检索结果-内蒙古大学图书馆

IEEE Transactions on Artificial intelligence 2025年第5期6卷 1373-1385页

作者： Dai, Yong Hong, Xiaopeng Wang, Yabin Ma, Zhiheng Jiang, Dongmei Wang, Yaowei Shenzhen Polytechnic University School of Artificial Intelligence Shenzhen518055 China Peng Cheng Laboratory Shenzhen518055 China Harbin Institute of Technology Faculty of Computing Harbin150001 China Xi’an Jiaotong University School of Electronic and Information Engineering Xiaan710049 China Shenzhen University of Advanced Technology Shenzhen518071 China Chinese Academy of Sciences Shenzhen Institutes of Advanced Technology Shenzhen518071 China Harbin Institute of Technology School of Computing Shenzhen518055 China

Contemporary continual learning approaches typically select prompts from a pool, which function as supplementary inputs to a pretrained model. However, this strategy is hindered by the inherent noise of its selection approach when handling increasing tasks. In response to these challenges, we reformulate the prompting approach for continual learning and propose the prompt customization (PC) method. PC mainly comprises a prompt generation module (PGM) and a prompt modulation module (PMM). In contrast to conventional methods that employ hard prompt selection, PGM assigns different coefficients to prompts from a fixed-sized pool of prompts and generates tailored prompts. Moreover, PMM further modulates the prompts by adaptively assigning weights according to the correlations between input data and corresponding prompts. We evaluate our method on four benchmark datasets for three diverse settings, including the class, domain, and task-agnostic incremental learning tasks. Experimental results demonstrate consistent improvement (by up to 16.2%), yielded by the proposed method, over the state-of-the-art (SOTA) techniques. The code has been released online. © 2020 IEEE.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Simple One-step Multi-view Clustering with Fast Similarity and Cluster Structure Learning

引用

IEEE Signal Processing Letters 2025年 32卷 1850-1854页

作者： Xia, Xianxian Huang, Dong Yang, Chen-Min He, Chaobo Wang, Chang-Dong South China Agricultural University College of Mathematics and Informatics Guangzhou China South China Normal University School of Computer Science Guangzhou China Sun Yat-sen University School of Computer Science and Engineering Guangzhou China Ministry of Education Key Laboratory of Machine Intelligence and Advanced Computing China

Multi-view clustering (MVC) is essential for integrating heterogeneous data from multiple sources. However, many existing approaches are hindered by high computational complexity and the separate optimization of similarity and cluster structures. In light of these challenges, this paper presents a novel anchor-based MVC method termed simple one-step multi-view clustering with fast similarity and cluster structure learning (SONIC), which models adaptive anchor learning, multi-view similarity structure learning, and discrete cluster structure learning in a joint framework. In particular, we employ the anchor-based multi-view similarity learning to capture the consensus manifold structure latent in multiple views, thereby constructing a unified bipartite graph with adaptive anchor learning and view weighting. Then we impose a low-rank constraint on the bipartite graph structure to directly reveal the desired number of clusters without additional post-processing. An efficient alternating minimization algorithm is developed to optimize the model, resulting in a computational complexity that scales linearly with the number of samples. Extensive experiments on eight benchmark datasets demonstrate the superior performance of SONIC in both clustering quality and computational efficiency. © 1994-2012 IEEE.

关键词： Data assimilation

来源：评论

学校读者我要写书评

暂无评论

iGrasp: An Interactive 2D-3D Framework for 6-DoF Grasp Detection 27th

iGrasp: An Interactive 2D-3D Framework for 6-DoF Grasp Detec...

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Jiang, Jian-Jian Wu, Xiao-Ming Chen, Zibo Wei, Yi-Lin Zheng, Wei-Shi School of Computer Science and Engineering Sun Yat-sen University Guangzhou China Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education Guangzhou China Guangdong Province Key Laboratory of Information Security Technology Sun Yat-sen University Guangzhou China

