Bingyi Jing
Chair Professor of Statistics and Data Science, Southern University of Science and Technology
Bingyi Jing, Chair Professor of Statistics and Data Science, Southern University of Science and Technology, National Specially Appointed Expert, Second Prize of National Natural Science Award, Chair Professor of Changjiang Scholars of Ministry of Education, Second Prize of Natural Science Award of Higher Education Institutions of Ministry of Education, Fellow of American Statistical Association, Fellow of Institute of Mathematical Statistics, Elected Member of International Statistical Institute, Fellow of International Statistical Institute. Member of the International Statistical Institute (ISI). He is the chairman of the Multivariate Analysis Committee of the Chinese Society of Field Statistics, and has served as associate editor of seven international academic journals, including Ann Appl Probab, JBES, and Science China. His research interests include: probability statistics, data science, artificial intelligence, big models and other fields. He has published more than 110 papers in top journals and top conferences such as Ann Stat, Ann Prob, JASA, JRSSB, Biometrika, J of Econometrics, JBES, Bioinformatics, JMLR, Science China, NeurIPS and so on.
Topic
Large-scale distributed systems for large models
With the rapid growth in model size and context length requirements, current algorithmic innovations in large-scale deep learning algorithms, both for LLM and multimodal LVLM, rely heavily on the infrastructure of distributed systems. Distributed systems are ubiquitous throughout the model lifecycle, from streaming data transfer to multidimensional parallelism of the model to inference in the subsequent deployment phase. We address the design of a good distributed system for large models, oriented to the entire model lifecycle. Outline: Stream-oriented and cloud-storage data design Distributed parallel training schemes for very long sequences Construction of high-throughput inference systems
