410082 Peoples R China Hunan Univ Sci & Technol Sch Comp Sci & Engn Xiangtan 411201 Peoples R China Changsha Univ Sci & Technol Sch Comp & Commun Engn Changsha 410004 Peoples R China Chinese Acad Sci Comp Network Informat Ctr Beijing 100190 Peoples R China Univ Chinese Acad Sci Sch Comp Sci & Technol Beijing 100190 Peoples R China Hangzhou Inst Adv Technol Hangzhou 310024 Peoples R China Hunan Vanguard Grp Corp Ltd Changsha 410100 Peoples R China

Comprehensive network monitoring data is crucial for anomaly detection and network optimization tasks. However, due to factors such as sampling strategies and failures in data transmission or storage, only incomplete monitoring data can be obtained. Traditional techniques for completing network monitoring data matrices have limitations in leveraging network-related features and lack the adaptability required for offline and online execution. In this paper, we introduce a novel approach that significantly improves the integration of network features and operational flexibility in data completion tasks. By converting the data matrix into a bipartite graph and integrating network features into the graph's node attributes, we redefine the problem of missing data completion. This transformation reframes the issue as estimating unobserved edges in the bipartite graph. We propose the Bi-directional Bipartite Graph Completion (BGC) model, a flexible framework that seamlessly adapts to both offline and online data completion tasks. This model encapsulates static, dynamic, bi-directional temporal features and network topology, thereby improving the accuracy of unobserved edge estimation. Experiments conducted on two public data traces demonstrate the superiority of our method over six baseline models. Our method not only achieves higher accuracy in offline scenarios but also displays remarkable speed in online situations.

关键词： Monitoring Tensors Data models Neural networks Feature extraction Bipartite graph Adaptation models Matrix decomposition computational modeling Accuracy Network monitoring data completion bipartite graph

来源：评论

学校读者我要写书评

暂无评论

Defending Data Inference Attacks Against Machine Learning Models by Mitigating Prediction Distinguishability

引用

IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING 2025年第3期22卷 2687-2704页

作者： Yang, Ziqi Zhu, Yiran Wan, Jie Xiang, ChuXiao Tang, Tong Wang, Yilin Xu, Ruite Wang, Lijin Zhang, Fan Xu, Jiarong Qin, Zhan Zhejiang Univ State Key Lab Blockchain & Data Secur Hangzhou 310027 Peoples R China Hangzhou High Tech Zone Binjiang Inst Blockchain & Hangzhou 310027 Peoples R China Zhejiang Univ Coll Comp Sci & Technol Hangzhou 310027 Peoples R China Fudan Univ Dept Informat Management & Business Intelligence Shanghai 200437 Peoples R China

Neural networks are vulnerable to data inference attacks, including the membership inference attack, the model inversion attack, and the attribute inference attack. In this paper, we propose Purifier to defend against membership inference attacks by quantifying the differences between dataset members and non-members in three dimensions: individual shape, statistical distribution, and prediction label. Purifier involves transforming the confidence scores produced by the target classifier, resulting in purified confidence scores that are indistinguishable across the dimensions above. We conduct experiments on widely-used datasets and models. The results show that Purifier offers robust defense against membership inference attacks with superior efficacy compared to prior defense techniques while maintaining minimal utility degradation (e.g., less than 0.7% classification accuracy drop of most datasets). Additionally, our extended experiments explore the effectiveness of Purifier in defending against the model inversion attack and the attribute inference attack.

关键词： Data models Predictive models Shape Training Statistical distributions Vectors Multiprotocol label switching Machine learning computational modeling Accuracy Data privacy machine learning security membership inference

来源：评论

学校读者我要写书评

暂无评论

SOC Estimation of Lithium-Ion Battery Pack Based on Discharge Stage Division and Fusion modeling

引用

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT 2025年 74卷

作者： Chen, Yuan Liu, Yanzhong He, Yigang Lyu, Zhiqiang Cai, Yujing Zhang, Siyuan Anhui Univ Sch Artificial Intelligence Hefei 230000 Peoples R China Wuhan Univ Sch Elect & Automat Wuhan 430000 Peoples R China Anhui Univ Sch Internet Hefei 230000 Peoples R China

