bet365 sports betting login 1
bet365 sports betting login title:Towards more scalable and interpretable natural language processing
bet365 sports betting login Announcement People:Cai Deng
Reporting location:School Main bet365 sports betting login Building313 Lecture Hall
bet365 sports betting login time：2022Year9month24日8:30-9:05
bet365 sports betting login Introduction:In recent years，Pre-training technology greatly improves the accuracy of natural language processing，But the existing system still has obvious shortcomings: on the one hand，Neural network models with huge parameters require huge computing resources when used。On the other hand，The knowledge and reasoning capabilities of the neural network model are contained in a large number of model parameters，There are problems that are difficult to explain and expand。Difficult to explain means that the decision-making basis and reasoning process of the model when making predictions are unknown，Difficult expansion means that updating knowledge of the model often requires retraining the huge model on new corpus，These have caused obstacles to practical application。forThe above questions, weProposedKnowledge in machine learning modelsSplit (memory) and reasoning (generalization)，Storing knowledge outside the neural network model，Make the neural network model itself more focused on inference problems。Knowledge and reasoning can be organically related through retrieval、Interconnection、Form a whole。Specifically，We construct explicit external memory to store knowledge，Memory extraction through retrieval，Symbolic representation and reasoning of simultaneous reference semantics，Convert text data into structured semantic representation and perform reasoning on semantic graph。This can make the basis for decision-making transparent and verifiable，Knowledge update plug and play，The reasoning process is clearly visible。This bet365 sports betting login will introduce the speaker’s research work in these two areas: (1) Explicit external memory，includes memory retrieval、Memory fusion、Three key links in feedback optimization；(2) Symbolic representation and reasoning of semantics，Extraction including symbolic representation、Modeling、Three key issues in reasoning，And future research directions。
Speaker Profile:Cai Deng, Ph.D. from the Chinese University of Hong Kong, mentor isProfessor Wai Lam。Research direction is natural language processing and machine learning，Including semantic parsing、Dialogue system、Machine translation, etc.。Especially the fusion of deep learning models and external explicit memory、Symbolic representation and reasoning of semantics。 in ACL、EMNLP、NAACL、NeurIPS,Published more than 20 papers in top conferences and journals such as AAAI，Google Scholar statistics cited more than 900 times. everWinACL2021 Outstanding Paper Award(first author),Once inIJCAI2022 and SIGIR2022 hold cutting-edge workshops(first speaker). long termACL, EMNLP, NeurIPS, Reviewer of top journals such as ICML，formerly COLING2022Text generation directionField Chairman. Was on AmazonAWS AI、Microsoft Research Redmond、Tencent AI Lab、Visiting Internship at Alibaba Damo Academy。
bet365 sports betting login二
bet365 sports betting login title:Task-based dialogue system based on transfer learning
bet365 sports betting login Announcement Person:Qin Libo
Reporting location:School Main bet365 sports betting login Building313 Lecture Hall
bet365 sports betting login time:2022Year9month24日9:05-9:40
bet365 sports betting login Introduction:Task-based dialogue system can help users complete some specific tasks，For example, check traffic、Check phone bill、Ordering、Ticket booking and consultation, etc.，Increasingly attracting attention from academia and industry。Although existing dialogue systems have made great progress，But they still require a large amount of labeled data for training，Making it difficult to generalize to data scarce scenarios。Therefore，In order to solve the problem of data scarcity in task-based dialogue scenes，This bet365 sports betting login explores the application of transfer learning in task-based dialogue systems: That is, using transfer learning technology to transfer knowledge from the source with sufficient data to the target with insufficient data to alleviate the problem of data scarcity in task-based dialogue systems。Specifically，This bet365 sports betting login will detail our work across tasks，Cross-domain，Work on alleviating data scarcity in task-based dialogue systems across language dimensions。
Speaker profile:Qin Libo，PhD student at Harbin Institute of Technology。Mainly research task-based dialogue systems、Natural language processing and other directions。2022Graduated from Harbin Institute of Technology with a PhD in Engineering，PhD at the National University of Singapore (remotely)，Westlake University，Microsoft Research Asia Internship Visit。atACL、EMNLP、COLING、AAAI、IJCAI、TASLPPublished many academic papers in important international journals and conferences in the fields of natural language processing and artificial intelligence。
