检索结果-内蒙古大学图书馆

A medical visual question-answering model based on multi-scale feature fusion and question Feature enhancement

Multimedia Systems 2025年第4期31卷

作者： Xia, Hong Zhang, Yifan Jia, Hui Chen, Yanping Xu, Jing Li, Shiyong School of Computer Science and Technology Xi’an University of Posts and Telecommunications Shaanxi Xi’an710121 China Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing Shaanxi Xi’an710121 China Xi’an Key Laboratory of Big Data and Intelligent Computing Shaanxi Xi’an710121 China

The core task of Medical Visual Question Answering (Med-VQA) is to generate accurate answers to clinical questions based on medical images. However, existing models often struggle to simultaneously extract fine-grained details and global semantics due to their reliance on single-scale feature processing. In addition, most methods model the interaction between questions and visual features using bilinear attention, failing to fully leverage visual information to enhance text representations, which leads to suboptimal cross-modal fusion. To address these issues, we propose a novel multi-scale and cross-modal enhancement framework for Med-VQA. Our approach introduces two key innovations: (1) A multi-scale visual encoder that combines vision transformer (ViT) and convolutional neural network (CNN) features through a global–local fusion mechanism, enabling the model to simultaneously capture fine-grained details and global semantics, overcoming the limitations of single-scale processing. (2) A question enhancement module that dynamically refines text representations using visual information before multimodal fusion, enabling more effective cross-modal interactions. By jointly optimizing multi-scale visual feature fusion and text enhancement, our approach more comprehensively integrates image details and global information, effectively capturing cross-modal interactions. This enables the model to better understand the complex relationships between images and text, thereby improving reasoning accuracy. Extensive experiments on the VQA-RAD and SLAKE benchmark datasets show that our model achieves an accuracy of 77.2% on the VQA-RAD dataset and 83.1% on the SLAKE dataset. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2025.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Interventional Multi-Instance Learning with Deconfounded Instance-Level Prediction

arXiv

引用

arXiv 2022年

作者： Lin, Tiancheng Xu, Hongteng Yang, Canqian Xu, Yi Shanghai Key Lab of Digital Media Processing and Transmission Shanghai Jiao Tong University China MoE Key Lab of Artificial Intelligence AI Institute Shanghai Jiao Tong University China Gaoling School of Artificial Intelligence Renmin University of China China Beijing Key Laboratory of Big Data Management and Analysis Methods China JD Explore Academy China

When applying multi-instance learning (MIL) to make predictions for bags of instances, the prediction accuracy of an instance often depends on not only the instance itself but also its context in the corresponding bag. From the viewpoint of causal inference, such bag contextual prior works as a confounder and may result in model robustness and interpretability issues. Focusing on this problem, we propose a novel interventional multi-instance learning (IMIL) framework to achieve deconfounded instance-level prediction. Unlike traditional likelihood-based strategies, we design an Expectation-Maximization (EM) algorithm based on causal intervention, providing a robust instance selection in the training phase and suppressing the bias caused by the bag contextual prior. Experiments on pathological image analysis demonstrate that our IMIL method substantially reduces false positives and outperforms state-of-the-art MIL methods. Copyright © 2022, The Authors. All rights reserved.

关键词： Forecasting

来源：评论

学校读者我要写书评

暂无评论

Efficient meta neural heuristic for multi-objective combinatorial optimization 23

Efficient meta neural heuristic for multi-objective combinat...

引用

Proceedings of the 37th International Conference on Neural Information processing Systems

作者： Jinbiao Chen Jiahai Wang Zizhen Zhang Zhiguang Cao Te Ye Siyuan Chen School of Computer Science and Engineering Sun Yat-sen University P.R. China School of Computer Science and Engineering Sun Yat-sen University P.R. China and Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education Sun Yat-sen University P.R. China and Guangdong Key Laboratory of Big Data Analysis and Processing Guangzhou P.R. China School of Computing and Information Systems Singapore Management University Singapore

Recently, neural heuristics based on deep reinforcement learning have exhibited promise in solving multi-objective combinatorial optimization problems (MOCOPs). However, they are still struggling to achieve high learning efficiency and solution quality. To tackle this issue, we propose an efficient meta neural heuristic (EMNH), in which a meta-model is first trained and then fine-tuned with a few steps to solve corresponding single-objective subproblems. Specifically, for the training process, a (partial) architecture-shared multi-task model is leveraged to achieve parallel learning for the meta-model, so as to speed up the training; meanwhile, a scaled symmetric sampling method with respect to the weight vectors is designed to stabilize the training. For the fine-tuning process, an efficient hierarchical method is proposed to systematically tackle all the subproblems. Experimental results on the multi-objective traveling salesman problem (MOTSP), multi-objective capacitated vehicle routing problem (MOCVRP), and multi-objective knapsack problem (MOKP) show that, EMNH is able to outperform the state-of-the-art neural heuristics in terms of solution quality and learning efficiency, and yield competitive solutions to the strong traditional heuristics while consuming much shorter time.

