Discovering Novel Communication Algorithms via Machine Learning

Abstract

Coding theory is a central discipline underpinning wireless communications that are the workhorses of the information age. Significant progresses have been made over the past century, with enormous impact on humanity. Landmarks include convolutional codes, Turbo codes, Low-Density Parity-Check codes, and most recently polar codes, which are standards in cellular, satellite, and deep space communications. However, such progresses in coding theory is largely driven by individual human ingenuity, and breakthroughs have inevitably been sporadic over the past century. The objective of this proposal is to automate the discovery of coding and decoding algorithms via deep learning. Canonical channel models of communication systems provide excellent platforms for training neural networks, as unlimited training examples can be generated, there is no issues of over-fitting, and there are gold standards on evaluating the learned coding schemes. If successful, this provides a new data-driven framework for designing codes, which can fill-in the current gaps in performance of existing schemes in the feedback channel.

Document Details

Document Type
DoD Grant Award
Publication Date
Feb 14, 2019
Source ID
W911NF1810384

Entities

People

  • Sewoong Oh

Organizations

  • Army Contracting Command
  • United States Army
  • University of Illinois Urbana–Champaign

Tags

Fields of Study

  • Computer science

Readers

  • Computer Programming and Software Development.
  • Educational Psychology
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • AI & ML - DoD AI Strategy
  • AI & ML - Neural Networks
  • Space