Out-of-distribution Detection In Real-world Environments

Abstract

Modern deep neural networks have achieved remarkable success in known contexts for which they are trained. Yet, they often strugglewith unknown situations in complex open-world environments---e.g., samples from unknown classes that the network has not been exposed to during training, but can be blindly classified as a known class with high confidence in testing. To unlock the full potential of machine intelligence in open-world environments, the project will contribute new algorithmic techniques for building safe and resilient learning methods for real-world environments. New out-of-distribution (OOD) detection algorithms will be developed for imperfect data regimes, such as insufficient or imbalanced training data. Additionally, this project will develop strategies for applying algorithmic advances to Naval applications. It will demonstrate the utility of new techniques on operationally relevant data, as well as document specific use cases for existing programs. These use cases willalso identify gaps that will inform subsequent algorithmdevelopment.#Approved for Public Release.#

Document Details

Document Type
DoD Grant Award
Publication Date
Jul 24, 2023
Source ID
N000142312643

Entities

People

  • Sharon Li

Organizations

  • Office of Naval Research
  • United States Navy
  • University of Wisconsin System

Tags

Fields of Study

  • Computer science

Readers

  • Agent-Based Social Robotics and Mobile-Assisted Learning in Virtual Environments.
  • Neural Network Machine Learning.
  • Software Engineering.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - DoD AI Strategy
  • AI & ML - Neural Networks