检索结果-内蒙古大学图书馆

SSRN 2024年

作者： Hu, Wentao Jiang, Dawei Wu, Sai Chen, Ke Chen, Gang Zhejiang Police College Hangzhou China The Key Laboratory of Big Data Intelligent Computing of Zhejiang Province Zhejiang University China

Experts in fraud detection and data security commonly encounter hurdles in the automatic extraction of constraints from relational databases, a vital step for the identification of anomalous data. The aim is to unearth intrinsic database constraints, so that any tuple infringements can be marked as anomalies. The crux of IceBerg is the KD-VAE, an innovative generative network. By integrating \underline{K}ernel \underline{D}ensity estimation with \underline{V}ariational \underline{A}uto\underline{E}ncoder, KD-VAE addresses the challenge of capturing complex data distributions, a critical aspect for effective anomaly detection. KD-VAE synergizes Kernel Density estimation with Variational Autoencoders, effectively addressing the intricacies of modeling complex data distributions, which is essential for proficient anomaly detection. The distinct combination of kernel density estimation with variational autoencoders in KD-VAE markedly improves the modeling of normal tuple distributions, thus facilitating enhanced anomaly detection. Anomalous data are pinpointed by evaluating the congruence of a tuple with the normal distributions, and anomalies are interpreted by contrasting detected aberrations against their normal analogs. Principal improvements encompass methodological refinements for generating interpretable rules, the addition of Jaccard and NMI to the evaluation metrics, and extensive empirical analyses over 13 real-world datasets. Furthermore, comparative analyses of temporal performance and visualizations for rule interpretation are introduced, yielding deeper insights into anomaly detection, and significantly surpassing our prior work's benchmarks. These progressions confirm IceBerg's exceptional capacity for modeling intricate, high-dimensional data distributions, establishing it as a formidable asset in the analysis of large-scale datasets. © 2024, The Authors. All rights reserved.

关键词： Anomaly detection

来源：评论

学校读者我要写书评

暂无评论

Training on the Benchmark Is Not All You Need 39

Training on the Benchmark Is Not All You Need

引用

39th Annual AAAI Conference on Artificial Intelligence, AAAI 2025

作者： Ni, Shiwen Kong, Xiangtao Li, Chengming Hu, Xiping Xu, Ruifeng Zhu, Jia Yang, Min Shenzhen Key Laboratory for High Performance Data Mining Shenzhen Institutes of Advanced Technology CAS China Key Laboratory of Intelligent Education Technology Application of Zhejiang Province Zhejiang Normal University China University of Science and Technology of China China Shenzhen MSU-BIT University China Harbin Institute of Technology Shenzhen China

ISBN: (纸本)157735897X

The success of Large Language Models (LLMs) relies heavily on the huge amount of pre-training data learned in the pre-training phase. The opacity of the pre-training process and the training data causes the results of many benchmark tests to become unreliable. If any model has been trained on a benchmark test set, it can seriously hinder the health of the field. In order to automate and efficiently test the capabilities of large language models, numerous mainstream benchmarks adopt a multiple-choice format. As the swapping of the contents of multiple-choice options does not affect the meaning of the question itself, we propose a simple and effective data leakage detection method based on this property. Specifically, we shuffle the contents of the options in the data to generate the corresponding derived data sets, and then detect data leakage based on the model's log probability distribution over the derived data sets. If there is a maximum and outlier in the set of log probabilities, it indicates that the data is leaked. Our method is able to work under gray-box conditions without access to model training data or weights, effectively identifying data leakage from benchmark test sets in model pre-training data, including both normal scenarios and complex scenarios where options may have been shuffled intentionally or unintentionally. Through experiments based on two LLMs and benchmark designs, we demonstrate the effectiveness of our method. In addition, we evaluate the degree of data leakage of 35 mainstream open-source LLMs on four benchmark datasets and give a ranking of the leaked LLMs for each benchmark, and we find that the Qwen family of LLMs has the highest degree of data leakage. Copyright © 2025, Association for the Advancement of Artificial Intelligence (***). All rights reserved.

关键词： Normal distribution

来源：评论

学校读者我要写书评

暂无评论

Mixup data Augmentation and Channel-Spatial Attention Mechanisms for Facial Emotion Recognition 14

Mixup Data Augmentation and Channel-Spatial Attention Mechan...

