Reporter: Ding Hu, University of Science and Technology of China
Reporting location: New Campus Information Building535
Report time:July 18, 2024 (Thursday) 10:00-12:00 am
Report title: Research on several high-dimensional computational geometry problems for artificial intelligence
Personal introduction:
Ding Hu,Graduated from the Department of Mathematics, Sun Yat-sen University in 2009，Graduated from the Department of Computer Science and Engineering, State University of New York at Buffalo in 2015，Won the best doctoral thesis in the department。Main research directions include computational seabet casino review geometry、Big data optimization algorithm，And in artificial intelligence、Chip Design、Applications in biomedicine and other fields。Formerly an assistant professor in the Department of Computer Science and Engineering (tenure-track), Michigan State University, USA，Doctoral Supervisor。Worked at the School of Computer Science, University of Science and Technology of China in June 2018，Specially Appointed Professor，Doctoral Supervisor。Has published nearly 50 articles in well-known international conferences and journals。The scientific research results of which he is the first/corresponding author were published in SODA、SoCG、ICALP and other algorithm theories，and NeurIPS、The top international conference in application fields such as ICML。Won the CRII Award from the National Science Foundation of the United States, seabet app download Berkeley Simons Fellowship, etc.。Invited many times to attend internationally renowned conferences、Reviewer or program committee member of the journal。As an expert in computer algorithms，Invited to participate in the project review process of the National Science Foundation。Presided over a number of national/provincial and ministerial level scientific research projects，Currently the person in charge of the Young Scientist Project of the National Ministry of Science and Technology’s Key Research and Development Program (2021-2026)，Leader of the Innovation Team Cultivation Project of the University of Science and Technology of China (2022-2024)。
Report Introduction:
With the development of big data and computing equipment，Artificial intelligence has affected every aspect of our lives。However，How to design efficient algorithms is still one of the main bottlenecks limiting the seabet sports betting application of artificial intelligence technology。Since the data required by many artificial intelligence models are in high-dimensional geometric space，Some difficult artificial intelligence algorithm problems can be solved through computational geometry technology in high-dimensional space。In this report，We will introduce three specific examples of research in this area。（1) Density clustering algorithm DBSCAN in high-dimensional space。Clustering is one of the basic methods in many large-scale data processing scenarios。DBSCANAs one of the most popular density clustering algorithms，Exhibits very excellent data processing capabilities in low-dimensional space。However，Due to heavy reliance on operations such as neighbor counting，DBSCAN is difficult to apply to high-dimensional data。Our new DBSCAN algorithm can effectively break seabet mobile through this limitation，Gives the linear time complexity in high-dimensional space，and can be easily applied to streaming data。(2) Minimum ball coverage problem in high-dimensional space。The minimum sphere coverage problem is a basic high-dimensional geometric optimization problem，Especially has a very important connection with classification problems in machine learning。However, the traditional minimum ball coverage algorithm often requires repeated iterative reading of data，The algorithm is less efficient。Here we will introduce a novel sublinear algorithm，Significantly reduces the complexity of existing algorithms。(3) Data dependent core set construction method。The core set is to reduce the data size、One of the conventional methods to improve calculation efficiency。However, existing core set construction methods seabet casino review often rely on the VC dime of the optimization targetnsion, And the construction complexity is usually higher。We will introduce a method suitable for the field of machine learningGeneral core set construction method of ERM model。Our method does not depend on the optimization goalVC dimension, Can effectively handle dynamic updates at the same time,OutliersAnd other questions.
