检索结果-南通市图书馆

Red Alarm for Pre-trained Models:Universal Vulnerability to Neuron-level Backdoor Attacks

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Machine Intelligence Research 2023年第2期20卷 180-193页

作者： Zhengyan Zhang Guangxuan Xiao Yongwei Li Tian Lv Fanchao Qi Zhiyuan Liu Yasheng Wang Xin Jiang Maosong sun Department of Computer science and Technology Tsinghua UniversityBeijing 100084China Institute for Artificial Intelligence Tsinghua UniversityBeijing 100084China Beijing National Research Center for Information science and Technology Beijing 100084China Huawei noah′s ark laboratory Hong Kong 999077China

The pre-training-then-fine-tuning paradigm has been widely used in deep *** to the huge computation cost for pre-training,practitioners usually download pre-trained models from the Internet and fine-tune them on downstream datasets,while the downloaded models may suffer backdoor *** from previous attacks aiming at a target task,we show that a backdoored pre-trained model can behave maliciously in various downstream tasks without foreknowing task *** can restrict the output representations(the values of output neurons)of trigger-embedded samples to arbitrary predefined values through additional training,namely neuron-level backdoor attack(NeuBA).since fine-tuning has little effect on model parameters,the fine-tuned model will retain the backdoor functionality and predict a specific label for the samples embedded with the same *** provoke multiple labels in a specific task,attackers can introduce several triggers with predefined contrastive *** the experiments of both natural language processing(NLP)and computer vision(CV),we show that NeuBA can well control the predictions for trigger-embedded instances with different trigger *** findings sound a red alarm for the wide use of pre-trained ***,we apply several defense methods to NeuBA and find that model pruning is a promising technique to resist NeuBA by omitting backdoored neurons.

关键词： Pre-trained language models backdoor attacks transformers natural language processing(NLP) computer vision(CV)

Deep learning for natural language processing:advantages and challenges

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

National science Review 2018年第1期5卷 24-26页

作者： Hang Li noah’s ark Lab Huawei Technologies

INTRODUCTION Deep learning refers to machine learning technologies for learning and utilizing’deep’artificial neural networks,such as deep neural networks(DNN),convolutional neural networks(CNN)and recurrent neural networks(RNN).Recently,deep leaminghasbeen successfiully applied to natural language processing and significant progress has been *** paper summarizes the recent advancement of deep learning for natural language processing and discusses its advantages and challenges.

关键词： Deep learning for natural language processing:advantages and challenges

Protein structure Prediction:Challenges,Advances,and the shift of Research Paradigms

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Genomics, Proteomics & Bioinformatics 2023年第5期21卷 913-925页

作者： Bin Huang Lupeng Kong Chao Wang Fusong Ju Qi Zhang Jianwei Zhu Tiansu Gong Haicang Zhang Chungong Yu Wei-Mou Zheng Dongbo Bu Key laboratory of Intelligent Information Processing Institute of Computing TechnologyChinese Academy of SciencesBeijing 100190China University of Chinese Academy of sciences Beijing 100049China Changping laboratory Beijing 102206China Microsoft Research AI4science Beijing 100080China Huawei noah’s ark Lab Wuhan 430206China Zhongke Big Data Academy Zhengzhou 450046China Institute of Theoretical Physics Chinese Academy of SciencesBeijing 100190China

Protein structure prediction is an interdisciplinary research topic that has attracted researchers from multiple fields,including biochemistry,medicine,physics,mathematics,and computer *** researchers adopt various research paradigms to attack the same structure prediction problem:biochemists and physicists attempt to reveal the principles governing protein folding;mathematicians,especially statisticians,usually start from assuming a probability distribution of protein structures given a target sequence and then find the most likely structure,while computer scientists formulate protein structure prediction as an optimization problem-finding the structural conformation with the lowest energy or minimizing the difference between predicted structure and native *** research paradigms fall into the two statistical modeling cultures proposed by Leo Breiman,namely,data modeling and algorithmic ***,we have also witnessed the great success of deep learning in protein structure *** this review,we present a survey of the efforts for protein structure *** compare the research paradigms adopted by researchers from different fields,with an emphasis on the shift of research paradigms in the era of deep *** short,the algorithmic modeling techniques,especially deep neural networks,have considerably improved the accuracy of protein structure prediction;however,theories interpreting the neural networks and knowledge on protein folding are still highly desired.