ISBN: (纸本)9783031781124

For 6-DoF grasp detection, we aim at introducing a new interactive 2D-3D framework which filters out irrelevant information and makes both modalities collaborate effectively to generate robust grasps and accelerate inference speed greatly. This cannot be accomplished by existing works on learning to grasp that merely utilize 3D point clouds or leverage both 2D textures and 3D point clouds. Our framework is called iGrasp, a novel three-step design between 2D textures and 3D point clouds, where the interaction modelling enhances both modalities. Concretely, we propose the 2D-to-3D interaction to leverage objectness masks generated from 2D textures to filter out target-irrelevant information in 3D point clouds. Then, we introduce the 3D-to-2D interaction to leverage structural priors from 3D point cloud features with cross-attention and cylinder grouping to refine 2D texture features. Finally, we combine the refined 2D texture features and 3D point cloud features for generating high-quality 6-DoF grasp poses. Our experiments on the large-scale real-world dataset, namely GraspNet-1Billion, demonstrate that iGrasp surpasses state-of-the-art methods by 4.66/3.53 mAP on RealSense/Kinect and reduces the inference time by 28%. Real-world experiments further verify the effectiveness of iGrasp. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Deep learning

来源：评论

学校读者我要写书评

暂无评论

Depth-Enhanced Alignment for Label-Free 3D Semantic Segmentation 27th

Depth-Enhanced Alignment for Label-Free 3D Semantic Segment...

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Xie, Shangjin Feng, Jiawei Chen, Zibo Liu, Zhixuan Zheng, Wei-Shi School of Computer Science and Engineering Sun Yat-sen University Guangzhou China Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education Guangzhou China Guangdong Province Key Laboratory of Information Security Technology Sun Yat-sen University Guangzhou China

ISBN: (纸本)9783031784552

Labeling every point in a scene is a laborious journey for 3D understanding. To achieve annotation-free training, existing works introduce Contrastive Language-Image Pre-training (CLIP) to transfer the pre-trained capability of visual-linguistic correspondence to 3D-linguistic matching. However, directly adopting this CLIP-driven strategy can inevitably introduce bias: The overrated roles of the color and texture from an RGB image could overshadow the geometric nature of the corresponding 3D scene, resulting in a sub-optimal alignment. We note that different from RGB images, a depth map contains rich geometric information. Inspired by this, we propose Depth-Enhanced Alignment (D-EA) for label-free 3D semantic segmentation. D-EA aims to explore the rich geometric cues in depth maps and mitigate the color and texture biases rooted in the original CLIP-driven strategy. Specifically, we first tune a geometry-enhanced CLIP by aligning its depth prediction to the paired RGB prediction given by the original CLIP. Next, the point cloud feature space is matched with the RGB-Depth aggregated CLIP space by aligning point prediction to RGB and depth predictions. Moreover, to mitigate the semantic ambiguity caused by view-specific noise, we propose a View-Integrated Pseudo Label Generation paradigm. Experiments demonstrate the effectiveness of the proposed D-EA on the ScanNet (indoor) and GraspNet-1Billion (desktop) datasets in the label-free setting. Our method is also competitive in limited annotation semantic segmentation. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Semantic Segmentation

来源：评论

学校读者我要写书评

暂无评论

Privacy-Preserving Face Recognition with Adaptive Generative Perturbations 27th

Privacy-Preserving Face Recognition with Adaptive Generative...

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Zhang, Delong Peng, Yixing Wu, Ancong Zheng, Wei-Shi School of Computer Science and Engineering Sun Yat-sen University Guangzhou China Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education Guangzhou China Guangdong Province Key Laboratory of Information Security Technology Sun Yat-sen University Guangzhou China

ISBN: (纸本)9783031783401

For online face recognition services, the potential leakage of facial features and reconstruction techniques gives malicious attackers the opportunity to reconstruct face images, raising public concern about privacy. Previous privacy-preserving face recognition methods either require retraining the face recognition model or iterative perturbation that increases inference time. To overcome these limitations, we propose an efficient plug-and-play method that crafts Adaptive Generative perturbations for frozen Face recognition model (AGFace) to defend the reconstruction attacks. To generate perturbation with a single forward for shallow features extracted by frozen recognition model, we learn a Perturbation Generator to efficiently mine the adversarial perturbations by simultaneously minimizing the perturbation and maximizing the reconstruction error of a proxy reconstructor. To achieve privacy-utility trade-off, we propose the Adaptive Channel Selector to identify top-k reconstruction-sensitive channels for the features. By selecting these channels for perturbation, the dominant visual privacy information is protected with reconstruction-insensitive discriminative information preserved. Extensive experiments demonstrate that AGFace achieves state-of-the-art performance in terms of both privacy and utility among retraining-free methods and is comparable to retraining-dependent methods. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Differential privacy

来源：评论

学校读者我要写书评

暂无评论

Towards Completeness: A Generalizable Action Proposal Generator for Zero-Shot Temporal Action Localization 27th

Towards Completeness: A Generalizable Action Proposal Genera...

