检索结果-内蒙古大学图书馆

Multi-Granularity Open Intent Classification via Adaptive Granular-Ball Decision Boundary

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Li, Yanhua Ouyang, Xiaocao Pan, Chaofan Zhang, Jie Zhao, Sen Xia, Shuyin Yang, Xin Wang, Guoyin Li, Tianrui School of Computing and Artificial Intelligence Southwestern University of Finance and Economics Chengdu611130 China Engineering Research Center of Intelligent Finance Ministry of Education Chengdu611130 China Key Laboratory of Big Data Intelligent Computing Chongqing Key Laboratory of Computational Intelligence Chongqing University of Posts and Telecommunications Chongqing400065 China National Center for Applied Mathematics in Chongqing Chongqing Normal University Chongqing401331 China School of Computing and Artificial Intelligence Southwest Jiaotong University Chengdu611756 China

Open intent classification is critical for the development of dialogue systems, aiming to accurately classify known intents into their corresponding classes while identifying unknown intents. Prior boundary-based methods assumed known intents fit within compact spherical regions, focusing on coarse-grained representation and precise spherical decision boundaries. However, these assumptions are often violated in practical scenarios, making it difficult to distinguish known intent classes from unknowns using a single spherical boundary. To tackle these issues, we propose a Multi-granularity Open intent classification method via adaptive Granular-Ball decision boundary (MOGB). Our MOGB method consists of two modules: representation learning and decision boundary acquiring. To effectively represent the intent distribution, we design a hierarchical representation learning method. This involves iteratively alternating between adaptive granular-ball clustering and nearest sub-centroid classification to capture fine-grained semantic structures within known intent classes. Furthermore, multi-granularity decision boundaries are constructed for open intent classification by employing granular-balls with varying centroids and radii. Extensive experiments conducted on three public datasets demonstrate the effectiveness of our proposed method. Copyright © 2024, The Authors. All rights reserved.

关键词： Semantics

DEGSTalk: Decomposed Per-Embedding Gaussian Fields for Hair-Preserving Talking Face Synthesis

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Deng, Kaijun Zheng, Dezhi Xie, Jindong Wang, Jinbao Xie, Weicheng Shen, Linlin Song, Siyang Computer Vision Institute School of Computer Science and Software Engineering Shenzhen University China National Engineering Laboratory for Big Data System Computing Technology Shenzhen University China Guangdong Provincial Key Laboratory of Intelligent Information Processing China Department of Computer Science University of Exeter United Kingdom

Accurately synthesizing talking face videos and capturing fine facial features for individuals with long hair presents a significant challenge. To tackle these challenges in existing methods, we propose a decomposed per-embedding Gaussian fields (DEGSTalk), a 3D Gaussian Splatting (3DGS)-based talking face synthesis method for generating realistic talking faces with long hairs. Our DEGSTalk employs Deformable Pre-Embedding Gaussian Fields, which dynamically adjust pre-embedding Gaussian primitives using implicit expression coefficients. This enables precise capture of dynamic facial regions and subtle expressions. Additionally, we propose a Dynamic Hair-Preserving Portrait Rendering technique to enhance the realism of long hair motions in the synthesized videos. Results show that DEGSTalk achieves improved realism and synthesis quality compared to existing approaches, particularly in handling complex facial dynamics and hair preservation. Our code will be publicly available at https://***/CVI-SZU/DEGSTalk. Copyright © 2024, The Authors. All rights reserved.

