检索结果-内蒙古大学图书馆

SSFam: Scribble Supervised Salient Object Detection Family

IEEE TRANSACTIONS ON MULTIMEDIA 2025年 27卷 1988-2000页

作者： Liu, Zhengyi Deng, Sheng Wang, Xinrui Wang, Linbo Fang, Xianyong Tang, Bin Anhui Univ Sch Comp Sci & Technol Key Lab Intelligent Comp & Signal Proc Minist Educ Hefei 230039 Peoples R China Hefei Univ Sch Artificial Intelligence & Big Data Hefei 230601 Peoples R China

Scribble supervised salient object detection (SSSOD) constructs segmentation ability of attractive objects from surroundings under the supervision of sparse scribble labels. For the better segmentation, depth and thermal infrared modalities serve as the supplement to RGB images in the complex scenes. Existing methods specifically design various feature extraction and multi-modal fusion strategies for RGB, RGB-Depth, RGB-Thermal, and Visual-Depth-Thermal image input respectively, leading to similar model flood. As the recently proposed Segment Anything Model (SAM) possesses extraordinary segmentation and prompt interactive capability, we propose an SSSOD family based on SAM, named SSFam, for the combination input with different modalities. Firstly, different modal-aware modulators are designed to attain modal-specific knowledge which cooperates with modal-agnostic information extracted from the frozen SAM encoder for the better feature ensemble. Secondly, a siamese decoder is tailored to bridge the gap between the training with scribble prompt and the testing with no prompt for the stronger decoding ability. Our model demonstrates the remarkable performance among combinations of different modalities and refreshes the highest level of scribble supervised methods and comes close to the ones of fully supervised methods.

关键词： Object detection Decoding Training Feature extraction Modulation Image segmentation Annotations Testing data mining Visualization Salient object detection scribble supervision multi-modal segment anything model unified model

来源：评论

学校读者我要写书评

暂无评论

Class-Incremental Learning: A Survey

引用

IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE intelligence 2024年第12期46卷 9851-9873页

作者： Zhou, Da-Wei Wang, Qi-Wei Qi, Zhi-Hong Ye, Han-Jia Zhan, De-Chuan Liu, Ziwei Nanjing Univ Sch Artificial Intelligence Natl Key Lab Novel Software Technol Nanjing 210023 Peoples R China Nanyang Technol Univ Coll Comp & Data Sci S Lab Singapore 639798 Singapore

Deep models, e.g., CNNs and Vision Transformers, have achieved impressive achievements in many vision tasks in the closed world. However, novel classes emerge from time to time in our ever-changing world, requiring a learning system to acquire new knowledge continually. Class-Incremental Learning (CIL) enables the learner to incorporate the knowledge of new classes incrementally and build a universal classifier among all seen classes. Correspondingly, when directly training the model with new class instances, a fatal problem occurs - the model tends to catastrophically forget the characteristics of former ones, and its performance drastically degrades. There have been numerous efforts to tackle catastrophic forgetting in the machine learning community. In this paper, we survey comprehensively recent advances in class-incremental learning and summarize these methods from several aspects. We also provide a rigorous and unified evaluation of 17 methods in benchmark image classification tasks to find out the characteristics of different algorithms empirically. Furthermore, we notice that the current comparison protocol ignores the influence of memory budget in model storage, which may result in unfair comparison and biased results. Hence, we advocate fair comparison by aligning the memory budget in evaluation, as well as several memory-agnostic performance measures.

关键词： Catastrophic forgetting class-incremental learning continual learning lifelong learning

来源：评论

学校读者我要写书评

暂无评论

Information-controlled graph convolutional network for multi-view semi-supervised classification

引用

NEURAL NETWORKS 2025年 184卷 107102页

作者： Shi, Yongquan Pi, Yueyang Liu, Zhanghui Zhao, Hong Wang, Shiping Fuzhou Univ Coll Comp & Data Sci Fuzhou 350108 Peoples R China Fujian Prov Univ Key Lab Intelligent Metro Fuzhou 350108 Peoples R China Minnan Normal Univ Sch Comp Sci Zhangzhou 363000 Peoples R China Minnan Normal Univ Key Lab Data Sci & Intelligence Applicat Zhangzhou 363000 Peoples R China