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Du, Jia-Run Lin, Kun-Yu Meng, Jingke Zheng, Wei-Shi School of Computer Science and Engineering Sun Yat-sen University Guangzhou China Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education Guangzhou China Guangdong Province Key Laboratory of Information Security Technology Sun Yat-sen University Guangzhou China

ISBN: (纸本)9783031784439

To address the zero-shot temporal action localization (ZSTAL) task, existing works develop models that are generalizable to detect and classify actions from unseen categories. They typically develop a category-agnostic action detector and combine it with the Contrastive Language-Image Pre-training (CLIP) model to solve ZSTAL. However, these methods suffer from incomplete action proposals generated for unseen categories, since they follow a frame-level prediction paradigm and require hand-crafted post-processing to generate action proposals. To address this problem, in this work, we propose a novel model named Generalizable Action Proposal generator (GAP), which can interface seamlessly with CLIP and generate action proposals in a holistic way. Our GAP is built in a query-based architecture and trained with a proposal-level objective, enabling it to estimate proposal completeness and eliminate the hand-crafted post-processing. Based on this architecture, we propose an Action-aware Discrimination loss to enhance the category-agnostic dynamic information of actions. Besides, we introduce a Static-Dynamic Rectifying module that incorporates the generalizable static information from CLIP to refine the predicted proposals, which improves proposal completeness in a generalizable manner. Our experiments show that our GAP achieves state-of-the-art performance on two challenging ZSTAL benchmarks, i.e., Thumos14 and ActivityNet1.3. Specifically, our model obtains significant performance improvement over previous works on the two benchmarks, i.e., +3.2% and +3.4% average mAP, respectively. The code is available at https://***/Run542968/GAP. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Zero-shot learning

来源：评论

学校读者我要写书评

暂无评论

Byzantine-robust distributed support vector machine

引用

Science China Mathematics 2025年第3期68卷 707-728页

作者： Xiaozhou Wang Weidong Liu Xiaojun Mao School of Statistics East China Normal UniversityShanghai 200062China Key Laboratory of Advanced Theory and Application in Statistics and Data Science(Ministry of Education) East China Normal UniversityShanghai 200062China School of Mathematical Sciences Shanghai Jiao Tong UniversityShanghai 200240China Key Laboratory of Artificial Intelligence(Ministry of Education) Shanghai Jiao Tong UniversityShanghai 200240China Key Laboratory of Scientific and Engineering Computing(Ministry of Education) Shanghai Jiao Tong UniversityShanghai 200240China

The development of information technology brings diversification of data sources and large-scale data sets and calls for the exploration of distributed learning algorithms. In distributed systems, some local machines may behave abnormally and send arbitrary information to the central machine(known as Byzantine failures), which can invalidate the distributed algorithms based on the assumption of faultless systems. This paper studies Byzantine-robust distributed algorithms for support vector machines(SVMs) in the context of binary classification. Despite a vast literature on Byzantine problems, much less is known about the theoretical properties of Byzantine-robust SVMs due to their unique challenges. In this paper, we propose two distributed gradient descent algorithms for SVMs. The median and trimmed mean operations in aggregation can effectively defend against Byzantine failures. Theoretically, we show the convergence of the proposed estimators and provide the statistical error rates. After a certain number of iterations, our estimators achieve near-optimal rates. Simulation studies and real data analysis are conducted to demonstrate the performance of the proposed Byzantine-robust distributed algorithms.

关键词： Byzantine robustness convergence distributed learning support vector machine

来源：评论

学校读者我要写书评

暂无评论

Perception-Enhanced Generative Transformer for Key Information Extraction from Documents 27th

Perception-Enhanced Generative Transformer for Key Informat...

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Zhao, Runbo Ou Yang, Jun Jie Gao, Chen Qin, Xugong Zeng, Gangyan Hu, Xiaoxu Zhang, Peng School of Cyber Science and Engineering Nanjing University of Science and Technology Nanjing China Viterbi School of Engineering University of Southern California Los Angeles United States National Computer Network Response Technical Team/Coordination Center Technology Beijing China Laboratory for Advanced Computing and Intelligence Engineering Wuxi China