关键词： Gaussian distribution

Gnn-Mgrpool: Enhanced Graph Neural Networks with Multi-Granularity Pooling for Graph Classification

学校读者我要写书评

暂无评论

SSRN

SSRN 2023年

作者： Sun, Haichao Wang, Guoyin Liu, Qun Guo, Yike Chongqing Key Laboratory of Computational Intelligence Chongqing University of Posts and Telecommunications Chongqing400065 China Key Laboratory of Big Data Intelligent Computing Chongqing University of Posts and Telecommunications Chongqing400065 China Key Laboratory of Cyberspace Big Data Intelligent Security Ministry of Education Chongqing University of Posts and Telecommunications Chongqing400065 China Department of Computer Science and Engineering The Hong Kong University of Science and Technology 999077 Hong Kong School of Computer Engineering and Science Shanghai University Shanghai200444 China

Graph neural networks (GNNs) have gained significant attention and have been applied in various domain tasks. Currently, numerous pooling approaches have been proposed to aggregate node features and obtain node embeddings. However, current GNNs are almost black-box models that typically use a flat or single pooling step to aggregate nodes by only considering the similarity between nodes within the cluster. These approaches ignore the influence of relationships within and between clusters. To address this issue, we propose a novel multi-granular pooling method to aggregate nodes by simultaneously considering density and relationships among nodes and clusters. This method allows to obtain multi-granular node embedding clusters. The clusters in the current layer are built upon the previous layer, and these clusters change from fine to coarse as the number of clusters decreases, which is achieved by using multi-granular pooling (MgrPool for short). Additionally, the node representation of each layer is established through the ratio of node distance within clusters to that of between clusters. Finally, we conduct several experiments on node and graph classification tasks by this pooling approach. The results demonstrate that our GNN-MgrPool model outperforms the state-of-the-art similar algorithms and largely improves the interpretability of the learning process. © 2023, The Authors. All rights reserved.

关键词： Graph neural networks

Research on Improved Ant Colony Path Planning Algorithm for Updating Pheromone of Subway Inspection Mobile Robot

学校读者我要写书评

暂无评论

SSRN

SSRN 2023年

作者： Wang, Lina Yang, Xin Chen, Zeling Wang, Binrui Key Laboratory of Intelligent Manufacturing Quality Big Data Tracing and Analysis of Zhejiang Province China Jiliang University Hangzhou310018 China School of Mechanical and Electrical Engineering China Jiliang University Hangzhou310018 China

One enhanced ant colony optimization algorithm is proposed to address the issues related to slow convergence speed and susceptibility to local optima in mobile robot path planning. The algorithm employs several key improvements to achieve these goals. Firstly, the initial path is generated using the artificial potential field method. Pheromones are then added to this initial path to enhance its guidance for ants and expedite the algorithm's convergence during its initial stages. Secondly, an adaptive principle is introduced to optimize the original heuristic function by incorporating the Euclidean distance from optional nodes to the target node. This adjustment further accelerates the algorithm's convergence. Thirdly, enhancements are made to the pheromone update rule, which helps in avoiding local optima and enhances the algorithm's optimization capability. Specifically, if the length of the optimal path remains unchanged for several consecutive iterations, the pheromone concentration on the optimal path is deliberately reduced. Finally, a node optimization strategy is applied to the generated path, eliminating redundant path nodes and effectively shortening the path length. Simulation results demonstrate that the improved algorithm exhibits faster convergence and shorter path lengths compared to the classic ant colony optimization algorithm, resulting in better overall performance. In simple environments, the path length is reduced by 9.19%, and in complex environments, it is reduced by 4.38%. © 2023, The Authors. All rights reserved.

关键词： Ant colony optimization

HairDiffusion: Vivid Multi-Colored Hair Editing via Latent Diffusion

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Zeng, Yu Zhang, Yang Liu, Jiachen Shen, Linlin Deng, Kaijun He, Weizhao Wang, Jinbao Computer Vision Institute School of Computer Science & Software Engineering Shenzhen University China Shenzhen Institute of Artificial Intelligence and Robotics for Society China National Engineering Laboratory for Big Data System Computing Technology Shenzhen University China Guangdong Provincial Key Laboratory of Intelligent Information Processing China