bet365 sports betting login三
bet365 sports betting login title:Design of large-scale reinforcement learning framework for intelligent transportation systems
bet365 sports betting login Announcement Person:Wang Enshu
Reporting location:School Main bet365 sports betting login Building313 Lecture Hall; Tencent Conference935-979-861
bet365 sports betting login time:2022Year9month24日9:40-10:15
bet365 sports betting login Introduction:With the rapid development of reinforcement learning and the rapid popularization of intelligent transportation systems，The combination of the two has become an inevitable trend。More and more intelligent transportation systems based on reinforcement learning will emerge。However，Existing reinforcement learning frameworks face highly dynamic large-scale urban traffic environments，And issues unique to different intelligent transportation systems，Unable to effectively make adaptive decisions。Therefore，Aiming at the characteristics of different intelligent transportation systems，Need to design a new reinforcement learning framework and related optimization mechanisms。This bet365 sports betting login will take the traffic flow control problem of the online car-hailing platform and the information transmission problem of the collaborative driving system as examples，Discussing the framework design of large-scale competitive reinforcement learning and cooperative reinforcement learning。
Speaker profile:Wang Enshu，State University of New York at Buffalo, USA，PhD student in bet365 sports betting login science。Supervisor: Professor Qiao Chunming，IEEE Fellow。Research directions include deep learning、Reinforcement Learning、Intelligent transportation, etc.。2013Graduated from the School of Communication Engineering, Jilin University with a bachelor's degree，Won the National Scholarship for two consecutive years while studying。2016Master's degree in Communications and Information Systems from Jilin University。While studying for a PhD in the United States, I worked inIEEE Transactions on Mobile Computing、IEEE Transactions on Network Science and Engineering、IEEE Transactions on Intelligent Transportation Systems、IEEE International Conference on Distributed Computing Systems Published academic papers in internationally renowned journals and conferences10Remaining articles. He has presided over scientific research projects and obtained a total of scientific research funds74,638USD.
bet365 sports betting loginfour
bet365 sports betting login title:G-Mixup Graph bet365 sports betting login enhancement
bet365 sports betting login Announcement Person:Han Xiaotian
Reporting location:School Main bet365 sports betting login Building313 Lecture Hall; Tencent Conference935-979-861
bet365 sports betting login time:2022Year9month24日10:15-10:50
bet365 sports betting login Introduction:MixupBy interpolating features and labels between two random samples，Shows superiority in improving the generalization and robustness of neural networks。Traditionally，MixupCan be applied to rules、Gridded European bet365 sports betting login，such as image or tabular bet365 sports betting login。However，Adopt directlyMixupIt is difficult to enhance graph bet365 sports betting login，Because two different graph bet365 sports betting login are usually:1) has a different number of nodes;2) Not easy to align;3) There are unique types in non-Euclidean spaces。For this，We proposedG-Mixup, by interpolating different classes of graph generators (i.e.graphon) to enhance graph bet365 sports betting login。Specifically，We first estimate a using graphs in the same categorygraphon。Then，We do not directly operate graph bet365 sports betting login，But a graph generator that interpolates different categories in Euclidean space，Get hybrid graph generator，The synthesized graph is generated by sampling from the hybrid graph generator。Experimental results，Our method greatly improves the generalization and robustness of graph neural networks。
Speaker profile:Han Xiaotian，Fourth-year doctoral student at Texas A&M University。Mainly research graph neural networks、Graph bet365 sports betting login enhancement，Algorithmic fairness and other directions。2015Graduated from the School of Information, Shandong University with a bachelor's degree,2019Graduated from the School of bet365 sports betting login Science, Beijing University of Posts and Telecommunications with a master's degree。atICML、ICLR、TheWebConf、IJCAI、AAAI，TKDEPublished academic papers in international journals and conferences in the fields of machine learning and bet365 sports betting login mining10Remaining articles, in chargeICML、ICLR，NeurIPSMember of the meeting program committee. getICML2022 Outstanding Paper Award.