引用

14th International Conference on Information Technology in Medicine and Education, ITME 2024

作者： Yu, Dewei Kong, Chuijian Tang, Xianpiao Xia, Daoxun Guizhou Normal University School of Big Data and Computer Science Guiyang China Guizhou Normal University Key Laboratory of Information and Computing Science Guizhou Province Guiyang China Guizhou Normal Universitya Engineering Laboratory for Applied Technology of Big Data in Education Guiyang China

ISBN: (纸本)9798350356670

With the rapid advancement of computer vision, facial expression recognition has made significant progress in areas like human-computer interaction, mental health, driver assistance, and education. However, traditional facial expression recognition techniques exhibit poor robustness under complex backgrounds and varying lighting conditions, making them susceptible to environmental noise and challenging to accurately capture and classify complex emotional states. Additionally, these techniques often lack in data augmentation and model generalization capabilities, leading to overfitting on training samples and insufficient recognition of unseen samples. To address these limitations, this paper proposes a novel convolutional neural network model - HSCNet (A Hybrid data Augmentation and Channel-Spatial Attention Mechanisms Neural Network). The HSCNet model first employs the Mixup data augmentation technique to generate new training samples, then utilizes the Res2Net50 backbone network combined with the CBAM attention mechanism and average-max pooling techniques to extract significant features while suppressing redundant ones. This approach not only enhances feature extraction accuracy but also significantly improves the model's generalization ability and robustness in facial expression recognition tasks. Finally, tested on the public dataset FERPlus, which includes eight types of emotions, the HSCNet model achieved an accuracy rate of 89.72%, demonstrating excellent performance and robustness. © 2024 IEEE.

关键词： attention mechanism average-max pooling Computer vision data augmentation facial expression recognition HSCNet

来源：评论

学校读者我要写书评

暂无评论

Hierarchical Context Pruning: Optimizing Real-World Code Completion with Repository-Level Pretrained Code LLMs 39

Hierarchical Context Pruning: Optimizing Real-World Code Com...

引用

39th Annual AAAI Conference on Artificial Intelligence, AAAI 2025

作者： Zhang, Lei Li, Yunshui Li, Jiaming Xia, Xiaobo Yang, Jiaxi Luo, Run Wang, Minzheng Chen, Longze Liu, Junhao Qu, Qiang Yang, Min Shenzhen Key Laboratory for High Performance Data Mining Shenzhen Institutes of Advanced Technology Chinese Academy of Sciences China University of Chinese Academy of Sciences China Key Laboratory of Intelligent Education Technology Application of Zhejiang Province Zhejiang Normal University China School of Computing National University of Singapore Singapore University of Science and Technology of China China University of California Irvine United States MAIS Institute of Automation Chinese Academy of Sciences China

ISBN: (纸本)157735897X

Some of the latest released Code Large Language Models (Code LLMs) have been trained on repository-level code data, enabling them to perceive repository structures and utilize cross-file code information. This capability allows us to directly concatenate the content of repository code files in prompts to achieve repository-level code completion. However, in real development scenarios, directly concatenating all code repository files in a prompt can easily exceed the context window of Code LLMs, leading to a significant decline in completion performance. Additionally, overly long prompts can increase completion latency, negatively impacting the user experience. In this study, we conducted extensive experiments, including completion error analysis, topology dependency analysis, and cross-file content analysis, to investigate the factors affecting repository-level code completion. Based on the conclusions drawn from these preliminary experiments, we proposed a strategy called Hierarchical Context Pruning (HCP) to construct high-quality completion prompts. We applied the HCP to six Code LLMs and evaluated them on the CrossCodeEval dataset. The experimental results showed that, compared to previous methods, the prompts constructed using our HCP strategy achieved higher completion accuracy on five out of six Code LLMs. Additionally, the HCP managed to keep the prompt length around 8k tokens (whereas the full repository code is approximately 50k tokens), significantly improving completion throughput. Our code and data will be publicly available. © 2025, Association for the Advancement of Artificial Intelligence (***). All rights reserved.

关键词： High level languages

来源：评论

学校读者我要写书评

暂无评论

Dynamic Neural Radiance Field From Defocused Monocular Video

arXiv

引用

arXiv 2024年

作者： Luo, Xianrui Sun, Huiqiang Peng, Juewen Cao, Zhiguo Key Laboratory of Image Processing and Intelligent Control Ministry of Education School of AIA Huazhong University of Science and Technology China College of Computing and Data Science Nanyang Technological University Singapore

Dynamic Neural Radiance Field (NeRF) from monocular videos has recently been explored for space-time novel view synthesis and achieved excellent results. However, defocus blur caused by depth variation often occurs in video capture, compromising the quality of dynamic reconstruction because the lack of sharp details interferes with modeling temporal consistency between input views. To tackle this issue, we propose D2RF, the first dynamic NeRF method designed to restore sharp novel views from defocused monocular videos. We introduce layered Depth-of-Field (DoF) volume rendering to model the defocus blur and reconstruct a sharp NeRF supervised by defocused views. The blur model is inspired by the connection between DoF rendering and volume rendering. The opacity in volume rendering aligns with the layer visibility in DoF rendering. To execute the blurring, we modify the layered blur kernel to the ray-based kernel and employ an optimized sparse kernel to gather the input rays efficiently and render the optimized rays with our layered DoF volume rendering. We synthesize a dataset with defocused dynamic scenes for our task, and extensive experiments on our dataset show that our method outperforms existing approaches in synthesizing all-in-focus novel views from defocus blur while maintaining spatial-temporal consistency in the scene. Copyright © 2024, The Authors. All rights reserved.