Graph convolutional networks have achieved remarkable success in the field of multi-view learning. Unfortunately, most graph convolutional network-based multi-view learning methods fail to capture long-range dependencies due to the over-smoothing problem. Many studies have attempted to mitigate this issue by decoupling graph convolution operations. However, these decoupled architectures lead to the absence of feature transformation module, thus limiting the expressive power of the model. To this end, we propose an information-controlled graph convolutional network for multi-view semi-supervised classification. In the proposed method, we maintain the paradigm of node embeddings during propagation by imposing orthogonality constraints on the feature transformation module. By further introducing a damping factor based on residual connections, we theoretically demonstrate that the proposed method can alleviate the over-smoothing problem while retaining the feature transformation module. Furthermore, we prove that the proposed model can stabilize both forward inference and backward propagation in graph convolutional networks. Extensive experimental results on benchmark datasets demonstrate the effectiveness of the proposed method.

关键词： Multi-view learning Semi-supervised classification Graph convolutional network Layer normalization

来源：评论

学校读者我要写书评

暂无评论

Identifying the hierarchical emotional areas in the human brain through information fusion

引用

INFORMATION FUSION 2025年 113卷

作者： Huang, Zhongyu Du, Changde Li, Chaozhuo Fu, Kaicheng He, Huiguang Chinese Acad Sci Inst Automat Key Lab Brain Cognit & Brain inspired Intelligence Beijing 100190 Peoples R China Univ Chinese Acad Sci Sch Artificial Intelligence Beijing 100049 Peoples R China Beijing Univ Posts & Telecommun MoE Key Lab Trustworthy Distributed Comp & Serv Beijing 100876 Peoples R China

The brain basis of emotion has consistently received widespread attention, attracting a large number of studies to explore this cutting-edge topic. However, the methods employed in these studies typically only model the pairwise relationship between two brain regions, while neglecting the interactions and information fusion among multiple brain regions-one of the key ideas of the psychological constructionist hypothesis. To overcome the limitations of traditional methods, this study provides an in-depth theoretical analysis of how to maximize interactions and information fusion among brain regions. Building on the results of this analysis, we propose to identify the hierarchical emotional areas in the human brain through multi-source information fusion and graph machine learning methods. comprehensive experiments reveal that the identified hierarchical emotional areas, from lower to higher levels, primarily facilitate the fundamental process of emotion perception, the construction of basic psychological operations, and the coordination and integration of these operations. Overall, our findings provide unique insights into the brain mechanisms underlying specific emotions based on the psychological constructionist hypothesis.

关键词： Emotion Human brain Hierarchical emotional areas Multi-source information fusion Graph machine learning Psychological constructionism

来源：评论

学校读者我要写书评

暂无评论

TITE: A transformer-based deep reinforcement learning approach for traffic engineering in hybrid SDN with dynamic traffic

引用

FUTURE GENERATION compUTER SYSTEMS-THE INTERNATIONAL JOURNAL OF ESCIENCE 2024年 161卷 95-105页

作者： Lin, Bin Guo, Yingya Luo, Huan Ding, Mingjie Fuzhou Univ Coll Comp & Data Sci Fuzhou Peoples R China Fuzhou Univ Fujian Prov Key Lab Network Comp & Intelligent Inf Fuzhou Peoples R China Minist Educ Engn Res Ctr Big Data Intelligence Fuzhou Peoples R China

Hybrid Software Defined Networks (Hybrid SDNs), with a partial upgrade of legacy routers to SDN switches in traditional distributed networks, currently stand as a prevailing network architecture. Traffic Engineering (TE) in hybrid SDN requires the efficient and timely acquisition of a routing policy to adapt to dynamically changing traffic demands, which has recently become a hot topic. Ignoring the hidden relations of consecutive states and the compact representation of the network environment, previous Deep Reinforcement Learning (DRL)-based studies suffer from the convergence problem by only establishing the direct relationship between individual Traffic Matrix (TM) and routing policy. Therefore, to enhance TE performance in hybrid SDNs under dynamically changing traffic demands, we propose to integrate the Transformer model with DRL to establish the relationship between consecutive states and routing policies. The temporal characteristic among consecutive states can effectively assist DRL in solving the convergence problem. To obtain a compact and accurate description of the network environment, we propose to jointly consider TM, routing action, and reward in designing the state of the network environment. To better capture the temporal relations among consecutive states of the network environment, we design a multi-feature embedding module and achieve positional encodings in the Transformer model. The extensive experiments demonstrate that once the convergence problem is solved, the proposed Transformer-based DRL method can efficiently generate routing policies that adapt well to dynamic network traffic.