To meet practical usage requirements, lithium-ion batteries usually need to form a battery pack. However, due to production deviations and different usage environments, there are inconsistencies between batteries within the battery pack. This makes it challenging to estimate the state of charge (SOC) of the battery pack accurately. This article proposes a battery pack SOC estimation approach based on discharge stage division and fusion modeling. According to the battery discharge characteristics and SOC inconsistency, three stages are divided in the battery pack discharge process. In the first stage, the second-order RC model and extended Kalman filter (EKF) algorithm are employed for SOC estimation as the consistency between batteries is good. In the second stage, the representative battery and long short-term memory (LSTM) recurrent neural network (RNN) will be used to consider the impact of battery inconsistency on battery SOC estimation. An ampere-hour integration with a constraint factor for smoothing is applied to enhance the estimation accuracy of the LSTM network. In the third stage, EKF is utilized to estimate the SOC of all batteries as the inconsistency between batteries increases significantly and reaches a maximum at the end of the discharge. Finally, experiments and actual vehicle tests under the Worldwide Harmonized Light Vehicles Test Cycle (WLTC) working conditions are conducted to verify the effectiveness of proposed methods on the ternary lithium battery pack at 0 C-degrees, 25 C-degrees, and 45 C-degrees. Compared with other algorithms, the proposed method has good estimation performance, with a maximum RMSE of only 0.8%. The maximum prediction error under actual vehicle operating conditions is less than 1.2%.

关键词： Batteries State of charge Discharges (electric) Estimation Integrated circuit modeling Accuracy computational modeling Long short term memory Mathematical models Prediction algorithms Ampere-hour integration with constraint factor battery pack discharge stage division fusion model state of charge (SOC)

来源：评论

学校读者我要写书评

暂无评论

Contrastive Neuron Pruning for Backdoor Defense

引用

IEEE TRANSACTIONS ON IMAGE PROCESSING 2025年 34卷 1234-1245页

作者： Feng, Yu Ma, Benteng Liu, Dongnan Zhang, Yanning Cai, Weidong Xia, Yong NorthwesternPolytechn Univ Sch Comp Sci & Engn Natl Engn Lab Integrated Aerosp Ground Ocean Big D Xian 710072 Peoples R China Univ Sydney Sch Comp Sci Sydney NSW 2006 Australia Northwestern Polytech Univ Shenzhen Res & Dev Inst Shenzhen 518057 Peoples R China

Recent studies have revealed that deep neural networks (DNNs) are susceptible to backdoor attacks, in which attackers insert a pre-defined backdoor into a DNN model by poisoning a few training samples. A small subset of neurons in DNN is responsible for activating this backdoor and pruning these backdoor-associated neurons has been shown to mitigate the impact of such attacks. Current neuron pruning techniques often face challenges in accurately identifying these critical neurons, and they typically depend on the availability of labeled clean data, which is not always feasible. To address these challenges, we propose a novel defense strategy called Contrastive Neuron Pruning (CNP). This approach is based on the observation that poisoned samples tend to cluster together and are distinguishable from benign samples in the feature space of a backdoored model. Given a backdoored model, we initially apply a reversed trigger to benign samples, generating multiple positive (benign-benign) and negative (benign-poisoned) feature pairs from the backdoored model. We then employ contrastive learning on these pairs to improve the separation between benign and poisoned features. Subsequently, we identify and prune neurons in the Batch Normalization layers that show significant response differences to the generated pairs. By removing these backdoor-associated neurons, CNP effectively defends against backdoor attacks while requiring the pruning of only about 1% of the total neurons. Comprehensive experiments conducted on various benchmarks validate the efficacy of CNP, demonstrating its robustness and effectiveness in mitigating backdoor attacks compared to existing methods.

关键词： Neurons Training Data models Feature extraction Robustness Perturbation methods Face recognition Electronic mail computational modeling Batch normalization Backdoor defense image classification contrastive learning

来源：评论

学校读者我要写书评

暂无评论

A Sparse Cross Attention-Based Graph Convolution Network With Auxiliary Information Awareness for Traffic Flow Prediction

引用

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS 2025年第3期26卷 3210-3222页

作者： Chen, Lingqiang Zhao, Qinglin Li, Guanghui Zhou, Mengchu Dai, Chenglong Feng, Yiming Liu, Xiaowei Li, Jinjiang Hebei Univ Engn Sch Informat & Elect Engn Handan 056038 Peoples R China Macau Univ Sci & Technol Sch Comp Sci & Engn Macau Peoples R China Jiangnan Univ Sch Artificial Intelligence & Comp Sci Wuxi 214122 Peoples R China New Jersey Inst Technol Dept Elect & Comp Engn Newark NJ 07102 USA