bet365 sports betting login 5
bet365 sports betting login title:Adversarial machine learning and robust optimization
bet365 sports betting login Announcement Person:Zhao Yue
Reporting location:School Main bet365 sports betting login Building313 Lecture Hall; Tencent Conference935-979-861
bet365 sports betting login time:2022Year9month24日10:50-11:25
bet365 sports betting login Introduction: This bet365 sports betting login will introduce adversarial machine learning from the perspective of Lupine optimization，And introduces the application of adversarial machine learning from the two perspectives of adversarial examples and data pollution3DResearch in bet365 sports betting login vision，Finally, we will briefly discuss the potential research topics of adversarial machine learning in related operations research。
About the speaker: Zhao Yue,2017Graduated from the Mathematics Department of Nanjing University,2016Visited the Department of Statistics at the University of California, Berkeley, in 2018,2022Graduated from the National University of Singapore with a Ph.D.，Currently a researcher at the Operations Research and Analysis Center of the National University of Singapore。Main research areas include machine learning、Operations Research and the intersection of the two。Research results in the field of machine learning have been publishedCVPR、ICCV (CCF AConference)Posted on，Mainly interested in trustworthy machine learning、Geometric Deep Learning。Research interest in the field of operations research is robust optimization、bet365 sports betting login-driven optimization and the application of optimization theory in operations management。
bet365 sports betting loginsix
bet365 sports betting login title:Privacy-preserving machine learning research for multi-omics bet365 sports betting login
bet365 sports betting login Announcement Person:Wang Qingyong
Reporting location:School Main bet365 sports betting login Building313 Lecture Hall; Tencent Conference935-979-861
bet365 sports betting login time:2022Year9month24日11:25-12:00
bet365 sports betting login Introduction:In recent years，Existing genes、Protein、Various omics bet365 sports betting login such as transcription and metabolism can help improve the performance of machine learning methods。However，Using multi-institutional datasets is a challenge，Due to privacy and security issues, medical bet365 sports betting login cannot be directly shared for model use。Therefore，Privacy-preserving machine learning methods are needed here。We use a federated learning model to train a shared global model to keep all sensitive bet365 sports betting login within our own institution，In order to improve the performance of the model without ignoring bet365 sports betting login privacy protection。In view of the personalized problems encountered in the federation training process。We propose a variety of new federal privacy protection training models，The results show，The method we proposed is safe，Original records will not be exposed to other parties，And the computational overhead during training is acceptable。
Speaker introduction: Wang Qingyong mainly studies machine learning and privacy computing。2016Graduated from Hubei University of Economics with a bachelor's degree in bet365 sports betting login science and technology，2018Graduated from Macau University of Science and Technology with a Master of Science in bet365 sports betting login and Information Systems，2022Graduated from the National University of Defense Technology in Management Science and Engineering and received a PhD in Engineering。atACM TOMM, Expert systems with applications, IEEE UIC, IEEE ISPA, Briefings in BioinformaticsPublished academic papers in important international journals and conferences in machine learning related fields10More articles. Served as a reviewer for more than ten international journals, such asIEEE T-SMC: systems, Neurocomputing, CILS, AIC, BSPC, CIBM (Elsevier)etc. He also received the National Postgraduate Scholarship from the Ministry of Education.
bet365 sports betting loginseven
bet365 sports betting login title:Research on expression recognition methods in complex scenes
bet365 sports betting login Announcement Person:Li Yingjian
Reporting location:Tencent Conference935-979-861
bet365 sports betting login time:2022Year9month24日14:00-14:35
bet365 sports betting login Introduction:Expressions are the most direct way for humans to express emotions、One of the most effective ways。With the development of Internet technology，More and more large-scale real scene expression data are being collected，These unconstrained data may have facial angle changes、Uneven lighting、Noise information irrelevant to emotional and semantics such as complex background and local occlusion，Severely affects the accuracy of visual emotion analysis in application scenarios。The research content of this bet365 sports betting login mainly focuses on two goals: improving the robustness of the model to unconstrained data noise in the same scene and improving the generalization ability of the model to diverse data across scenes。This bet365 sports betting login will analyze the current status and key challenges of expression recognition research，Introducing solutions with semantic mining as the core technology，And experimental verification results，Thus achieving the goal of improving model robustness and generalization capabilities。
Speaker profile:Li Yingjian，Harbin Institute of Technology，Fresh PhD graduates，Chinese Image and Graphics Society/ACM SIGMMStudent Member。The main research direction is visual emotion analysis，Transfer learning，Image classification。separated from2016Nianhe2018Received Bachelor of Science and Master of Engineering from Harbin Institute of Technology。Ph.D. student at Harbin Institute of Technology，One year of joint training at Nanyang Technological University in Singapore。atIEEE TMM，TAFFC，TCSVTandACM MMPublished more than ten papers in important journals and conferences, etc., served asIEEE TNNLS，TCSVTReviewers and other journalsACM MMMember of the meeting program committee.
bet365 sports betting logineight
bet365 sports betting login title:Based on deep learning3DHeart grid reconstruction
bet365 sports betting login Announcement Person:Chen Xiang
Reporting location:Tencent Conference935-979-861
bet365 sports betting login time:2022Year9month24日 14：35-15：10
bet365 sports betting login Introduction:Cardiovascular disease is one of the leading causes of death in the world。Reconstruct the corresponding image based on the patient’s scanned image3DHeart mesh model in simulation，Heart movement analysis，Surgical planning and navigation，It has huge application value in the analysis and diagnosis of cardiovascular diseases。However，Traditional heart3DMost mesh reconstruction algorithms have long reconstruction times，Reconstruction accuracy is not high enough，As well as the need to spend more time adjusting hyperparameters and other issues。This bet365 sports betting login will focus on improving the reconstruction speed，Introducing two deep learning-based methods from image to heart3DMesh reconstruction algorithm. They greatly enhance the heart3DGrid reconstruction speed，And can guarantee accuracy，Stable，Consistent reconstruction of cardiac topology，for3DThe heart grid provides more possibilities in practical applications.