Hair editing is a critical image synthesis task that aims to edit hair color and hairstyle using text descriptions or reference images, while preserving irrelevant attributes (e.g., identity, background, cloth). Many existing methods are based on StyleGAN to address this task. However, due to the limited spatial distribution of StyleGAN, it struggles with multiple hair color editing and facial preservation. Considering the advancements in diffusion models, we utilize Latent Diffusion Models (LDMs) for hairstyle editing. Our approach introduces Multi-stage Hairstyle Blend (MHB), effectively separating control of hair color and hairstyle in diffusion latent space. Additionally, we train a warping module to align the hair color with the target region. To further enhance multi-color hairstyle editing, we fine-tuned a CLIP model using a multi-color hairstyle dataset. Our method not only tackles the complexity of multi-color hairstyles but also addresses the challenge of preserving original colors during diffusion editing. Extensive experiments showcase the superiority of our method in editing multi-color hairstyles while preserving facial attributes given textual descriptions and reference images. © 2024, CC BY.

关键词： Color image processing

Improved RUL prediction method of wind turbine gearbox bearings under strong noises

学校读者我要写书评

暂无评论

Improved RUL prediction method of wind turbine gearbox beari...

Youth Academic Annual Conference of Chinese Association of Automation (YAC)

作者： Sheng Xiang Penghua Li Jie Hou Mamadsho Ilolov Jiwei Wang Deqi Huang Lirong Xie School of Automation& Key Laboratory of Intelligent Computing for Big Data Chongqing University of Posts and Telecommunications Chongqing China Center for Innovative Development of Science and New Technologies National Academy of Sciences of Tajikistan Dushanbe Tajikistan School of Electrical Engineering Xinjiang University Urumqi China

ISBN: (数字)9798350379228

ISBN: (纸本)9798350390780

As essential components within the wind turbine gearbox, bearings are critical, and forecasting their Remaining Useful Life (RUL) is paramount. To enhance the integration of current inputs with historical data in Recurrent Neural Network (RNN) models, the Mogrifier Long Short-Term Memory (LSTM) approach has been devised. It interacts with inputs and recurrent data in a predetermined fashion, marking its initial application in wind turbine gearbox bearing RUL prediction. Furthermore, the Relative Standard Deviation of Envelope Spectrum (RSDES), a reliable health indicator (HI) under high noise conditions, is introduced for precise evaluation of the gearbox bearing health. Consequently, a novel RUL prediction technique is introduced. Initially, RSDES from vibration data serves as HIs. Subsequently, utilizing these HIs, the Mogrifier LSTM predicts future HIs, and the bearing’s RUL is determined once the predicted HI surpasses the failure threshold. Empirical outcomes underscore the proposed method’s superiority over existing techniques in bearing RUL prediction.

关键词： Vibrations Recurrent neural networks Accuracy Noise data models Wind turbines Reliability

Privacy-Preserving Federated Learning Framework with Resisting Source Inference Attacks in Smart Healthcare

学校读者我要写书评

暂无评论

SSRN

SSRN 2024年

作者： Chenghe, Dong Xu, Guangquan Zhang, Jianhong Bai, Hongpeng Hao, Peng College of Intelligence and Computing Tianjin University Tianjin300354 China School of Big Data Qingdao Huanghai University Qingdao China Tianjin300350 China Department of Electrical and Computer Engineering North China University of Technology Beijing100144 China Key Laboratory of Intelligent Education Technology Application of Zhejiang Province School of Computer Science and Technology Zhejiang Normal University Jinhua321004 China

Federated Learning (FL) is a distributed machine learning technology that is extensive applications in smart healthcare, since it allows multiple medical institutions to jointly train medical diagnostic models without sharing their local sensitive dataset with others. Despite its benefits, FL is susceptible to source inference attacks and model poisoning. To avoid these issues, we propose a federated learning framework for smart healthcare that preserves privacy and employs certificateless ring signatures to counteract potential attacks. During the collaborative training process of the FL model, certificateless ring signatures are used to obscure the origin of parameter updates, ensuring the anonymity of the parameter sources. This approach has the advantage of being computationally efficient due to no complicated pairing operations. Furthermore, compared with the previous certificateless ring signature schemes, our proposed scheme is more effective in reducing the success rate of source inference attacks, and is efficient in terms of computation and communication overhead. © 2024, The Authors. All rights reserved.