关键词： Traffic Engineering Dynamic traffic Transformer Deep Reinforcement Learning Hybrid SDN

来源：评论

学校读者我要写书评

暂无评论

Detecting Major Depressive Disorder by Graph Neural Network Exploiting Resting-State Functional MRI 29th

Detecting Major Depressive Disorder by Graph Neural Network ...

引用

29th International Conference on Neural Information Processing

作者： Zhao, Tianyi Zhang, Gaoyan Tianjin Univ Coll Intelligence & Comp Tianjin Key Lab Cognit Comp & Applicat Tianjin 300350 Peoples R China

ISBN: (纸本)9789819916412;9789819916429

Major Depressive Disorder (MDD) has raised concern worldwide because of its prevalence and ambiguous neuropathophysiology. Resting-state functional MRI (rs-fMRI) is an applicable tool for measuring abnormal brain functional connectivity in MDD. However, effective method for early diagnosis and treatment for MDD is still lacking. In this study, we propose a three-stage classification framework to analyze rs-fMRI data for the diagnosis of MDD. We first apply self-supervised pretraining on developed graph encoder, incorporating triplet relationship among input subjects, to enable higher ability to learn robust and discriminative graph representations. Then, supervised classification is performed utilizing the pretrained encoder. Specifically, to better model subjects' brain as functional connectivity network, our developed graph encoder consists of following modules: non-linear feature transformation, graph isomorphism convolution, topk pooling and hierarchical readout. Afterwards, ensemble learning is implemented to further boost model's performance. Finally, we identify salient ROIs by investigating pooling scores learned by topk pooling layers, which implies brain areas potentially related to MDD and equips our model with fair interpretability. Experimental results on Rest-meta-MDD, a large-scale multisite dataset, suggest the efficacy of our method.

关键词： Major depressive disorder diagnosis Graph neural network Deep learning Resting-state functional magnetic resonance imaging Brain functional connectivity

来源：评论

学校读者我要写书评

暂无评论

Local perturbation-based black-box federated learning attack for time series classification

引用

FUTURE GENERATION compUTER SYSTEMS-THE INTERNATIONAL JOURNAL OF ESCIENCE 2024年 158卷 488-500页

作者： Chen, Shengbo Yuan, Jidong Wang, Zhihai Sun, Yongqi Minist Educ Key Lab Big Data & Artificial Intelligence Transpo Beijing Peoples R China Beijing Jiaotong Univ Sch Comp Sci & Technol Beijing 100044 Peoples R China

The widespread adoption of intelligent machines and sensors has generated vast amounts of time series data, leading to the increasing use of neural networks in time series classification. Federated learning has emerged as a promising machine learning paradigm that reduces the risk of user privacy leakage. However, federated learning is vulnerable to backdoor attacks, which pose significant security threats. Furthermore, existing unrealistic white -box methods for attacking time series result in insufficient adaptation and inferior stealthiness. To overcome these limitations, this paper proposes a gradient -free black -box method called local perturbationbased backdoor F ederated L earning A ttack for T ime S eries classification (FLATS). The attack is formulated as a constrained optimization problem and is solved using a differential evolution algorithm, without requiring any knowledge of the internal architecture of the target model. In addition, the proposed method considers the time series shapelet interval as a local perturbation range and adopts a soft target poisoning approach to minimize the difference between the attacker model and the benign model. Experimental results demonstrate that our proposed method can effectively attack federated learning time series classification models with potential security issues while generating imperceptible poisoned samples that can evade various defence methods.

