A0815
Title: Temporal label recovery via manifold learning
Authors: Wanjie Wang - National University of Singapore (Singapore) [presenting]
Yuehaw Khoo - University of Chicago (United States)
Xin Tong - National University of Singapore (Singapore)
Yuguan Wang - University of Chicago (United States)
Abstract: Often, dynamical data are collected without clear temporal labels. In order to recover the temporal labels from noisy data points, spectral algorithms are developed based on the graph Laplacian. The method does not require the similarity matrix to have certain monotone properties, commonly assumed in existing spectral seriation algorithms. The L-infinity error of the estimators is analyzed for the ordering, and the algorithms are applied to datasets from toy cryo-electron microscopy (cryo-EM) examples to demonstrate the efficacy of the methods.
