检索结果-内蒙古大学图书馆

Sequential Causal Effect Estimation by Jointly modeling the Unmeasured Confounders and Instrumental Variables

IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING 2025年第2期37卷 910-922页

作者： Sun, Zexu He, Bowei Shen, Shiqi Wang, Zhipeng Gong, Zhi Ma, Chen Qi, Qi Chen, Xu Renmin Univ China Gaoling Sch Artificial Intelligence Beijing 100872 Peoples R China Beijing Key Lab Big Data Management & Anal Methods Beijing 100872 Peoples R China City Univ Hong Kong Dept Comp Sci Hong Kong 999077 Peoples R China Tencent Wechat Beijing 100160 Peoples R China

Sequential causal effect estimation has recently attracted increasing attention from research and industry. While the existing models have achieved many successes, there are still many limitations. Existing models usually assume the causal graphs to be sufficient, i.e., there are no latent factors, such as the unmeasured confounders and instrumental variables. However, in real-world scenarios, it is hard to record all of the factors in the observational data, which makes the causally sufficient assumptions not hold. Moreover, existing models mainly focus on discrete treatments rather than continuous ones. To alleviate the above problems, in this paper, we propose a novel Continous Causal Model by explicitly capturing the Latent Factors (called (CM)-M-2-LF for short). Specifically, we define a sequential causal graph by simultaneously considering the unmeasured confounders and instrumental variables. Second, we describe the independence that should be satisfied among different variables from the mutual information perspective and further propose our learning objective. Then, we reweight different samples in the continuous treatment space to optimize our model unbiasedly. Beyond the above designs, we also theoretically analyze our model's causal identifiability and unbiasedness. Finally, we conduct extensive experiments on both simulation and real-world datasets to demonstrate the effectiveness of our proposed model.

关键词： Instruments Estimation History Mutual information Mathematical models computational modeling Sun Reliability Predictive models Long short term memory Continuous treatment instrumental variables sequential causal effects unmeasured confounders

来源：评论

学校读者我要写书评

暂无评论

Thompson Sampling and Proportional-Greedy Algorithm for Uncertain Coded Edge Computing

引用

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY 2025年第3期74卷 4865-4876页

作者： Kong, Linglin Sung, Chi Wan Shum, Kenneth W. City Univ Hong Kong Dept Elect Engn Hong Kong Peoples R China Chinese Univ Hong Kong Shenzhen Sch Sci & Engn Shenzhen 518172 Peoples R China

This paper investigates the allocation of coded computing tasks in edge networks with uncertain conditions. Specifically, it focuses on scenarios where the computing speed-related parameters of edge devices are heterogeneous and unknown. The challenge lies in mitigating the impact of stragglers, which involves identifying fast workers and distributing a reasonable workload among them. In this paper, we put the fast worker identification into a multi-armed bandit (MAB) framework and propose a Thompson sampling (TS)-based approach to tackle it. Then, our approach leverages the heterogeneity of computing speeds by formulating the task allocation problem to minimize the expected computing delay. We derive a lower bound for the delay and prove that our proposed algorithm minimizes this bound. Importantly, the time complexity of our algorithm is independent of the number of tasks to be assigned, which is typically large in practical scenarios. When our scheme is applied to solve the linear regression problem, simulation results show that it reduces the computing delay of a state-of-the-art method by more than 15% in two different scenarios.

关键词： Delays Vectors computational modeling Edge computing Resource management Decoding Optimization Load modeling computational efficiency Codes Coded edge computing distributed computing computation offloading straggler mitigation multi-armed bandits Thompson sampling

来源：评论

学校读者我要写书评

暂无评论

Token Masking Transformer for Weakly Supervised Object Localization

引用

IEEE TRANSACTIONS ON MULTIMEDIA 2025年 27卷 2059-2069页

作者： Xu, Wenhao Wang, Changwei Xu, Rongtao Xu, Shibiao Meng, Weiliang Zhang, Man Zhang, Xiaopeng Beijing Univ Posts & Telecommun Sch Artificial Intelligence Beijing 100876 Peoples R China Qilu Univ Technol Shandong Acad Sci Shandong Comp Sci Ctr Shandong Fundamental Res Ctr Comp SciNatl Superco Jinan 250013 Peoples R China Chinese Acad Sci Inst Automat State Key Lab Multimodal Artificial Intelligence S Beijing 100045 Peoples R China Mohamed Bin Zayed Univ Artificial Intelligence Abu Dhabi U Arab Emirates Chinese Acad Sci Inst Automat Beijing 100190 Peoples R China