关键词： Volume rendering

来源：评论

学校读者我要写书评

暂无评论

Streaming Media Load Balancing with Improved Genetic Algorithm

Streaming Media Load Balancing with Improved Genetic Algorit...

引用

Chinese Control and Decision Conference, CCDC

作者： Yang Zhang Xing Yang Yuan Xu Yanlin He Mingqing Zhang Qunxiong Zhu College of Information Science and Technology Beijing University of Chemical Technology Beijing China Key Laboratory of Data Science and Intelligent Computing Institute of International Innovation Beihang University Hangzhou China

ISBN: (数字)9798350387780

ISBN: (纸本)9798350387797

The load balancing algorithm of streaming media server is facing unique challenges. Different from the traditional Web server, the streaming media server needs to consider the characteristics of real-time, bandwidth occupation and multimedia data transmission. Therefore, the traditional load balancing algorithm is not necessarily suitable for streaming media servers, and needs special design and research. Aiming at the network requests of streaming media server clusters, a multi-dimensional matrix evaluation method based on streaming media characteristics is proposed, the genetic algorithm is improved, and it is applied to the dynamic load balancing algorithm scheduling scenario, and analyzed and simulated. Experimental results show that the algorithm can effectively reduce the average execution time of all requests, speed up the average response time and improve throughput.

关键词： Heuristic algorithms Clustering algorithms Streaming media Throughput Load management Dynamic scheduling Web servers

来源：评论

学校读者我要写书评

暂无评论

FR2Seg: Continual Segmentation Across Multiple Sites via Fourier Style Replay and Adaptive Consistency Regularization 39

FR2Seg: Continual Segmentation Across Multiple Sites via Fou...

引用

39th Annual AAAI Conference on Artificial Intelligence, AAAI 2025

作者： Xu, Cheng Zhang, Weiwen Zhang, Hongrui Xu, Xuemiao Zhang, Huaidong Zou, Jing Qin, Jing South China University of Technology China The Hong Kong Polytechnic University Hong Kong Guangdong Engineering Center for Large Model and GenAI Technology China State Key Laboratory of Subtropical Building Science China Ministry of Education Key Laboratory of Big Data and Intelligent Robot China Guangdong Provincial Key Lab of Computational Intelligence and Cyberspace Information China

ISBN: (纸本)157735897X

In clinical imaging, medical segmentation networks typically require continually adapting to new data from multiple sites over time, as aggregating all data for learning at once can be impractical due to storage limitations and privacy concerns. However, existing methods basically overlook domain-specific characteristics and fall short of adequately capturing domain-invariant knowledge during continual learning, leading to undesired catastrophic forgetting of previous sites and inferior generalization to new sites. To tackle this issue, this paper introduces FR2Seg, to sufficiently exploit both domain-specific and domain-invariant knowledge for efficient continual learning with the aid of low-frequency cues. For the former aspect, we propose a Fourier style replay module to synthesize pseudo images with old-site styles for data augmentation during new-site training, effectively preventing catastrophic forgetting without sacrificing data privacy. For the latter, we present a Fourier adaptive consistency regularization to identify and constrain the optimization of domain-invariant parameters with explicit awareness of knowledge transferability across sites, ensuring excellent generalizability to new sites. Experimental results on two public datasets confirm our method's superiority over existing state-of-the-art continual learning methods. Copyright © 2025, Association for the Advancement of Artificial Intelligence (***). All rights reserved.

关键词： Image segmentation

来源：评论

学校读者我要写书评

暂无评论

PP-Grading: Hierarchical Authorization using Polarization Pruning 14

PP-Grading: Hierarchical Authorization using Polarization Pr...

