检索结果-内蒙古大学图书馆

arXiv 2024年

作者： Zhang, Chaoran Zou, Lixin Luo, Dan Tang, Min Luo, Xiangyang Li, Zihao Li, Chenliang Wuhan University China Lehigh University United States Monash University Australia State Key Lab of Mathematical Engineering and Advanced Computing China

In recent years, Large Language Models (LLMs) have demonstrated remarkable capabilities across a wide array of text-centric tasks. However, their 'large' scale introduces significant computational and storage challenges, particularly in managing the key-value states of the transformer, which limits their wider applicability. Therefore, we propose to adaptively release resources from caches and rebuild the necessary key-value states. Particularly, we accomplish this by a lightweight controller module to approximate an ideal top-K sparse attention. This module retains the tokens with the highest top-K attention weights and simultaneously rebuilds the discarded but necessary tokens, which may become essential for future decoding. Comprehensive experiments in natural language generation and modeling reveal that our method is not only competitive with full attention in terms of performance but also achieves a significant throughput improvement of up to 221.8%. The code for replication is available on the https://***/WHUIR/ADORE. © 2024, CC BY.

关键词： Modeling languages

来源：评论

学校读者我要写书评

暂无评论

RDGFuzz: A directed greybox fuzzing optimization method based on Rich-Branch nodes 22

RDGFuzz: A directed greybox fuzzing optimization method base...

引用

Proceedings of the 2022 6th International Conference on Electronic Information Technology and Computer engineering

作者： Zejun Wu Li Lu Qiong Jia Zhihao Chen State Key Laboratory of Mathematical Engineering and advanced Computing China National Local Joint Engineering Laboratory of Network Space Security Technology China Beijing Institute of Computer Technology and Application China

ISBN: (纸本)9781450397148

Directed fuzzing technology is one of the key technologies to quickly reach a specific location of software, and to conduct targeted testing or bug recurrence. However, directed fuzzing technology has some problems, such as unreasonable seed energy allocation, low code coverage and incomplete testing. To solve the above problems, this paper proposes an optimization method of directed fuzzing based on Rich-Branch nodes. In this method, the concept of Rich-Branch nodes is defined and the algorithm of extracting Rich-Branch nodes is given. The optimization method collects the coverage information of the target program in the running process, calculates the weights of covered functions and nodes in real time by combining CG and CFG of the target program, and generates a list of Rich-Branch nodes. According to the weights of Rich-Branch nodes, the seed energy allocation algorithm of AFLGo is optimized and improved. Compared with AFLGo, this optimization method improves the average code coverage of each targeted point by 56.79%, and has the same target reaching ability as AFLGo.

关键词： Rich-Branch nodes Directed fuzzing Code coverage rate Vulnerability discovery

来源：评论

学校读者我要写书评

暂无评论

Taking Myopic Best Response Against the Hedge Algorithm 42

Taking Myopic Best Response Against the Hedge Algorithm

引用

42nd Chinese Control Conference, CCC 2023

作者： Guo, Xinxiang Mu, Yifen School of Mathematical Sciences University of Chinese Academy of Sciences Beijing China Academy of Mathematics and Systems Science The State Key Laboratory of Scientific and Engineering Computing Chinese Academy of Sciences Beijing China Academy of Mathematics and Systems Science The Key Laboratory of Systems and Control Chinese Academy of Sciences Beijing China

ISBN: (纸本)9789887581543

With the rapid development of artificial intelligence (AI), the game between the human and machine/AI becomes more and more common. The related theoretical analysis becomes significant and necessary, which however is still rare. This problem involves evolution analysis and optimal control of dynamic game systems driven by learning algorithms. In this paper, we consider a finitely repeated two-player zero-sum game. We study the dynamic evolution process of the game system, in which one player employs the Hedge algorithm and the other player takes the myopic best response. By the updating formula of the Hedge algorithm, we define a quantity called the state and construct the State Transition Triangle Graph (STTG). Then, we prove that the game system is periodic after o(T) stages. © 2023 Technical Committee on Control Theory, Chinese Association of Automation.

关键词： Artificial intelligence

来源：评论

学校读者我要写书评

暂无评论

HHL Analysis and Simulation Verification Based on Origin Quantum Platform 4

HHL Analysis and Simulation Verification Based on Origin Qua...

引用

2021 4th International Conference on Mechatronics and Computer Technology engineering, MCTE 2021

作者： Liu, Xiaonan Jing, Lina Han, Lin Gao, Jie State Key Laboratory of Mathematical Engineering and Advanced Computing Zhengzhou450001 China Henan Supercomputing Center Zhengzhou University Zhengzhou450001 China

Solving large-scale linear equations is of great significance in many engineering fields, such as weather forecasting and bioengineering. The classical computer solves the linear equations, no matter adopting the elimination method or Kramer's rule, the time required for solving is in a polynomial relationship with the scale of the equation system. With the advent of the era of big data, the integration of transistors is getting higher and higher. When the size of transistors is close to the order of electron diameter, quantum tunneling will occur, and Moore's Law will not be followed. Therefore, the traditional computing model will not be able to meet the demand. In this paper, through an in-depth study of the classic HHL algorithm, a small-scale quantum circuit model is proposed to solve a 2×2 linear equations, and the circuit diagram and programming are used to simulate and verify on the Origin Quantum Platform. The fidelity under different parameter values reaches more than 90%. For the case where the matrix to be solved is a sparse matrix, the quantum algorithm has an exponential speed improvement over the best known classical algorithm. © 2021 Institute of Physics Publishing. All rights reserved.