Weakly supervised object localization (WSOL) is both a promising and challenging task that aims to achieve object localization exclusively through image category labels for supervision. Visual transformers have recently been applied to WSOL, demonstrating significant success through the exploitation of long-range feature dependencies in self-attention mechanisms. However, the transformer-based approach suffers from the same partial activation problem as the CNN-based approach due to the use of the classification task to train self-attention map, i.e., only a few discriminative regions are assigned high attention response and thus the localization map does not cover the whole object. To alleviate this problem, we propose a plug-and-play Token Masking Transformer (TMT) method to help transformer-based WSOL methods to obtain a more complete localization map by dynamic discriminative token masking. Specifically, a batch-wise discriminative token selection strategy is first introduced to flexibly determine the tokens to be masked in each image. Then, we design a token masking transformer block to perform token masking and inspire the network to mine more object-related tokens. Besides, we also design an intermediate token activation loss to further improve the performance of TMT by imposing constraints on intermediate tokens. Extensive experiments demonstrate that our TMT can substantially improve the performance of existing transformer-based methods without increasing the computational cost, and achieves state-of-the-art performance on two mainstream benchmarks.

关键词： Transformers Location awareness Visualization Training Computer vision computational modeling Semantics Hands Computer architecture Image resolution Weakly supervised object localization visual transformer token selection

来源：评论

学校读者我要写书评

暂无评论

How to Collaborate: Towards Maximizing the Generalization Performance in Cross-Silo Federated Learning

引用

IEEE TRANSACTIONS ON MOBILE COMPUTING 2025年第4期24卷 3211-3222页

作者： Sun, Yuchang Kountouris, Marios Zhang, Jun Hong Kong Univ Sci & Technol Dept Elect & Comp Engn Clear Water Bay Hong Kong Peoples R China Univ Granada Andalusian Res Inst Data Sci & Computat Intelligen Dept Comp Sci & Artificial Intelligence Granada 18012 Spain EURECOM F-06410 Biot France

Federated learning (FL) has attracted vivid attention as a privacy-preserving distributed learning framework. In this work, we focus on cross-silo FL, where clients become the model owners after training and are only concerned about the model's generalization performance on their local data. Due to the data heterogeneity issue, asking all the clients to join a single FL training process may result in model performance degradation. To investigate the effectiveness of collaboration, we first derive a generalization bound for each client when collaborating with others or when training independently. We show that the generalization performance of a client can be improved by collaborating with other clients that have more training data and similar data distributions. Our analysis allows us to formulate a client utility maximization problem by partitioning clients into multiple collaborating groups. A hierarchical clustering-based collaborative training (HCCT) scheme is then proposed, which does not need to fix in advance the number of groups. We further analyze the convergence of HCCT for general non-convex loss functions which unveils the effect of data similarity among clients. Extensive simulations show that HCCT achieves better generalization performance than baseline schemes, whereas it degenerates to independent training and conventional FL in specific scenarios.

关键词： Training Data models Collaboration computational modeling Training data Servers Games Convergence Analytical models Mobile computing Federated learning generalization collaboration pattern hierarchical cluster

来源：评论

学校读者我要写书评

暂无评论

Enhancing Physical Layer Communication Security through Generative AI with Mixture of Experts

引用

IEEE WIRELESS COMMUNICATIONS 2025年第3期32卷 176-184页

作者： Zhao, Changyuan Du, Hongyang Niyato, Dusit Kang, Jiawen Xiong, Zehui Kim, Dong In Shen, Xuemin (Sherman) Letaief, Khaled B. Nanyang Technol Univ Singapore Singapore CNRS CREATE Singapore Singapore Univ Hong Kong Hong Kong Peoples R China Guangdong Univ Technol Guangzhou Peoples R China Singapore Univ Technol & Design Singapore Singapore Sungkyunkwan Univ Seoul South Korea Univ Waterloo Waterloo ON Canada Hong Kong Univ Sci & Technol Hong Kong Peoples R China