关键词：

来源：评论

学校读者我要写书评

暂无评论

An Improved 2P-CFAR Method for Ship Detection in SAR images

An Improved 2P-CFAR Method for Ship Detection in SAR Images

引用

IEEE International Conference on Radar

作者： Jiahui Chang Qing Wang Jianhui Zhao Ning Li Henan Key Laboratory of Big Data Analysis and Processing (of Henan University) College of Computer and Information Engineering (of Henan University) Kaifeng China Air Force Research Institute Beijing China

ISBN: (纸本)9781665468893

Synthetic Aperture Radar (SAR) systems can be used for discovery and surveillance. According to the current requirements, accurate ship detection is becoming more and more important. However, the interference factors such as coherent speckle noise and strong target sidelobes will reduce the detection accuracy. In this paper, an improved two-parameter Constant False Alarm Rate (2P-CFAR) method is proposed to solve this problem. After using the non-local means (NLM) filtering method to suppress coherent speckle noise, the 2P-CFAR method is performed to obtain candidate targets, and an equal proportion segmentation procedure is carried out to optimize the aspect ratios of the candidate targets and weaken the sidelobes. Experimental results based on the SAR images acquired by the Gaofen-3 satellite verify the performance of the proposed method.

关键词： image segmentation Satellites Filtering Spaceborne radar Surveillance Radar detection Speckle

来源：评论

学校读者我要写书评

暂无评论

Efficient Document Retrieval by End-to-End Refining and Quantizing BERT Embedding with Contrastive Product Quantization

arXiv

引用

arXiv 2022年

作者： Qiu, Zexuan Su, Qinliang Yu, Jianxing Si, Shijing School of Computer Science and Engineering Sun Yat-sen University Guangzhou China Guangdong Key Laboratory of Big Data Analysis and Processing Guangzhou China School of Artificial Intelligence Sun Yat-sen University Guangzhou China School of Economics and Finance Shanghai International Studies University China

Efficient document retrieval heavily relies on the technique of semantic hashing, which learns a binary code for every document and employs Hamming distance to evaluate document distances. However, existing semantic hashing methods are mostly established on outdated TFIDF features, which obviously do not contain lots of important semantic information about documents. Furthermore, the Hamming distance can only be equal to one of several integer values, significantly limiting its representational ability for document distances. To address these issues, in this paper, we propose to leverage BERT embeddings to perform efficient retrieval based on the product quantization technique, which will assign for every document a real-valued codeword from the codebook, instead of a binary code as in semantic hashing. Specifically, we first transform the original BERT embeddings via a learnable mapping and feed the transformed embedding into a probabilistic product quantization module to output the assigned codeword. The refining and quantizing modules can be optimized in an end-to-end manner by minimizing the probabilistic contrastive loss. A mutual information maximization based method is further proposed to improve the representativeness of codewords, so that documents can be quantized more accurately. Extensive experiments conducted on three benchmarks demonstrate that our proposed method significantly outperforms current state-of-the-art baselines. Copyright © 2022, The Authors. All rights reserved.

关键词： Hamming distance

来源：评论

学校读者我要写书评

暂无评论

Dynamic Pricing for Client Recruitment in Federated Learning

arXiv

引用

arXiv 2022年

作者： Wang, Xuehe Zheng, Shensheng Duan, Lingjie School of Artificial Intelligence Sun Yat-sen University Zhuhai The Guangdong Key Laboratory of Big Data Analysis and Processing Guangzhou China School of Artificial Intelligence Sun Yat-sen University Zhuhai China Pillar of Engineering Systems and Design Singapore University of Technology and Design Singapore Singapore