关键词： Federated learning Black-box backdoor attack Local perturbation Shapelet interval Stealthy

来源：评论

学校读者我要写书评

暂无评论

RADICAL CONSTRAINT-BASED GENERATIVE ADVERSARIAL NETWORK FOR HANDWRITTEN CHINESE CHARACTER GENERATION

引用

compUTING AND INFORMATICS 2024年第2期43卷 482-504页

作者： Ye, Xi-Ling Zhang, Hong-Bo Yang, Li-Jie Du, Ji-Xiang Huaqiao Univ Dept Comp Sci & Technol Xiamen 361021 Peoples R China Huaqiao Univ Fujian Key Lab Big Data Intelligence & Secur Xiamen 361021 Peoples R China

Generative adversarial networks (GANs) have been used as a solution to handwritten Chinese character automatic generation (HCCAG) in recent years. However, most existing GAN-based methods adopt a pixel -based strategy, which ignores the radical structure of Chinese characters. To achieve better HCCAG, a radical constraint -based GAN (RC-GAN) is proposed in this work. In the proposed method, a gated recurrent unit (GRU)-based radical learning network is designed to calculate the radical components among Chinese characters, and radical consistent loss is applied to train this module. Finally, the radical learning module is fused with a cyclic structure GAN to improve the performance of Chinese character generation. The experimental results show that compared with the existing GAN, the proposed method has better performance. Specifically, the proposed method can reduce the stroke error in the generated Chinese character images.

关键词： Handwritten Chinese character generation generative adversarial network radical constraint cyclic structure

来源：评论

学校读者我要写书评

暂无评论

Short-term wind speed prediction method based on prior wind direction knowledge and multi-period decoupling

引用

ENGINEERING APPLICATIONS OF ARTIFICIAL intelligence 2025年 153卷

作者： Shang, Zewen Li, Xuewei Liu, Zhiqiang Sun, Yingzhou Yu, Jian Yu, Mei Xu, Tianyi Xiong, Wei Tianjin Univ Coll Intelligence & Comp Tianjin 300350 Peoples R China Tianjin Key Lab Cognit Comp & Applicat Tianjin 300350 Peoples R China Tianjin Key Lab Adv Networking Tianjin 300350 Peoples R China Tianjin Chengjian Univ TCU Sch Civil Engn Tianjin 300384 Peoples R China

Accurate wind speed prediction improves power system stability and efficiency. Wind direction contains airflow dynamic information, but existing methods fail to fully capture it. The multi-periodicity of meteorological data causes irregular wind speed fluctuations, complicating the capture of correlations and variances across periods. To address these issues, this paper proposes the Multi-Period Wind Direction Graph Network(MWGN). Gaussian Spatial Module(GSM) calculates the relationship between wind turbines using Gaussian distribution along the crosswind direction, thus enhancing airflow motion capture. Multi-Period Time Series Decoupling Module(MPTSDM) selects the primary periods by analyzing the frequency domain and extracts local correlation and periodic correlation, to better model the changing pattern of time series. compared with the existing models, MWGN achieves the consistent state-of-the-art in three datasets.

关键词： Wind speed forecasting Wind direction Deep learning Neural networks

来源：评论

学校读者我要写书评

暂无评论

TSCF: An Improved Deep Forest Model for Time Series Classification

引用

NEURAL PROCESSING LETTERS 2024年第1期56卷 13-13页

作者： Dai, Mingxin Yuan, Jidong Liu, Haiyang Wang, Jinfeng Minist Educ Key Lab Big Data & Artificial Intelligence Transpo Beijing Peoples R China Beijing Jiaotong Univ Sch Comp & Informat Technol Beijing 100044 Peoples R China

The deep forest presents a novel approach that yields competitive performance when compared to deep neural networks. Nevertheless, there are limited studies on the application of deep forest to time series classification (TSC) tasks, and the direct use of deep forest cannot effectively capture the relevant characteristics of time series. For that, this paper proposes time series cascade forest (TSCF), a model specifically designed for TSC tasks. TSCF relies on four base classifiers, i.e., random forest, completely random forest, random shapelet forest, and diverse representation canonical interval forest, allowing for feature learning on the original data from three granularities: point, subsequence, and summary statistics calculated based on intervals. The major contribution of this work, is to define an ensemble and deep classifier that significantly outperforms the individual classifiers and the original deep forest. Experimental results show that TSCF outperforms other forest-based algorithms for solving TSC problems.

关键词： Time series classification Deep forest Shapelet Interval

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：