报告摘要

在深度学习及相关领域的一个长期以来广受批评，但是至今仍在理论和实际中经常应用的一个重要假设是关于样本数据(包括训练数据和测试数据)的独立和同分布假设特别是当人们需要获得样本或者样本的函数的预期值，如预期效用，预期误差，预期风险暴露这些关键的统计数据时。但是另一方面，对于国内外的一些典型的金融数据的研究和分析表明这个在金融领域经常使用的假设导致了一些明显的错误，研究表明我们需要使用更加普适的非线性期望的理论和算法，以科学而稳健的消除这些积累误差和风险。本报告将深入浅出分析和解释为什么非线性期望算法可以用来更精准的计算藏在数据背后的不确定性。

Biography

Academician Peng is an eminent mathematician based at Shandong University where he serves as the Director of the Mathematics Institute and the Finance Institute.

Prof. Peng obtained his bachelor degree in physics from Shandong University in 1974, and his Ph.D. in mathematics from Paris Dauphine. He has been a professor with Shandong University since 1990. Among the numerous honors that he has received in his career, we mention his plenary lecture at the 2010 ICM, plenary lecture at the 2015 ICIAM, and his election to the Chinese Academy of Sciences in 2005.  Indeed, he is the sole ICM plenary speaker from China so far.  He is highly regarded for his exceptional contributions to mathematical sciences, his invention of backward stochastic differential equations and nonlinear expectation in particular. They have become indispensable tools in several branches of mathematics.