A federated learning (FL) system first needs to use heterogeneous clients to recruit data, and then perform data training locally on the client to finish model training for preserving users’ data privacy. Yet clients still face the cost of computational energy consumption on model training and parameter transmission and are reluctant to join federated learning platform. Thus, incentive mechanisms are needed when inviting numerous clients to help train the global model. However, static pricing cannot adjust data size adaptively to minimize cost due to client random arrival for participation and incomplete information about computing cost. Moreover, how to balance data sampling time and model training for finite time horizons is under-explored. Therefore, in this paper, a two-phase FL model consisting of the data recruitment phase and model training phase is formulated to study the clients’ incentive mechanism design under incomplete information about clients’ arrivals and their private costs. First, we propose a dynamic pricing scheme for homogeneous clients that offer time-dependent monetary returns to clients by considering the tradeoff between the total payment to clients and the model accuracy loss. We show that the pricing should increase over time in the data-sampling phase due to data aging. Then we obtain the best time partition between data sampling and training phases under a closed-form dynamic pricing solution. It is shown that a small data recruitment threshold should be allocated if the client’ training time per global iteration is long. Further, the extension to heterogeneous clients with different data size and working time is discussed regarding to the optimal dynamic pricing and recruitment threshold. We prove monotonicity in multiple client-type choice and a linear algorithm is proposed to find the optimal recruitment threshold and client-type choice for total cost minimization. Finally, we present the robustness of our algorithm to data size and e

关键词： Costs

来源：评论

学校读者我要写书评

暂无评论

Opara: Exploiting Operator Parallelism for Expediting DNN Inference on GPUs

arXiv

引用

arXiv 2023年

作者： Chen, Aodong Xu, Fei Han, Li Dong, Yuan Chen, Li Zhou, Zhi Liu, Fangming The Shanghai Key Laboratory of Multidimensional Information Processing School of Computer Science and Technology East China Normal University 3663 N. Zhongshan Road Shanghai200062 China The School of Software Engineering East China Normal University 3663 N. Zhongshan Road Shanghai200062 China The School of Computing and Informatics University of Louisiana at Lafayette 301 East Lewis Street LafayetteLA70504 United States The Guangdong Key Laboratory of Big Data Analysis and Processing School of Computer Science and Engineering Sun Yat-sen University 132 E. Waihuan Road Guangzhou510006 China Peng Cheng Laboratory Huazhong University of Science and Technology China

GPUs have become the defacto hardware devices for accelerating Deep Neural Network (DNN) inference workloads. However, the conventional sequential execution mode of DNN operators in mainstream deep learning frameworks cannot fully utilize GPU resources, even with the operator fusion enabled, due to the increasing complexity of model structures and a greater diversity of operators. Moreover, the inadequate operator launch order in parallelized execution scenarios can lead to GPU resource wastage and unexpected performance interference among operators. In this paper, we propose Opara, a resource- and interference-aware DNN Operator parallel scheduling framework to accelerate DNN inference on GPUs. Specifically, Opara first employs CUDA Streams and CUDA Graph to parallelize the execution of multiple operators automatically. To further expedite DNN inference, Opara leverages the resource demands of operators to judiciously adjust the operator launch order on GPUs, overlapping the execution of compute-intensive and memory-intensive operators. We implement and open source a prototype of Opara based on PyTorch in a non-intrusive manner. Extensive prototype experiments with representative DNN and Transformer-based models demonstrate that Opara outperforms the default sequential CUDA Graph in PyTorch and the state-of-the-art operator parallelism systems by up to 1.68× and 1.29×, respectively, yet with acceptable runtime overhead. Copyright © 2023, The Authors. All rights reserved.

关键词： Deep neural networks

来源：评论

学校读者我要写书评

暂无评论

Diffusion Model-based Contrastive Learning for Human Activity Recognition

arXiv

引用

arXiv 2024年

作者： Xiao, Chunjing Han, Yanhui Yang, Wei Hou, Yane Shi, Fangzhan Chetty, Kevin The School of Computer and Information Engineering Henan University Kaifeng475004 China University College London LondonWC1E 6BT United Kingdom The School of Computer and Information Engineering Henan University China The Henan Key Laboratory of Big Data Analysis and Processing Henan University Kaifeng475004 China The Department of Security and Crime Science University College London LondonWC1E 6BT United Kingdom

WiFi Channel State Information (CSI)-based activity recognition has sparked numerous studies due to its widespread availability and privacy protection. However, when applied in practical applications, general CSI-based recognition models may face challenges related to the limited generalization capability, since individuals with different behavior habits will cause various fluctuations in CSI data and it is difficult to gather enough training data to cover all kinds of motion habits. To tackle this problem, we design a diffusion model-based Contrastive Learning framework for human Activity Recognition (CLAR) using WiFi CSI. On the basis of the contrastive learning framework, we primarily introduce two components for CLAR to enhance CSI-based activity recognition. To generate diverse augmented data and complement limited training data, we propose a diffusion model-based time series-specific augmentation model. In contrast to typical diffusion models that directly apply conditions to the generative process, potentially resulting in distorted CSI data, our tailored model dissects these condition into the high-frequency and low-frequency components, and then applies these conditions to the generative process with varying weights. This can alleviate data distortion and yield high-quality augmented data. To efficiently capture the difference of the sample importance, we present an adaptive weight algorithm. Different from typical contrastive learning methods which equally consider all the training samples, this algorithm adaptively adjusts the weights of positive sample pairs for learning better data representations. The experiments suggest that CLAR achieves significant gains compared to state-of-the-art methods. Copyright © 2024, The Authors. All rights reserved.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Reversible adversarial information hiding based on the interpretation of neural network

