Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 319 - SLDS CSpeed 6
Type: Contributed
Date/Time: Wednesday, August 11, 2021 : 3:30 PM to 5:20 PM
Sponsor: Section on Statistical Learning and Data Science
Abstract #318888
Title: Efficient Path Following Algorithms and Its Applications to Case Influence Assessment
Author(s): Qiuyu Gu* and Renxiong Liu and Yunzhang Zhu
Companies: The Ohio State University and Ohio State University and Ohio State University
Keywords: path following algorithm; case influence; case weight; optimization

We present a path-following algorithm for a parametric optimization problem, which consists of a family of similar optimization problems indexed by a single parameter. The algorithm is motivated by case influence assessment using case influence graph, where case weight is introduced for each case [Cook, 1986], and computing a case influence graph is equivalent to solving a parametric optimization problem indexed by case weight. Our algorithm is based on the finding that the closeness between the approximated solution path and the true solution path depends on two quantities: how finely the grid points are and how accurate the optimization algorithm runs at each grid point. We call them interpolation error and optimization error respectively. Our algorithm is designed such that the two quantities are well balanced, without causing redundant computation. We demonstrate through simulation that our algorithm performs better than some standard numerical ODE solvers in terms of accuracy and runtime. We also show that our algorithm can be used to efficiently assess the case influence and identify the mislabeled images in the CIFAR-10 dataset.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2021 program