关键词： Linear equations

来源：评论

学校读者我要写书评

暂无评论

A Cross-Platform Instant Messaging User Association Method Based on Spatio-temporal Trajectory 8th

A Cross-Platform Instant Messaging User Association Method B...

引用

8th International Conference on Artificial Intelligence and Security , ICAIS 2022

作者： Zhou, Pei Luo, Xiangyang Du, Shaoyong Li, Lingling Yang, Yang Liu, Fenlin School of Cyber Science and Engineering Zhengzhou University Zhengzhou450000 China State Key Laboratory of Mathematical Engineering and Advanced Computing Zhengzhou450000 China State Key Laboratory of Information Security Institute of Information Engineering Chinese Academy of Sciences Beijing100093 China School of Intelligent Engineering Zhengzhou University of Aeronautics Zhengzhou450000 China School of Computing and Information Systems Singapore Management University Singapore188065 Singapore

ISBN: (纸本)9783031067600

The current research on cross-platform instant messaging user association is mainly divided into two categories: based on user attributes and based on user behavior. Methods based on user attributes mainly identify users through multiple attributes such as user name and multi-platform user information association based on cell phone number, but user association is not possible when multi-platform user information is inconsistent and users do not grant their address book permissions. Methods based on user behavior mainly calculate the similarity between user trajectories features such as geographic location frequency and co-occurrence, but this method lacks the user’s information, which leads to the inability to fully excavate the sequential features of the trajectory and affects the accuracy of trajectory matching. In order to solve this problem, this paper proposes a cross-platform instant messaging user association algorithm based on temporal trajectories (CPTrajst). We firstly place probes in the area where the target may appear so as to obtain user information, gets user trajectory, then processes the trajectory and performs two trajectory matches, finally associate users of different platforms whose trajectories match, thus increasing the accuracy and reliability of cross-platform instant messaging user association. We conducts specific experiments for users of WeChat and Momo, the most commonly used instant messengers in China. The results show that we can achieve reliable association for these two types of instant messaging users and the user association accuracy can reach 99.5%, which is better than the existing user association algorithms based on trajectory matching. © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.

关键词： Trajectories

来源：评论

学校读者我要写书评

暂无评论

Designing Uncorrelated Address Constrain for DNA Storage by DMVO Algorithm

引用

IEEE/ACM Transactions on Computational Biology and Bioinformatics 2022年第2期19卷 866-877页

作者： Cao, Ben Ii, Xue Zhang, Xiaokang Wang, Bin Zhang, Qiang Wei, Xiaopeng Key Laboratory of Advanced Design and Intelligent Computing Ministry of Education Ministry of Education School of Software Engineering Dalian University Dalian116622 China School of Computer Science and Technology Dalian University of Technology Dalian116024 China

At present, huge amounts of data are being produced every second, a situation that will gradually overwhelm current storage technology. DNA is a storage medium that features high storage density and long-term stability and is now considered to be a feasible storage solution. Errors are easily made during the sequencing and synthesis of DNA, however. In order to reduce the error rate, novel uncorrelated address constrain are reported, and a Damping Multi-Verse Optimizer (DMVO)algorithm is proposed to construct a set of DNA coding, which is used as the non-payload. The DMVO algorithm exchanges objects through black/white holes in order to achieve a stable state and adds damping factors as disturbances. Compared with previous work, the coding set obtained by the DMVO algorithm is larger in size and of higher quality. The results of this study reveal that the size of the DNA storage coding set obtained by the DMVO algorithm increased by 4-16 percent, and the variance of the melting temperature decreased by 3-18 percent. © 2004-2012 IEEE.

关键词： Damping

来源：评论

学校读者我要写书评

暂无评论

Image Electronic Evidence Screening Based on Improved SSD 6

Image Electronic Evidence Screening Based on Improved SSD

引用

6th IEEE International Conference on Intelligent computing and Signal Processing, ICSP 2021

作者： Liu, Yafei Jiang, Liehui Liu, Tieming Zhang, Youwei Zhengzhou University Zhongyuan Network Security Research Institute Zhengzhou China State Key Laboratory of Mathematical Engineering and Advanced Computing Zhengzhou China

ISBN: (纸本)9780738143705

With the development of information technology, electronic evidence plays an increasingly important role in the judicial trial. In Judicial Forensics, it is difficult to extract effective electronic evidence accurately and quickly from a large number of electronic data. Traditional means generally take artificial identification to collect, which takes a long time and is not efficient. Using SSD target detection and recognition algorithm instead of traditional means can effectively reduce the time of screening evidence, but the basic SSD neural network is prone to miss detection for small targets. To solve the above problems, an improved SSD-based image electronic evidence screening method is proposed. This method optimizes the SSD neural network adaptively, introduces the attention mechanism module in the shallow convolution layer of the network to improve the representation ability of the feature map, and fuses the image features obtained from different convolution layers with multi-scale features to increase the shallow feature information. The experimental data are used to test the improved algorithm and analyze the experimental results. It is found that compared with the original SSD neural network algorithm, the detection mean average precision of the improved algorithm is increased by 4.7%, reaching 84.3%, which shows the feasibility and effectiveness of the improved algorithm. © 2021 IEEE.