Deep graph convolutional networks (GCNs) have shown promising performance in traffic prediction tasks, but their practical deployment on resource-constrained devices faces challenges. First, few models consider the potential influence of historical and future auxiliary information, such as weather and holidays, on complex traffic patterns. Second, the computational complexity of dynamic graph convolution operations grows quadratically with the number of traffic nodes, limiting model scalability. To address these challenges, this study proposes a deep encoder-decoder model named AIMSAN, which comprises an auxiliary information-aware module (AIM) and a sparse cross-attention-based graph convolutional network (SAN). From historical or future perspectives, AIM prunes multi-attribute auxiliary data into diverse time frames, and embeds them into one tensor. SAN employs a cross-attention mechanism to merge traffic data with historical embedded data in each encoder layer, forming dynamic adjacency matrices. Subsequently, it applies diffusion GCN to capture rich spatial-temporal dynamics from the traffic data. Additionally, AIMSAN utilizes the spatial sparsity of traffic nodes as a mask to mitigate the quadratic computational complexity of SAN, thereby improving overall computational efficiency. In the decoder layer, future embedded data are fused with feed-forward traffic data to generate prediction results. Experimental evaluations on three public traffic datasets demonstrate that AIMSAN achieves competitive performance compared to state-of-the-art algorithms, while reducing GPU memory consumption by 41.24%, training time by 62.09%, and validation time by 65.17% on average.

关键词： computational modeling Data models Meteorology Convolution computational complexity Vehicle dynamics Traffic control Sparse matrices Predictive models Graph convolutional networks Traffic flow prediction auxiliary information sparse cross attention graph convolutional network

来源：评论

学校读者我要写书评

暂无评论

Hier-FUN: Hierarchical Federated Learning and Unlearning in Heterogeneous Edge Computing

引用

IEEE INTERNET OF THINGS JOURNAL 2025年第7期12卷 8653-8668页

作者： Ma, Zhenguo Tu, Huaqing Zhou, Li Ji, Pengli Yan, Xiaoran Xu, Hongli Wang, Zhiyuan Chen, Suo Zhejiang Lab Res Ctr Data Hub & Secur Hangzhou 311121 Zhejiang Peoples R China Zhejiang Lab Res Ctr Innovat Intelligent Comp Facil Hangzhou 311121 Zhejiang Peoples R China Univ Sci & Technol China Sch Comp Sci & Technol Hefei 230027 Anhui Peoples R China Univ Sci & Technol China Suzhou Inst Adv Res Suzhou 215123 Jiangsu Peoples R China

Federated learning (FL) has emerged as a pivotal paradigm for distributed model training in edge computing (EC), enabling cooperation among numerous Internet of Things devices while safeguarding their data privacy. Despite its successes in machine learning, concerns regarding data security and model fidelity necessitate the efficient unlearning of target device, i.e., federated unlearning (FUN). However, due to resource constraints, device heterogeneity, and non-independent and identically distributed (Non-IID) data, securely eliminating a device's impact without retraining the model from scratch presents a complex challenge. In response to these challenges, we propose a hierarchical FUN framework, called Hier-FUN. Hier-FUN organizes edge devices into K clusters, each managed by a head device responsible for aggregating local models within the cluster. To expedite both the learning and unlearning processes of Hier-FUN, we design a heuristic algorithm to determine an appropriate value for K based on devices' data distributions and available resources. In addition, Hier-FUN denies the communication between the server and cluster heads during training, which can constrain the influence sphere of target device and accelerate the unlearning process. We conduct extensive experiments using real-world datasets, and the experimental results illustrate that Hier-FUN can improve test accuracy by 3.19% during the learning phase and achieve a 6.8x speedup during unlearning compared with the baseline methods.

关键词： Data models Performance evaluation Servers computational modeling Training Internet of Things Federated learning Edge computing Magnetic heads Heuristic algorithms Device heterogeneity edge computing (EC) federated unlearning (FUN) hierarchical architectures non-independent and identically distributed (Non-IID)

来源：评论

学校读者我要写书评

暂无评论

modeling, Embedded Control, and Design of Soft Robots Using a Learned Condensed FEM Model

引用

IEEE TRANSACTIONS ON ROBOTICS 2025年 41卷 2441-2459页

作者： Navez, Tanguy Menager, Etienne Chaillou, Paul Goury, Olivier Kruszewski, Alexandre Duriez, Christian Univ Lille CNRS Cent Lille UMR CRIStAL 9189Inria F-59000 Lille France PSL Res Univ Ecole Normale Super Dept informat ENS InriaCNRS F-75013 Paris France

The finite element method (FEM) is a powerful modeling tool for predicting soft robots' behavior, but its computation time can limit practical applications. In this article, a learning-based approach based on condensation of the FEM model is detailed. The proposed method handles several kinds of actuators and contacts with the environment. We demonstrate that this compact model can be learned as a unified model across several designs and remains very efficient in terms of modeling since we can deduce the direct and inverse kinematics of the robot. Building upon the intuition introduced in (M & eacute;nager et al., 2023), the learned model is presented as a general framework for modeling, controlling, and designing soft manipulators. First, the method's adaptability and versatility are illustrated through optimization-based control problems involving positioning and manipulation tasks with mechanical contact-based coupling. Second, the low-memory consumption and the high prediction speed of the learned condensed model are leveraged for real-time embedding control without relying on costly online FEM simulation. Finally, the ability of the learned condensed FEM model to capture soft robot design variations and its differentiability are leveraged in calibration and design optimization applications.