ISBN: (纸本)9783031781186

Key information extraction (KIE) from scanned documents has attracted significant attention due to practical real-world applications. Despite impressive results achieved by incorporating multimodal information within the generative framework, existing methods fail to understand complex layouts and fuzzy semantics in document images. To settle these issues, we propose a perception-enhanced generative transformer (PEGT), which improves the model through fine-grained multimodal modeling and pre-training tasks tailored for the generative framework. Firstly, we introduce a pre-trained vision-language model to provide transferable knowledge for visual text perceptron. Then two auxiliary pre-training tasks including absolute position prediction (APP) and semantic relationship reasoning (SRR) are designed for the generative framework. APP learns to predict which grids the texts fall into, improving the model on utilization of the position information. SRR exploits prior information of semantic relationships, injecting the ability for better semantic discrimination into PEGT. Finally, well-designed prompts are leveraged to unleash the potential of PEGT for extracting key information from documents. Extensive experiments on several public datasets show that PEGT effectively generalizes over different types of documents. Especially, PEGT achieves state-of-the-art results in terms of F-measure, i.e., 97.47%, 98.04%, and 84.32% on the SROIE, CORD, and FUNSD datasets, demonstrating the superiority of the proposed method. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Generative adversarial networks

来源：评论

学校读者我要写书评

暂无评论

Rumor detection for emergency events via few-shot ensembled prompt learning

引用

Journal of Intelligent Information Systems 2025年 1-32页

作者： Su, Chen Zhou, Junkang Jiang, Zhentao Zhu, Shuwei Li, Chao Fang, Wei Lu, Heng-Yang School of Artificial Intelligence and Computer Science Jiangnan University Wuxi214122 China China Ship Scientific Research Center Wuxi214082 China The Laboratory for Advanced Computing and Intelligence Engineering Wuxi China

Rumors on social media can cause panic during emergency events. Unlike conventional rumor detection, it is more challenging to detect rumors about emergency events that have not happened in history, due to the shortage of relevant corpus. Therefore, we treat emergency events rumor detection as a few-shot learning problem. Recently, large language models (LLMs) such as ChatGPT have been widely considered in various NLP tasks. However, LLMs may face limitations due to non-real-time knowledge and unwillingness to provide direct answers. Prompt learning effectively leverages pre-trained language models (PLMs), and prompt tuning can inject the latest knowledge into PLMs with very few instances. Therefore, we propose a template-based Ensembled Prompt Tuning (EPT) model. We contribute an approach leveraging the knowledge in PLMs to generate label words for verbalizer construction. Furthermore, we treat few-shot rumor detection as an MLM problem and design two types of prompt templates for online posts and comments. An ensemble strategy is introduced to make the final prediction. Experimental results on three datasets demonstrate the effectiveness of the proposed EPT model, outperforming current SOTA on accuracy and F1-score. Ablation study has shown the necessity and effectiveness of the ensemble strategy. Besides, we make comparisons with prevalent LLMs under both few-shot and zero-shot settings, the results show the competitiveness of our EPT model. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2025.

关键词： Zero-shot learning

来源：评论

学校读者我要写书评

暂无评论

Code is not natural language: unlock the power of semantics-oriented graph representation for binary code similarity detection 24

Code is not natural language: unlock the power of semantics-...

引用

Proceedings of the 33rd USENIX Conference on Security Symposium

作者： Haojie He Xingwei Lin Ziang Weng Ruijie Zhao Shuitao Gan Libo Chen Yuede Ji Jiashui Wang Zhi Xue School of Electronic Information and Electrical Engineering Shanghai Jiao Tong University Ant Group Laboratory for Advanced Computing and Intelligence Engineering University of North Texas

ISBN: (纸本)9781939133441

Binary code similarity detection (BCSD) has garnered significant attention in recent years due to its crucial role in various binary code-related tasks, such as vulnerability search and software plagiarism detection. Currently, BCSD systems are typically based on either instruction streams or control flow graphs (CFGs). However, these approaches have limitations. Instruction stream-based approaches treat binary code as natural languages, overlooking well-defined semantic structures. CFG-based approaches exploit only the control flow structures, neglecting other essential aspects of code. Our key insight is that unlike natural languages, binary code has well-defined semantic structures, including intra-instruction structures, inter-instruction relations (e.g., def-use, branches), and implicit conventions (e.g. calling conventions). Motivated by that, we carefully examine the necessary relations and structures required to express the full semantics and expose them directly to the deep neural network through a novel semantics-oriented graph representation. Furthermore, we propose a lightweight multi-head softmax aggregator to effectively and efficiently fuse multiple aspects of the binary code. Extensive experiments show that our method significantly outperforms the state-of-the-art (e.g., in the x64-XC retrieval experiment with a pool size of 10000, our method achieves a recall score of 184%, 220%, and 153% over Trex, GMN, and jTrans, respectively).

关键词：

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：