清华主页 EN
导航菜单

[AMSS-YMSC-BIMSA Joint Seminar on Progress of Topology and Its Applications] Identifying Nonlinear Dynamics with High Confidence from Sparse Time Series Data

来源: 05-05

时间:2022/5/5 9:00-10:30

地点:1110 & ZOOM: 388 528 9728, 密码: BIMSA

组织者:Konstantin Mischaikow

主讲人:Konstantin Mischaikow

题目: Identifying Nonlinear Dynamics with High Confidence from Sparse Time Series Data


摘要:

I will introduce a novel 5 step procedure that given time series data generated by a stationary deterministic nonlinear dynamical system provides a lower bound on the probability that the system generates specific local and/or global dynamic behavior.
More precisely,  the time series data is used to define a Gaussian process (GP).
The mean of this GP provides a surrogate model.
The dynamics of the surrogate model is interrogated using combinatorial methods and characterized using algebraic topological invariants (Conley index).
The GP predictive distribution provides a lower bound on the probability that these topological invariants (and hence the characterized dynamics) apply to the unknown dynamical system (a random path of the GP).  
The focus of this talk is on explaining the ideas, thus I restrict our examples to one-dimensional systems and show how to capture the existence of fixed points, periodic orbits, connecting orbits, chaotic dynamics, and bistability.
This is based on joint work with B. Batko, M. Gameiro, Y. Hung, W. Kalies, E. Vieira, and C. Thieme


返回顶部
相关文章
  • [AMSS-YMSC-BIMSA Joint Seminar on Progress of Topology and Its Applications] Geometry and Topology of Data

    Title: Geometry and Topology of DataAbstract: Data sets are often equipped with distances between data points, and thereby constitute a discrete metric space. We develop general notions of curvature that capture local and global properties of such spaces and relate them to topological concepts such as hyperconvexity. This also leads to a new interpretation of TDA.

  • Factor Modeling for Clustering High-dimensional Time Series

    AbstractWe propose a new unsupervised learning method for clustering a large number of time series based on a latent factor structure. Each cluster is characterized by its own cluster-specific factors in addition to some common factors which impact on all the time series concerned. Our setting also offers the flexibility that some time series may not belong to any clusters. The consistency with...