关键词： Federated learning

Fast Latent Factor Analysis via a Fuzzy PID-Incorporated Stochastic Gradient Descent Algorithm

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Li, Jinli Yuan, Ye School of Computer Science and Technology Chongqing University of Posts and Telecommunications Chongqing400065 China Chongqing Key Laboratory of Big Data and Intelligent Computing Chongqing Engineering Research Center of Big Data Application for Smart Cities Chongqing Institute of Green and Intelligent Technology Chinese Academy of Sciences Chongqing400714 China College of Computer and Information Science Southwest University Chongqing400715 China

A high-dimensional and incomplete (HDI) matrix can describe the complex interactions among numerous nodes in various big data-related applications. A stochastic gradient descent (SGD)-based latent factor analysis (LFA) model is remarkably effective in extracting valuable information from an HDI matrix. However, such a model commonly encounters the problem of slow convergence because a standard SGD algorithm learns a latent factor relying on the stochastic gradient of current instance error only without considering past update information. To address this critical issue, this paper innovatively proposes a Fuzzy PID-incorporated SGD (FPS) algorithm with two-fold ideas: 1) rebuilding the instance learning error by considering the past update information in an efficient way following the principle of PID, and 2) implementing hyper-parameters and gain parameters adaptation following the fuzzy rules. With it, an FPS-incorporated LFA model is further achieved for fast processing an HDI matrix. Empirical studies on six HDI datasets demonstrate that the proposed FPS-incorporated LFA model significantly outperforms the state-of-the-art LFA models in terms of computational efficiency for predicting the missing data of an HDI matrix with competitive accuracy. Copyright © 2023, The Authors. All rights reserved.

关键词： Matrix algebra

GESH-NET: GRAPH-ENHANCED SPHERICAL HARMONIC CONVOLUTIONAL NETWORKS FOR CORTICAL SURFACE REGISTRATION

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Zhang, Ruoyu Wang, Lihui Tang, Kun Xu, Jingwen Wei, Hongjiang Engineering Research Center of Text Computing & Cognitive Intelligence Ministry of Education Key Laboratory of Intelligent Medical Image Analysis and Precise Diagnosis of Guizhou Province State Key Laboratory of Public Big Data College of Computer Science and Technology Guizhou University Guiyang550025 China School of Biomedical Engineering Shanghai Jiao Tong University Shanghai China

Cortical surface registration plays a crucial role in coordinating individual cortical functions and anatomical features, serving as a fundamental step in cortical surface analysis. Its aim is to align the anatomical or functional regions of different individuals, which is of great importance for neuroimaging studies across different populations. Currently, cortical surface registration techniques based on classical methods have been well developed. However, a key issue with classical registration methods is that for each pair of images to be registered, it is necessary to search for the optimal transformation in the deformation space according to a specific optimization algorithm until the similarity measure function converges, which cannot meet the requirements of real-time and high-precision in medical image registration. With the spectacular success of deep learning in the field of computer vision, researching cortical surface image registration techniques based on deep learning models has become a new direction. But so far, there are still only a few studies on cortical surface image registration based on deep learning. Moreover, although deep learning methods theoretically have stronger representation capabilities, surpassing the most advanced classical methods in registration accuracy and distortion control remains a challenge. Therefore, to address this challenge, this paper constructs a deep learning model to study the technology of cortical surface image registration. The specific work is as follows: (1) An unsupervised cortical surface registration network based on a multi-scale cascaded structure is designed, and a convolution method based on spherical harmonic transformation is introduced to register cortical surface data. This solves the problem of scale-inflexibility of spherical feature transformation and optimizes the multi-scale registration process. The results show that the proposed network outperforms the other deep learning-based registration m

关键词： Spheres