seabet online sports betting ：4month21Morning9:30-11:30
Reporting location：Computer Building of Central South University313Lecture Hall
Report 1
Title：The responsibilities and challenges of large models
Report Introduction：The latest breakthrough in artificial intelligence technology makesAI’s performance in various fields has reached human levels，But it also raises concerns about its possible impact on society。For this，At Microsoft Research Asia，We have conducted several research projects，Aimed at making large models more responsible，And solve its privacy protection、Issues such as values and interpretability。In this report，We will dive into the responsibilities and challenges of large models，And describe our ongoing efforts to develop more responsible AI technologies。At the same time，We will also explore potential directions for future seabet casino review research，To enhance the sustainability and social responsibility of artificial intelligence technology。
About the speaker：
Dr. Xie XingJoined Microsoft Research Asia in July 2001，Current Chief Researcher，And concurrently serves as a doctoral supervisor at the University of Science and Technology of China。His research team is in data mining、Innovative research in areas such as social computing and responsible artificial intelligence。so far，He has published more than 300 academic papers，Cited more than 40,000 times，H index is 99。He received the first Microsoft Scholar Award (1999)、ACM SIGSPATIAL Ten Years Influential Paper Award (2019)、China Computer Federation Green Bamboo Award (2019)、ACM SIGSPATIAL Ten Years of Influential Paper Honors Award (2020)、ACM SIGKDD China Time-Tested Paper Award (2021) and ACM SIGKDD China Time-Tested Paper Award (2022)，And in KDD、Won multiple best paper awards at ICDM and other conferences。He is a fellow of the China Computer Federation、IEEE Fellow seabet sports betting and ACM Distinguished Member。
Report二
Title：Deep representation learning inApplications and challenges in Xbox game recommendation system
Report Introduction:Representation learning aims to automatically extract effective and streamlined features from complex raw data，Used to solve practical tasks or support the training of downstream machine learning models。Representation learning is widely used in recommendation systems，For example，Collaborative filtering model based on representation learning to user-Bipartite graph formed by items as raw data，Learning embedding representations of users and items for recommendation or classification tasks。This report is combined with actual scenarios in Xbox game recommendations，Introducing us at WebConf 2023's two jobs，Respectively correspond to two challenges in deep representation learning applications: scalability and interpretability。
First, we found that graph neural network (GNN) still has many shortcomings in learning large-scale seabet casino review representations，For example, over-smoothing and scalability issues caused by multi-hop information propagation；Each node is assigned a learnable vector，Resulting in the overall number of parameters of the model being too high，Difficult to support large-scale training, etc.。To overcome these problems, we proposedA new oneGNN model: xGCN。xGCN embeds node ID into vector viewIt is a static feature rather than a parameter that requires gradient training，And use an unsupervised propagation process and aThe "refinement network" composed of MLP to continuously and iteratively update the embedding vector。xGCN can significantly reduce the number of model parameters、Accelerate model training，Also achieved better prediction accuracy on social recommendation tasks。On the other hand,Deep representation learning models are often uninterpretable，But improving the interpretability of the model will enhance the persuasiveness of the recommendation results，Improve user trust and seabet casino review satisfaction, etc.。For this，We propose an interpretable collaborative filtering frameworkECF，Aims to mine some interpretable interest clusters from users’ group behavior，By learning sparse mapping of users/items to interest clusters，So that the labels of interest clusters and the path of user-interest cluster-item can be used to achieve explainable recommendations。In order to learn high-quality interest clusters，usFrom semantic similarity、Different loss functions are designed in three aspects: label similarity and independence to optimize the model end-to-end。ECF has a wide range of application scenarios，For example user tag generation、Explainable recall、Recommended list of similar items with themes、Similar crowd models in targeted advertising, etc.。
ReportPerson profile：
Dr. Lian Jianxun, Lead Researcher at Microsoft Research Asia,2018Graduated from University of Science and Technology of China，Received a doctorate degree in computer application seabet casino review technology。His research interests include user modeling and recommendation systems。He has participated in many international conferences，includesKDD, IJCAI， WWW， SIGIR ，Published 2 on CIKM and WSDM0Remaining papers，And applied relevant research results to Bing advertising、On multiple actual recommended application scenarios such as Xbox games and Microsoft news，obtained significant improvement in key indicators。 He is also the Microsoft recommendation system open source library Microsoft Recommender.
