BIMSA >
Seminar on Control Theory and Nonlinear Filtering
Backprogation and adjoint differentiation of chaos
Backprogation and adjoint differentiation of chaos
Organizer
Speaker
Xiuang Ni
Time
Friday, December 29, 2023 9:00 PM - 10:00 PM
Venue
Online
Abstract
Computing the linear response, or the derivative of long-time-averaged observables with respect to system parameters, is a central problem for statistics and engineering. Conventionally, there are three straight-forward formulas for the linear response: the pathwise perturbation (including the backpropagation method), the divergence, and the kernel differentiation formula. We shall explain why none works for the general case, which is typically chaotic, high-dimensional, and small-noise.
We present the fast response formula, which is an 'ergodic-theorem' for the linear response of hyperbolic chaos. It is the average of some function of u-many vectors over an orbit, where u is the unstable dimension, and those vectors can be computed recursively. Then we discuss how to further incorporate the kernel differentiation trick to overcome non-hyperbolicity.
主讲人简介:
倪昂修现任丘成桐数学中心助理教授。他2021年毕业于加州大学伯克利分校(导师John Strain),2023年博士后出站于北京大学(导师张平文)。他的主要研究方向为计算动力系统和工程应用,尤其是混沌系统的参数导数计算。