学术报告（叶颀 2024.1.26）

摘要：In this talk, we show a new mathematical framework of machine learning to combine the data driven methods and model-driven methods. Usually the data-driven methods and model driven methods are used to introduce the black-box algorithms and white-box algorithms, respectively. The original idea is to use the local information of multimodal data and multiscale models to construct the global approximate solutions by the learning algorithms. The work of the composite algorithms provides another road to study the mathematical theory of machine learning including the interpretability in approximation theory, the nonconvexity and nonsmoothness in optimization theory, and the generalization and overfitting in regularization theory. For our project of computational medicine for pancreatic cancer, we study the composite algorithm of image processing and modeling simulation.