关键词： Soft robotics Robots computational modeling Finite element analysis Mathematical models Biological system modeling Numerical models Adaptation models Predictive models Geometry Control and design of soft robots finite element method (FEM) simulation model reduced modeling

来源：评论

学校读者我要写书评

暂无评论

Hybrid Channel Tracking for THz Massive MIMO Communication Systems in Dynamic Environments

引用

IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS 2025年第3期43卷 988-1003页

作者： Fan, Yuheng Yang, Chuang Sun, Yanran Peng, Mugen Beijing Univ Posts & Telecommun State Key Lab Networking & Switching Technol Beijing 100876 Peoples R China

With gigahertz-level bandwidth, terahertz (THz) holds promise for achieving exceptionally high transmission rates in prospective sixth-generation (6G) communications. However, considerable path loss poses an obstacle to THz communications. To compensate for this, massive multiple-input-multiple-output (MIMO) based beamforming is utilized to promote directional power with narrow beams in communications. In dynamic environments, the frequent adjustment of narrow beams results in fast time-varying channel state information (CSI), which constrains the application of the THz communication systems. While traditional deterministic-based and statistical-based channel tracking methods address different aspects of this issue, they suffer from balancing accuracy and complexity in the THz dynamic environments. To solve this problem, based on the cluster distribution of THz time-varying channel, we propose a novel hybrid channel tracking method that uses deterministic physical motion variation law to extract the cluster subspace, and then statistical Markov evolution models are applied within it. To achieve this, an integrated clustering and estimation method, clustering subspace matching pursuit (CSMP) is proposed for obtaining the channel clusters prior knowledge. Then based on above hybrid tracking method design, we propose a virtual cluster subspace turbo-approximate message passing (VCS-TAMP). Finally, several simulation results validate that our proposal achieves great improvement in both accuracy and computational time performance.

关键词： Terahertz communications Vectors Massive MIMO Accuracy Heuristic algorithms Estimation Tracking Solid modeling Mathematical models computational modeling Terahertz massive MIMO channel tracking

来源：评论

学校读者我要写书评

暂无评论

A Reduced-Parameter Mathematical Surrogate Model for Improving Measurement Accuracy of Mechanical Systems Using Experimental Techniques

引用

IEEE ACCESS 2025年 13卷 58281-58301页

作者： Vijay Deosant, Pranav Jawale, H. P. Thorat, H. T. Jaiswal, Ankur Visvesvaraya Natl Inst Technol VNIT Dept Mech Engn Nagpur 440010 Maharashtra India Manipal Acad Higher Educ MAHE Manipal Inst Technol Dept Mechatron Manipal 576104 Karnataka India

modeling of a larger mechanical system is inherently difficult due to the complexity of individual processes and the interaction between multiple processes. Using a polynomial model based on experimental results to obtain an input-output relationship offers an easier option than establishing a mathematical model in the case of an Open Loop Controller. Such input-output relationship is workable but sometimes experiences computational burden due to a higher degree of the polynomial. Surrogate modeling is commonly used as a method to overcome this difficulty through an approximation by reducing the degree of complexity of the relationship. Surrogate models based on statistical principles, which correlate multiple inputs and required outputs, are opaque to the system parameters. These system parameters can be optimized to improve the measurement performance and hence input-output relationship by either the polynomial model or its surrogate does not serve the purpose of optimization. A surrogate of a mathematical model based on experimentation is proposed here. The method involves developing mathematical models of subsystems using minimum parameters. The complexity of each subsystem is reduced by retaining dominant parameters and eliminating non-influential parameters. To improve the measurement accuracy of the optimization, model an experimental method has evolved where parameters are attuned appropriately to accommodate the effect of eliminated parameters. This technique is demonstrated by establishing a mathematical model for a pendulum-ball system.

关键词： Mathematical models Polynomials computational modeling Accuracy Mechanical systems Analytical models Complexity theory Motors Torque Optimization Mathematical modeling system parameters measurement surrogates input-output relationship simple pendulum

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：