关键词： Protein folding Protein structure prediction Deep learning Transformer Language model

BI-GreenNet:Learning Green’s Functions by Boundary Integral Network

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Communications in Mathematics and statistics 2023年第1期11卷 103-129页

作者： Guochang Lin Fukai Chen Pipi Hu Xiang Chen Junqing Chen Jun Wang Zuoqiang shi Yau Mathematical sciences Center Tsinghua UniversityBeijing 100084People’s Republic of China Department of Mathematical sciences Tsinghua UniversityBeijing 100084People’s Republic of China Yanqi Lake Beijing Institute of Mathematical sciences and Applications Beijing 101408People’s Republic of China noah’s ark Lab HuaweiNo.3 Xinxi RoadBeijing 100085People’s Republic of China University College London London WC1E 6EAUK

Green’s function plays a significant role in both theoretical analysis and numerical computing of partial differential equations(PDEs).However,in most cases,Green’s function is difficult to *** troubles arise in the following ***,compared with the original PDE,the dimension of Green’s function is doubled,making it impossible to be handled by traditional mesh-based ***,Green’s function usually contains singularities which increase in the difficulty to get a good ***,the computational domain may be very complex or even *** override these problems,we develop a new framework for computing Green’s function leveraging the fundamental solution,boundary integral method and neural networkswith reasonably high accuracy in this *** focus on Green’s function of Poisson and Helmholtz equations in bounded domains,unbounded *** also consider Poisson equation and Helmholtz equations in domains with *** numerical experiments illustrate the efficiency and accuracy of our method for solving Green’s *** addition,Green’s function provides the operator from the source term and boundary condition to *** applyGreen’s function to solve PDEs with different sources,and obtain reasonably high-precision solutions,which shows the good generalization ability of our ***,the requirements for explicit fundamental solutions to remove the singularity of Green’s function hinder the application of our method in more complex PDEs,such as variable coefficient equations,which will be investigated in our future work.

关键词： Green’s function Partial differential equation Boundary integral Neural network

On the minimum number of neighbors needed for consensus of flocks

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Control Theory and Technology 2017年第4期15卷 327-339页

作者： Chen Chen Ge Chen Lei Guo noah’s ark laboratory 2012 Lab Huawei Technologies Co. Ltd Beijing China LsC & NCMIs Academy of Mathematics and Systems Science Chinese Academy of Sciences Beijing China

This paper investigates consensus of flocks consisting of n autonomous agents in the plane, where each agent has the same constant moving speed v and updates its heading by the average value of the kn nearest agents from it, with vn and kn being two prescribed parameters depending on n. such a topological interaction rule is referred to as k,-nearest-neighbors rule, which has been validated for a class of birds by biologists and verified to be robust with respect to disturbances. A theoretical analysis will be presented for this flocking model under a random framework with large population, but without imposing any a priori connectivity assumptions. We will show that the minimum number of k~ needed for consensus is of the order O（log n） in a certain sense. To be precise, there exist two constants C1 〉 C2 〉 0 such that, if k 〉 C1 logn, then the flocking mode] will achieve consensus for any initial headings with high probability, provided that the speed vn is suitably small. On the other hand, if k 〈 Ca ]ogn, then for large n, with probability 1, there exist some initial headings such that consensus cannot be achieved, regardless of the value of Vn.

关键词： k-nearest-neighbor consensus topological interaction random geometric graph

On some Proximity Problems of Colored sets

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Journal of Computer science & Technology 2014年第5期29卷 879-886页

作者：范成林罗军王文成钟发荣朱滨海 shenzhen Institutes of Advanced Technology Chinese Academy of Sciences Huawei noah's ark laboratory state Key laboratory of Computer science Institute of Software Chinese Academy of Sciences College of Mathematics Physics and Information Technology Zhejiang Normal University Department of Computer science Montana State UniversityBozeman MT 59717 U.S.A. MemberACM

The maximum diameter color-spanning set problem（MaxDCs） is defined as follows： given n points with m colors, select m points with m distinct colors such that the diameter of the set of chosen points is maximized. In this paper, we design an optimal O（n log n） time algorithm using rotating calipers for MaxDCs in the plane. Our algorithm can also be used to solve the maximum diameter problem of imprecise points modeled as polygons. We also give an optimal algorithm for the all farthest foreign neighbor problem（AFFN） in the plane, and propose algorithms to answer the farthest foreign neighbor query（FFNQ） of colored sets in two- and three-dimensional space. Furthermore, we study the problem of computing the closest pair of color-spanning set（CPCs） in d-dimensional space, and remove the log m factor in the best known time bound if d is a constant.