引用

14th International Conference on Information Technology in Medicine and Education, ITME 2024

作者： Zhu, Yujia Wang, Ruoxi Xia, Daoxun Guizhou Normal University School of Big Data and Computer Science Guiyang China Guizhou Normal University Key Laboratory of Information and Computing Science Guizhou Province Guiyang China Guizhou Normal University Engineering Laboratory for Applied Technology of Big Data in Education Guiyang China

ISBN: (纸本)9798350356670

With the popularity of neural networks, more and more service providers are providing us with convenient neural network services. In this situation, it is necessary for service providers to grant hierarchical authorization of the networks to meet the service requirements of different users. Pruning technology, as a mainstream method for modifying neural network structures, can also be applied in the hierarchical authorization of neural networks. This article uses the method of polarization pruning to decentralize and grade neural networks. Firstly, we extract the scale factors of the batch normalization(BN) layer as a measure of the importance of each channel. Then, sparse methods are used to screen for scale factors. In order to make the selected scale factors have more obvious and reasonable truncation intervals, we also use polarization methods to induce clustering of scale factors. According to different grading requirements, we prune the channels corresponding to different quantities of scale factors. The experiment proves that our method has good grading performance on different datasets and different neural networks. Moreover, good grading results can be achieved by pruning only a small portion of the channels. © 2024 IEEE.

关键词： grading hierarchical authorization pruning

来源：评论

学校读者我要写书评

暂无评论

TWO-GRID ALGORITHM OF H^(1)-GALERKIN MIXED FINITE ELEMENT METHODS FOR SEMILINEAR PARABOLIC INTEGRO-DIFFERENTIAL EQUATIONS

引用

Journal of Computational Mathematics 2022年第5期40卷 667-685页

作者： Tianliang Hou Chunmei Liu Chunlei Dai Luoping Chen Yin Yang School of Mathematics and Statistics Beihua UniversityJilin 132013China College of Science Hunan University of Science and EngineeringYongzhou 425199China School of Mathematics Southwest Jiaotong UniversityChengdu 611756China Hunan Key Laboratory for Computation and Simulation in Science and Engineering Key Laboratory of Intelligent Computing&Information Processing of Ministry of EducationSchool of Mathematics and Computational ScienceXiangtan UniversityXiangtan 411105China

In this paper,we present a two-grid discretization scheme for semilinear parabolic integro-differential equations by H1-Galerkin mixed finite element *** use the lowest order Raviart-Thomas mixed finite elements and continuous linear finite element for spatial discretization,and backward Euler scheme for temporal ***,a priori error estimates and some superclose properties are ***,a two-grid scheme is presented and its convergence is *** the proposed two-grid scheme,the solution of the nonlinear system on a fine grid is reduced to the solution of the nonlinear system on a much coarser grid and the solution of two symmetric and positive definite linear algebraic equations on the fine grid and the resulting solution still maintains optimal ***,a numerical experiment is implemented to verify theoretical results of the proposed *** theoretical and numerical results show that the two-grid method achieves the same convergence property as the one-grid method with the choice h=H^(2).

关键词： Semilinear parabolic integro-differential equations H^(1)-Galerkin mixed finite element method A priori error estimates Two-grid Superclose.

来源：评论

学校读者我要写书评

暂无评论

Invisible Watermark and Hypothesis Testing for Multi-Ownership Verification Framework of Deep Neural Network 14

Invisible Watermark and Hypothesis Testing for Multi-Ownersh...

引用

14th International Conference on Information Technology in Medicine and Education, ITME 2024

作者： Wang, Ruoxi Zhu, Yujia Xia, Daoxun Guizhou Normal University School of Big Data and Computer Science Guiyang China Guizhou Normal University Key Laboratory of Information and Computing Science Guizhou Province Guiyang China Guizhou Normal University Engineering Laboratory for Applied Technology of Big Data in Education Guiyang China

ISBN: (纸本)9798350356670

With the rapid development of information technology, artificial intelligence models have become the core asset of tech giants, highlighting the importance of copyright protection. Black-box watermarking has gained attention due to its ability to verify the legality of models remotely without relying on internal model information. The challenges lie in embedding and verifying watermarks without accessing internal model parameters, while ensuring the robustness and stealth of the watermark. This paper focuses on discussing effective ways to protect the copyright of black-box watermarking models and detect and prevent unauthorized user behavior that infringes on model ownership. Therefore, this paper proposes a multi-ownership verification framework based on invisible watermark and hypothesis testing, which mainly consists of two parts: watermark generation and copyright verification. The watermark generation uses a dynamic watermarking method to generate invisible backdoor watermark and verifies copyright through hypothesis testing. This framework has high assurance and effectiveness under various models of CIFAR-10 and MNIST datasets and shows strong robustness in VGG and Resnet model pruning attacks. This research provides new ideas and methods for the protection of model copyrights in deep neural networks. © 2024 IEEE.

关键词： digital watermarking hypothesis testing invisible watermark neural networks copyright protection ownership verification

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：