数学系 Seminar 第1996讲
报告主题：Model-based Data Assimilation versus Data-driven Machine Learning(基于模型的数据同化与数据驱动的机器学习）
报告人：Haixiang Ling 教授（荷兰代尔夫特理工大学应用数学系）
会议ID：会议ID：876 571 395
报告摘要：Uncertainty is common in real life, both mathematical-physical models and observations contain uncertainties. Data assimilation is a method which uses the information of observation data to reduce the uncertainty in the model consequently improving the forecast accuracy of the model. Machine learning is a data-driven method which tries to find the important features and their relations from the data, in contrast to model-based data assimilation, machine learning techniques do not require a mathematical-physical model and try to fit the data into some functional relationship through an optimization procedure. In this sense machine learning is therefore an “interpolation” method without paying attention to “extrapolation”. Combining the power of the model-based data assimilation method and the data-driven machine learning technique is the focus of many recent research, in this talk we will discuss some examples of this development.