Speaker profile:Chen Xiang，PhD student at University of Leeds。Mainly studies face generation、Key directions in the fields of bet365 sports betting login vision and medical image analysis such as medical image registration and three-dimensional mesh reconstruction of medical images。2016Graduated from the School of Electronic Information of Sichuan University with a bachelor's degree，In the same year, he was recommended to study for a master's degree at the Institute of Graphics, School of Electronic Information of our school。2019went to the University of Leeds in the UK to study for a PhD, joinedCISTIBLaboratory, apprenticeshipAlejandro F FrangiProfessor. existMICCAIandMedical Image AnalysisPublished many papers in top conferences and journals in the field of medical imaging，Authorized national invention patent1item。Won the national scholarship and various school-level scholarships during my undergraduate and master’s studies，Received a full PhD scholarship from the University of Leeds。Invited to review by many international authoritative journals，Being the top publication in the field IEEE Transactions Medical ImagingRated as an excellent reviewer.
bet365 sports betting loginNine
bet365 sports betting login title:Research on key technologies of smooth particle hydrodynamics method for high-performance computing
bet365 sports betting login Announcement Person:Long Sifan
Reporting location:Tencent Conference935-979-861
bet365 sports betting login time:2022Year9month24日15：10-15：45
bet365 sports betting login Introduction:Continuous innovation of random bet365 sports betting login hardware，The computing performance of the bet365 sports betting login has been greatly improved，Making it possible to handle a large number of computationally intensive tasks，Since the 1950s，Computational Fluid Dynamics(Computational Fluid Dynamics) was proposed for the first time，And with the development of computers, there is an intermediary mathematics、The intersection between fluid mechanics and computers。Its main research content is to solve the governing equations of fluid mechanics through computers and numerical methods，Simulate and analyze fluid dynamics problems。The corresponding numerical calculation methods and experimental physics experiments、Theoretical analysis has become the three major research methods of computational fluid dynamics。And is widely used in aerospace、Energy Development、Biological information、Industrial Manufacturing、Biomedicine、Meteorological Ocean、Weapons research and development and other fields。Depending on the discrete format used (Eulerian method or Lagrangian method)，The corresponding numerical methods can be divided into grid-based methods and gridless particle methods。Their principles are to divide the solution area in the space into different grids or particle nodes as interpolation，Grid/The denser the particles are divided，The more accurate the calculation results obtained，Therefore，Computational fluid dynamics requires a large amount of computing resources to obtain solutions with practical engineering significance，It gave birth to parallel computing，The application of parallel computing in numerical simulation has become an important manifestation of the country's overall scientific and technological level。This research mainly reports on the classic meshless method-Research progress of smooth particle hydrodynamics in the field of parallel computing，and willSPHThe method is applied in large-scale engineering examples and key technology research on multi-physics coupling，Committed to promotionSPHMethods should be used to solve practical engineering application problems.
Speaker profile:Long Sifan，PhD candidate at National University of Defense Technology (expected to graduate by the end of the year)。The main research direction is parallel computing、Computational Fluid Dynamics、Numerical simulation method、Artificial Intelligence、bet365 sports betting login Mining。2016Graduated from Changsha University in2019Received a master's degree in bet365 sports betting login science from Central South University。Joined the Galaxy Technical Service Department of the bet365 sports betting login Research Institute of the National University of Defense Technology in the same year to engage in the establishment of mathematical libraries、Optimization research work on physical models。In artificial intelligence、Parallel Computing、Number of publications in high-level international journals and conferences in the field of computational mechanics10Articles, including the following journals and conferencesInternational Conference on Parallel Processing (ICPP)、Journal of Computational Physics、Artificial Intelligence Review、EntropyAnd other journals, apply for patent3items and2software copyright.