AI technologies have become increasingly adopted in wireless communications. As an emerging type of AI technologies, generative artificial intelligence (GAI) is gaining attention in communication security. Due to its powerful learning ability, GAI models have demonstrated superiority over conventional AI methods. However, GAI still has several limitations, including high computational complexity and limited adaptability. Mixture of experts (MoE) technology, which uses multiple expert models for prediction through a gate mechanism, proposes possible solutions. In this article, we first review GAI model applications in physical layer communication security, discuss limitations, and explore how MoE can help GAI overcome these limitations. Furthermore, we propose an MoE-enabled GAI framework for network optimization problems for communication security. To demonstrate the framework's effectiveness, we provide a case study in a cooperative- friendly jamming scenario. The experimental results show that the MoE-enabled framework effectively assists the GAI algorithm, solves its limitations, and enhances communication security.

关键词： Security Adaptation models Data models Artificial intelligence Physical layer computational modeling Jamming Noise Generative AI Eavesdropping

来源：评论

学校读者我要写书评

暂无评论

A Physics-Enhanced Separation Framework for Spatiotemporal modeling of Distributed Parameter Systems With Multi-Fidelity Data

引用

IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING 2025年 22卷 13968-13982页

作者： Wang, Bing-Chuan He, Yan-Bo Wang, Yong Li, Han-Xiong Cent South Univ Sch Automat Changsha 410083 Peoples R China East China Univ Sci & Technol Key Lab Smart Mfg Energy Chem Proc Minist Educ Shanghai 200237 Peoples R China City Univ Hong Kong Dept Syst Engn Hong Kong Peoples R China

Many industrial processes are typically distributed parameter systems (DPSs) described by partial differential equations. Data-driven methods have become popular for spatiotemporal modeling of DPSs, which is crucial for system understanding, simulation, and control improvement. However, current data-driven methods rely heavily on data volume and fidelity. With limited high-fidelity data, they exhibit unsatisfactory long-term predictions for out-of-sample scenarios. To remedy this issue, a novel physics-enhanced separation framework (called PhysiT/S) is proposed. PhysiT/S is composed of a physics-enhanced spatiotemporal unit and a coefficient network. By enhancing physics utilization through the physics-enhanced spatiotemporal unit, PhysiT/S becomes less data-dependent while computationally efficient. Multi-fidelity learning of the coefficient network further alleviates the request for a large volume of high-fidelity data. As a result, PhysiT/S can accurately predict out-of-sample distributions, even when only limited high-fidelity data is available. PhysiT/S introduces a novel way of combining physics with multi-fidelity data for spatiotemporal modeling of DPSs. Extensive simulations on two benchmark DPSs and the thermal process of lithium-ion batteries demonstrate the merits of PhysiT/S. Note to Practitioners-This paper is motivated by the problem of data-driven spatiotemporal modeling of DPSs, which is also applicable to other prediction tasks for complex spatiotemporal systems. Existing data-driven approaches rely heavily on high-fidelity data, while physics-informed approaches struggle to extract coupled spatiotemporal features. These drawbacks limit their practical applications to engineering modeling. In this paper, we propose a new generic modeling framework that integrates physics with data through the theory of time/space separation. It aims to achieve accurate long-term predictions for out-of-sample scenarios by constructing spatiotemporal enhancement unit

关键词： Spatiotemporal phenomena Feature extraction Physics Data models modeling computational modeling Data mining Accuracy Mathematical models Automation Distributed parameter systems (DPSs) spatiotemporal modeling time/space separation physics-enhanced learning multi-fidelity data

来源：评论

学校读者我要写书评

暂无评论

Splitting State-Space Method for Converter-Integrated Power Systems EMT Simulations

