Towards a Theory of Long-Step Algorithms for Large Scale Optimization

Abstract

This research achieved its goal to study the design of large steps first-order methods by exploiting the geometry or the regularity of the problems. New algorithms for optimization were developed, complexity analysis were completed and new geometries were explored for the previously developed No Lips algorithm, an instance of the Bregman method, which has been exploited by the optimization community. This research generated 15 articles and has already garnered more than 250 Google Scholar citations. The final report attached references section has links to the produced articles which provide additional details beyond the overview contained within the final report.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Oct 29, 2021
Accession Number
AD1153818

Entities

People

  • Jerome Bolte

Tags

Communities of Interest

  • Air Platforms
  • Autonomy
  • C4I
  • Materials and Manufacturing Processes
  • Space

DTIC Thesaurus Topics

  • Acquisition
  • Air Force
  • Air Force Research Laboratories
  • Algorithms
  • Artificial Intelligence
  • Artificial Intelligence Software
  • Convergence
  • Deep Learning
  • Geometry
  • Image Classification
  • Inverse Problems
  • Kernel Functions
  • Learning
  • Machine Learning
  • Mathematical Programming
  • Neural Networks
  • Nonlinear Programming
  • Operations Research
  • Optimization
  • Sequences

Readers

  • Distributed Systems and Data Platform Development
  • Operations Research
  • Technical Research and Report Writing.