bet365 sports betting login十
bet365 sports betting login title:Learning for Cross-Layer Optimization: Joint Subcarrier Allocation and Beamforming Design in MEC-Aided Joint Transmission Networks
bet365 sports betting login Announcement Person:Zheng Chong
Reporting location:Tencent Conference935-979-861
bet365 sports betting login time:2022Year9month24日15:45-16:20
bet365 sports betting login Introduction:To address ongoing challenges in future wireless networks，A large number of applications such as large-scale multiple input and multiple output have emerged (MIMO), mobile edge computing (MEC)Collaborate with multiple points (CoMP) United Transmission (JT) and other promising wireless communication technologies. Among them, forMECAuxiliaryCoMP JTThe system performs cross-layer optimization design of transmission scheme，Be able to fully utilize the transmission and computing resources in the network to significantly increase the bet365 sports betting login transmission rate。This contains two key issues: subcarrier allocation of the media access control layer (SA) Beamforming with the physical layer (BF) Design。However，The technical bottleneck of traditional optimization methods poses a major challenge to solving this type of cross-layer optimization problems。This bet365 sports betting login will focus onMECAuxiliaryCoMP JTExpand the communication system and introduce the allianceSAandBFResearch status of cross-layer optimization problems，Key Challenges，And how to proceed from the perspective of deep learningSAandBF’s joint cross-layer optimization design，To maximize the system weighted sum rate。
Speaker profile:Zheng Chong，PhD student of Southeast University。Mainly researching intelligent wireless communications，Edge Intelligence，Privacy Protection，Key directions in future network fields such as machine learning。2017Graduated from the School of Communication Engineering, Chongqing University with a bachelor's degree，Currently studying for a PhD at the State Key Laboratory of Mobile Communications, School of Information Science and Engineering, Southeast University，One year of joint training at the Singapore University of Technology and Design during the PhD period。atIEEE JSAC, IEEE TWC, IEEE IoTJ, IEEE TVT, IEEE GLOBECOM, IEEE WCNCPublished in important international journals and conferences in related fields/In investment7Academic papers, authorized/Public national invention patent4item。Participate in the national key R&D program，National Natural Science Foundation of China，Huawei and other projects。
bet365 sports betting loginEleven
bet365 sports betting login title:Set reachability query on label graph
bet365 sports betting login Announcement Person:Zeng Yuanyuan
Reporting location:Tencent Conference935-979-861
bet365 sports betting login time:2022Year9month24日16:20-16:55
bet365 sports betting login Introduction:Graphic bet365 sports betting login is an important part of big bet365 sports betting login，Widely used in modeling of complex structured bet365 sports betting login，has broad application prospects。Over the past ten years，On many fundamental issues in managing and analyzing graph bet365 sports betting login，People have done a lot of research work。This speech will focus on a research problem: set reachability query on label graph；Focus on its distributed solutions。
Speaker profile:Zeng Yuanyuan，School of Information Science and Engineering, Hunan University，PhD candidate in bet365 sports betting login science and technology。The main research direction is graph data management、Parallel and distributed computing。Currently inTPDS, ICDEPublished many academic papers in important international journals and conferences.
bet365 sports betting loginTwelve
bet365 sports betting login title:Traffic management mechanism of bet365 sports betting login center bet365 sports betting login plane
bet365 sports betting login Announcement Person:Liu Jingling
Reporting location:Tencent Conference935-979-861
bet365 sports betting login time:2022Year9month24日16:55-17:30
bet365 sports betting login Introduction:As the traffic handled by data centers increases dramatically，Data center network becomes a potential performance bottleneck，Management of data center traffic is the key to improving data center network performance。However，Traffic in the data center network has strong bursts、High concurrency、Demand various features，And the data center topology has high bandwidth、Low latency、Shallow cache and multipath features，This brings great challenges to traffic management in data center networks。This bet365 sports betting login will focus on two key issues of data plane traffic management: load balancing and cache management。Will introduce the research status and key challenges of the two issues，And discuss how to ensure the performance of load balancing in asymmetric topology scenarios、How to solve the problem of fair cache allocation for ultra-high concurrent traffic。
Speaker profile:Liu Jingling，Ph.D. of Central South University。Main research interests include bet365 sports betting login center networking、Deterministic Network、Distributed machine learning and performance optimization of streaming media。2016Graduated from the School of Information Science and Engineering, Central South University in 2011 with a bachelor's degree in information security，One year of joint training at the Singapore University of Technology and Design during the PhD period。atIEEE/ACM TON、IEEE TCC、IEEE ICNP、IEEE ICDCS、Published academic papers in domestic and foreign academic journals and conferences in network system-related fields such as Journal of Software17Article, obtained4Authorization of national invention patents。Received the President’s Innovation Award from Central South University、Titles such as outstanding students of Central South University。Participated in multiple national key projects、Front Project、Provincial and ministerial key projects。
