GRAPH DEEP LEARNING FOR MULTIMODAL DATA WITH OPTIMAL TRANSPORT OPTIMIZATION METHODS

Abstract

Graph deep learning (GDL) is a fundamental model which can learn with graph structural representation. Graph Deep Learning showed that it is a powerful machine learning method and its successfully apply for many AI applications ranging from natural language processing to computer vision. Nowadays, with the growing of big data, the data now is including text, image, audio, and video. Some other data such as medicine which can consist of the information from manual document and molecular structure of medicine. The critical challenge for multi-modal data is that we need to find the common representation for those data. Besides, an essential requirement for GDL is that they need large annotated training data. Therefore, in this research, we would like to study the optimal transport optimization method (OPT) that can utilize unlabeled data and find common representations in multi-modal representation. In addition, we consider using OPT to improve the Generative Adversarial Network that aligns unlabeled data to the space of labeled data via the game theory model.

Document Details

Document Type
DoD Grant Award
Publication Date
Apr 20, 2023
Source ID
FA23862214039

Entities

People

  • Le-Minh Nguyen

Organizations

  • Air Force Office of Scientific Research
  • Japan Advanced Institute of Science and Technology
  • United States Air Force

Tags

Fields of Study

  • Computer science

Readers

  • Neural Network Machine Learning.
  • Trauma or Military Medicine

Technology Areas

  • AI & ML
  • AI & ML - Machine Learning Algorithms
  • AI & ML - Neural Networks
  • Space