引用

Digital Signal processing: A Review Journal 2025年 165卷

作者： Li, Ruiping Wang, Xi Cui, Qingchen Li, Ming Information Construction and Management Office Henan Normal University Xinxiang453007 China School of Artificial Intelligence and Software Engineering Nanyang Normal University Nanyang473061 China Henan Engineering Research Center of Intelligent Processing for Big Data of Digital Image Henan Nanyang473000 China College of Information Engineering Zhoukou Vocational College of Arts and Science Zhoukou466000 China College of Computer and Information Engineering Henan Normal University Xinxiang453007 China Key Laboratory of Artificial Intelligence and Personalized Learning in Education of Henan Province Xinxiang453007 China

The emerging technology of adversarial information hiding can generate adversarial example by embedding useful information instead of meaningless noise into the host image. The obtained image can deceive classification models and therefore equip the function of privacy protection. However, existing research mainly focuses on the effectiveness of attack rather than preserving the integrity of the protected image. It cannot meet the requirements in the fields such as medical imaging, financial transactions, and copyright protection where users' raw data should not be damaged. Therefore, we propose a novel reversible adversarial information hiding based on interpretability of neural networks. The Grad-CAM-generated heatmap is employed to identify a minimal set of high-impact pixels for embedding, ensuring that minor modifications can induce significant misclassification. Then, the user-defined secret bits are embedded into the identified pixels by using difference expansion. After extraction, the original image can be perfectly restored. The technologies of adversarial example and reversible information hiding are combined to accommodate wider applications. The proposed method is tested on CIFAR-10 with three different neural architectures (NiN, AlexNet, ResNet). Experimental results show that the proposed method can completely restore the cover image while ensuring the extraction of the embedded data and misleading the neural network. © 2025 Elsevier Inc.

关键词： Adversarial attacks Adversarial information hiding Neural network interpretability Reversibility

来源：评论

学校读者我要写书评

暂无评论

Online Attentive Kernel-Based Temporal Difference Learning

arXiv

引用

arXiv 2022年

作者： Yang, Guang Chen, Xingguo Yang, Shangdong Wang, Huihui Dong, Shaokang Gao, Yang The the Jiangsu Key Laboratory of Big Data Security & Intelligent Processing Nanjing University of Posts and Telecommunications National Engineering Laboratory for Agri-Product Quality Traceability Beijing Technology and Business University China The State Key Laboratory for Novel Software Technology Nanjing University China The PCA Lab Key Lab of Intelligent Perception and Systems for High-Dimensional Information Ministry of Education Jiangsu Key Lab of Image and Video Understanding for Social Security School of Computer Science and Engineering Nanjing University of Science and Technology China

With rising uncertainty in the real world, online Reinforcement Learning (RL) has been receiving increasing attention due to its fast learning capability and improving data efficiency. However, online RL often suffers from complex Value Function Approximation (VFA) and catastrophic interference, creating difficulty for the deep neural network to be applied to an online RL algorithm in a fully online setting. Therefore, a simpler and more adaptive approach is introduced to evaluate value function with the kernel-based model. Sparse representations are superior at handling interference, indicating that competitive sparse representations should be learnable, non-prior, non-truncated and explicit when compared with current sparse representation methods. Moreover, in learning sparse representations, attention mechanisms are utilized to represent the degree of sparsification, and a smooth attentive function is introduced into the kernel-based VFA. In this paper, we propose an Online Attentive Kernel-Based Temporal Difference (OAKTD) algorithm using two-timescale optimization and provide convergence analysis of our proposed algorithm. Experimental evaluations showed that OAKTD outperformed several Online Kernel-based Temporal Difference (OKTD) learning algorithms in addition to the Temporal Difference (TD) learning algorithm with Tile Coding on public Mountain Car, Acrobot, CartPole and Puddle World tasks. Copyright © 2022, The Authors. All rights reserved.

关键词： Learning algorithms

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：