关键词： Multilayer neural networks

来源：评论

学校读者我要写书评

暂无评论

Lightweight DCT-Like Domain Forensics Model for Adversarial Example 19th

Lightweight DCT-Like Domain Forensics Model for Adversarial ...

引用

19th International Workshop on Digital Forensics and Watermarking, IWDW 2020

作者： Zhao, Junjie Wang, Jinwei Nanjing University of Information Science and Technology Jiangsu210044 China State Key Laboratory of Mathematical Engineering and Advanced Computing Henan450001 China

ISBN: (纸本)9783030694487

Since the emergence of adversarial examples brings great security threat to deep neural network which is widely used in various fields, their forensics become very important. In this paper, a lightweight model for the forensics of adversarial example based on DCT-like domain is proposed. The DCT-like layer realizes the block conversion of data from the spatial domain to the frequency domain. Together with the color space transformation layer and the residual layer, the DCT-like layer realizes the simulation of JPEG quantization error. The feature statistical layer is used to obtain the statistical feature values of the feature map output by the frequency-division convolution, and at the same time, it also contains learnable hyperparameters. The group BN strategy ensures the effectiveness of the DCT-like layer and the feature statistical layer and promotes the accuracy of forensics. Experiments show that the proposed model not only reaches the highest accuracy we know, but also it only needs to train for one epoch to get a high-performance. © 2021, Springer Nature Switzerland AG.

关键词： Deep neural networks

来源：评论

学校读者我要写书评

暂无评论

Attentive Hawkes Process Application for Sequential Recommendation 28th

Attentive Hawkes Process Application for Sequential Recomme...

引用

28th International Conference on Database Systems for advanced Applications, DASFAA 2023

作者： Yu, Shuodian Ma, Li Gao, Xiaofeng Guo, Jianxiong Chen, Guihai Shanghai Key Laboratory of Scalable Computing and Systems Department of Computer Science and Engineering Shanghai Jiao Tong University Shanghai China Advanced Institute of Natural Sciences Beijing Normal University Zhuhai519087 China

ISBN: (纸本)9783031306716

Sequential recommendation systems which aim to predict the users’ future clicking items are essential in both research and industry services. Based on the contextual characteristics of users, a large number of existing methods exploit abundant information from user behavior sequences and gain users’ interests through expressive sequential models like recurrent neural networks and self-attention mechanism to predict the next item. However, the temporal generative process of behavior sequences, which is crucial to the complex relationships and distributions of users’ dynamic preferences, is rarely considered. In this paper, we propose HawRec (Hawkes process based sequential recommendation), which is a new representation learning approach to model the interacted sequences of users from a temporal point process perspective. In particular, we leverage temporal self-attention mechanism to make an analogy to the self-exciting mechanism of Hawkes process. We bring expressive superiority of neural network into the Hawkes process with non-parametric conditional intensity kernel, resulting in more accurate recommendation and inheriting robustness from the temporal point process. Furthermore, extensive experiments conducted on four public benchmarks demonstrate that our HawRec outperforms various baselines of state-of-the-art sequential recommendation. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

关键词： Recurrent neural networks

来源：评论

学校读者我要写书评

暂无评论

APU-D* Lite: Attack Planning under Uncertainty Based on D* Lite

引用

Computers, Materials & Continua 2020年第11期65卷 1795-1807页

作者： Tairan Hu Tianyang Zhou Yichao Zang Qingxian Wang Hang Li State Key Laboratory of Mathematical Engineering and Advanced Computing Zhengzhou450001China School of Electrical Engineering and Telecommunications UNSW KensingtonNSW 2033Australia

With serious cybersecurity situations and frequent network attacks,the demands for automated pentests continue to increase,and the key issue lies in attack *** the limited viewpoint of the attacker,attack planning under uncertainty is more suitable and practical for pentesting than is the traditional planning approach,but it also poses some *** address the efficiency problem in uncertainty planning,we propose the APU-D*Lite algorithm in this ***,the pentest framework is mapped to the planning problem with the Planning Domain Definition Language(PDDL).Next,we develop the pentest information graph to organize network information and assess relevant exploitation actions,which helps to simplify the problem ***,the APU-D*Lite algorithm is introduced based on the idea of incremental heuristic *** method plans for both hosts and actions,which meets the requirements of *** the pentest information graph as the input,the output is an alternating host and action *** experiments,we use the attack success rate to represent the uncertainty level of the *** result shows that APU-D*Lite displays better reliability and efficiency than classical planning algorithms at different attack success rates.

关键词： Attack planning under uncertainty automated pentest APU-D*Lite algorithm incremental heuristic search

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：