关键词： computational geometry colored set algorithm maximum diameter color-spanning set problem

维普期刊数据库

Protect You More Than Blank: Anti-Learning sensitive User Information in the social Networks

在线全文

学校读者我要写书评

暂无评论

Journal of Computer science & Technology 2014年第5期29卷 762-776页

作者：袁明轩陈雷 Philip s. Yu 梅宏 Department of Computer science and Engineering The Hong Kong University of Science and Technology Huawei noah's ark laboratory Department of Computer science University of Illinois at ChicagoChicago IL 60637 U.S.A. FellowACMIEEE Key laboratory of High Confidence software Technologies Ministry of Education Institute of software School of Electronics Engineering and Computer Science Peking University Department of Computer science and Engineering Shanghai Jiao Tong University

social networks are getting more and more attention in recent years. People join social networks to share their information with others. However, due to the different cultures and backgrounds, people have different requirements on what kind of information should be published. Currently, when social network websites publish data, they just leave the information that a user feels sensitive blank. This is not enough due to the existence of the label-structure relationship. A group of analyzing algorithms can be used to learn the blank information with high accuracy. In this paper, we propose a personalized model to protect private information in social networks. specifically, we break the label-structure association by slightly changing the edges in some users＇ neighborhoods. More importantly, in order to increase the usability of the published graph, we also preserve the influence value of each user during the privacy protection. We verify the effectiveness of our methods through extensive experiments. The results show that the proposed methods can protect sensitive labels against learning algorithms and at the same time, preserve certain graph utilities.

关键词： social network privacy protection

维普期刊数据库

Audio2AB:Audio-driven collaborative generation of virtual character animation

在线全文

学校读者我要写书评

暂无评论

虚拟现实与智能硬件（中英文） 2024年第1期6卷 56-70页

作者： Lichao NIU Wenjun XIE Dong WANG Zhongrui CAO Xiaoping LIU school of Computer science and Information Engineering Hefei University of TechnologyHeifei 230009China school of software Hefei University of TechnologyHeifei 230009China Anhui Province Key laboratory of Industry safety and Emergency Technology Hefei University of TechnologyHefei 230601China

Background Considerable research has been conducted in the areas of audio-driven virtual character gestures and facial animation with some degree of ***,few methods exist for generating full-body animations,and the portability of virtual character gestures and facial animations has not received sufficient *** Therefore,we propose a deep-learning-based audio-to-animation-and-blendshape(Audio2AB)network that generates gesture animations and ark it's 52 facial expression parameter blendshape weights based on audio,audio-corresponding text,emotion labels,and semantic relevance labels to generate parametric data for full-body *** parameterization method can be used to drive full-body animations of virtual characters and improve their *** the experiment,we first downsampled the gesture and facial data to achieve the same temporal resolution for the input,output,and facial *** Audio2AB network then encoded the audio,audio-corresponding text,emotion labels,and semantic relevance labels,and then fused the text,emotion labels,and semantic relevance labels into the audio to obtain better audio ***,we established links between the body,gestures,and facial decoders and generated the corresponding animation sequences through our proposed GAN-GF loss *** By using audio,audio-corresponding text,and emotional and semantic relevance labels as input,the trained Audio2AB network could generate gesture animation data containing blendshape ***,different 3D virtual character animations could be created through *** The experimental results showed that the proposed method could generate significant gestures and facial animations.

关键词： Audio-driven Virtual character Full-body animation Audio2AB Blendshape GAN-GF

维普期刊数据库

spatio-temporal flow maps for visualizing movement and contact patterns

在线全文

学校读者我要写书评

暂无评论

Visual Informatics 2017年第1期1卷 57-64页

作者： Bing Ni Qiaomu shen Jiayi Xu Huamin Qu Department of Computer science and Engineering Hong Kong University of Science and TechnologyClear Water BayKowloonHong Kong noah’s ark Lab Huawei TechnologiesRoom 525-530Core Building 2HKSPShatinHong Kong

The advanced telecom technologies and massive volumes of intelligent mobile phone users have yielded a huge amount of real-time data of people’s all-in-one telecommunication records,which we call telco big *** telco data and the domain knowledge of an urban city,we are now able to analyze the movement and contact patterns of humans in an unprecedented *** map is widely used to display the movements of humans from one single source to multiple destinations by representing locations as nodes and movements as ***,it fails the task of visualizing both movement and contact *** addition,analysts often need to compare and examine the patterns side by side,and do various quantitative *** this work,we propose a novel spatio-temporal flow map layout to visualize when and where people from different locations move into the same places and make *** also propose integrating the spatiotemporal flow maps into existing spatiotemporal visualization techniques to form a suite of techniques for visualizing the movement and contact *** report a potential application the proposed techniques can be applied *** results show that our design and techniques properly unveil hidden information,while analysis can be achieved efficiently.

关键词： spatio-temporal data Flow map Urban mobility