引用

IEEE TRANSACTIONS ON POWER DELIVERY 2025年第1期40卷 584-595页

作者： Fu, Xiaopeng Wu, Wei Li, Peng Mahseredjian, Jean Wu, Jianzhong Wang, Chengshan Tianjin Univ Key Lab Smart Grid Minist Educ Tianjin 300072 Peoples R China Beijing Huairou Lab Beijing 102206 Peoples R China Polytech Montreal Dept Elect Engn Montreal PQ H3T 1J4 Canada Cardiff Univ Inst Energy Sch Engn Cardiff CF24 3AA Wales

As the utilization of power electronic-based components in power systems continues to grow, a comprehensive understanding of their dynamics becomes increasingly important for system design, control and protection analysis. To meet practical needs, the high-fidelity but time-consuming electromagnetic transient (EMT) simulations are often required. To improve the performance of these simulations, a highly efficient splitting state-space method with numerical error control is proposed that reduces the computation workload. The method employs a generic decoupling principle to split the state-space equations of the converter-integrated power system and introduces the exponential splitting formulas of multiple orders accuracy to solve and then compose the splitting state-space equations. The decoupling principle is designed based on separation of time-varying portions of the state matrix, which is realized by locating the smallest subcircuit topology that is switch state-dependent, through automatic switch grouping and switch adjacent state variables (SASV) identification. A family of exponential splitting schemes is employed to accelerate the demanding matrix exponential calculation. The splitting state-space method undergoes comprehensive testing across various cases, including a distribution network with DC load, an LLC resonant converter, a large-scale wind farm, and an MMC circuit. The accuracy of the proposed method is thoroughly evaluated, and its efficiency is validated.

关键词： Matrix converters Switches Mathematical models Integrated circuit modeling Accuracy computational modeling Topology Transient analysis State-space methods Power system dynamics Electromagnetic transient exponential integrator power converter splitting method

来源：评论

学校读者我要写书评

暂无评论

A Physics-Coupled Deep Learning Framework for Hydrodynamic Diffusion modeling in Watershed Systems: Integrating Spatiotemporal Networks and Environmental Constraints

引用

IEEE ACCESS 2025年 13卷 34985-35003页

作者： Jia, L. Yen, N. Pei, Y. Univ Aizu Grad Sch Comp Sci & Engn Aizu Wakamatsu Fukushima 9658580 Japan Univ Aizu Comp Sci Div Aizu Wakamatsu Fukushima 9658580 Japan

The modeling and simulation of complex spatiotemporal systems are crucial for understanding and solving multidimensional dynamical systems, particularly in earth and environmental sciences. Accurate comprehension and computational modeling of fluid transport, environmental processes, and substance diffusion depend heavily on solving governing equations. Despite significant advancements in artificial intelligence techniques, such as deep learning and neural operator methods, challenges persist regarding robustness, scalability, and adherence to physical laws in hydrodynamic systems. This paper introduces a multi-scale interdisciplinary hybrid learning framework that integrates physics-informed neural networks with neural operator-based deep learning techniques to model hydrodynamic transport processes. By incorporating convolutional neural networks for multi-scale feature extraction and implementing hard constraints to enforce physical boundary conditions, the proposed framework enhances the stability and accuracy of predictions in dynamic fluid systems. The approach facilitates efficient reconstruction of spatiotemporal characteristics and parameterized dynamics while ensuring physical consistency. Through case studies of two-dimensional solute diffusion equations, the framework demonstrates superior generalizability and robustness in addressing high-dimensional and nonlinear fluid systems. Comparative experiments with multiple baseline models highlight significant improvements in prediction accuracy, noise resistance, and adaptability to sparse datasets. The proposed method effectively addresses limitations of traditional data-driven approaches by integrating domain-specific physical principles, providing a reliable and high-fidelity solution for hydrodynamic modeling and environmental system monitoring. This research offers a transformative approach to feature extraction, dynamic representation, and physically consistent modeling of complex spatiotemporal systems.

