Moving toward zero-shot learning via analogy and imagination

Abstract

While computer vision has progressed rapidly in recent years, humans are still far superior in tasks where there are few (or even zero) examples of the target object in a visual search task. This project will address applications in visual object recognition where a human user must communicate to an autonomous system a desired object. How can the user communicate to the agent most effectively about which object to search for in the environment? How should the user communicate to the agent about a desired object that is not currently known to the agent? The proposed project will explore two algorithmic frameworks that leverage the low-dimensional structure of natural images for addressing these questions: learning through visual imagination and learning by analogy.

Document Details

Document Type
DoD Grant Award
Publication Date
May 22, 2016
Source ID
N000141512619

Entities

People

  • Christopher J. Rozell

Organizations

  • Georgia Tech Research Corporation
  • Office of Naval Research
  • United States Navy

Tags

Fields of Study

  • Computer science

Readers

  • Artificial Intelligence
  • Computer Vision.
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • Autonomy