关键词： Mathematical models Data models computational modeling Spatiotemporal phenomena Deep learning Boundary conditions Watersheds Biological neural networks Feature extraction Adaptation models Deep neural networks fluid dynamics modeling physical-coupled machine learning watershed systems environmental constraints

来源：评论

学校读者我要写书评

暂无评论

FPGA Acceleration With Hessian-Based Comprehensive Intra-Layer Mixed-Precision Quantization for Transformer Models

引用

IEEE ACCESS 2025年 13卷 70282-70297页

作者： Byun, Woohong Woo, Jongseok Mukhopadhyay, Saibal Georgia Inst Technol Sch Elect & Comp Engn Atlanta GA 30332 USA Samsung Display Co Ltd Mobile Display Design Team Yongin 17096 South Korea

Recent advancements in using FPGAs as co-processors for language model acceleration, particularly for energy efficiency and flexibility, face challenges due to limited memory capacity. This limitation hinders the deployment of transformer-based language models. To address this challenge, we propose a novel software-hardware co-optimization framework that integrates Hessian-based intra-layer mixed-precision quantization with a runtime bit-configurable FPGA accelerator. Our proposed Hessian-based row-wise weight quantization addresses hardware inefficiencies in traditional parameter-wise and channel-wise approaches by enabling mixed-precision weight matrices to be split into two uniform-precision matrices, thereby simplifying hardware requirements. Additionally, our Query-Key coupled attention activation quantization optimally aligns precision within each outer product pair in attention calculations, reducing hardware complexity and memory management overhead. Our concurrent quantization method balances the benefits of row-wise weight quantization and Query-Key coupled activation quantization while maximizing energy efficiency through multi-precision optimization. To support this algorithm, we design a multi-precision FPGA accelerator capable of handling both 2(n)-based and non-2(n) mixed-precision operations. It is implemented on a single Xilinx ZCU102 FPGA board, operating at 200MHz with a power consumption of 15.08W during inference on the 110-million-parameter BERT-Base and 345-million-parameter GPT-2 Medium transformer models. Coupled with the proposed algorithm and dataflow optimization, it enables on-chip storage of all necessary parameters, minimizing off-chip memory access. Experimental results demonstrate that our FPGA accelerator significantly outperforms existing solutions, achieving energy efficiency improvements ranging from 2.22 x to 17.23 x over state-of-the-art FPGA accelerators.

关键词： Quantization (signal) Hardware Field programmable gate arrays Transformers Energy efficiency Optimization Memory management computational modeling Accuracy Sparse matrices Accelerator activation BERT FPGA hardware acceleration Hessian language model mixed-precision NLP quantization transformer

来源：评论

学校读者我要写书评

暂无评论

Rethinking Lightweight RGB-Thermal Salient Object Detection With Local and Global Perception Network

引用

IEEE INTERNET OF THINGS JOURNAL 2025年第11期12卷 18056-18069页

作者： Jin, Dongze Shao, Feng Xie, Zhengxuan Mu, Baoyang Chen, Hangwei Ningbo Univ Fac Informat Sci & Engn Ningbo 315211 Peoples R China

RGB-thermal salient object detection (RGB-T SOD) aims to segment the most intriguing parts through RGB and thermal images. However, the high computational costs and large model sizes of existing methods prevent its deployment in edge computing platforms. To cope with it, a new lightweight RGB-T SOD method is proposed, namely, local and global perception network (LGPNet), which relieves the pressure of data transmission and processing and improves robustness in complicated surroundings. Specifically, high-level lightweight fusion (HLF) blocks and low-level lightweight fusion (LLF) blocks are designed to reduce the difference between two modalities and perform multimodal feature fusion. Unlike existing convolutional neural network-based lightweight SOD methods, HLF and LLF blocks combine the spatial inductive bias of convolutional neural networks with the global perception of Transformer, which is able to perform feature extraction and fusion with fewer parameters and larger receptive field. Experimental results show that our method is able to compete with the state-of-the-art RGB-T SOD methods while having only 7.35 learnable parameters[M], 6.40G floating-point operations and real-time speed (33 frames per second (FPS) on PyTorch framework and 224 FPS on TensorRT framework).

关键词： computational modeling Feature extraction Transformers Visualization Internet of Things Data models Convolution computational efficiency Semantics Predictive models computational cost lightweight framework RGB-thermal salient object detection (RGB-T SOD) Transformer

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：