Domain-Adaptive Active Meta-Learning (DAAML)

Abstract

Our Domain-Adaptive Active Meta-Learning (DAAML) system leverages semi-supervised few-shot learning including unsupervised representation learning for maximized data economy, active sampling (i.e., active learning) to maximize information gain per label query, and adaptive module selection for cross-domain few-shot learning. In this report, we detail our technical approach and experimental results regarding these three major building modules. This effort is funded by the DARPA Learning with Less Labels (LwLL) program.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Sep 01, 2021
Accession Number
AD1148460

Entities

People

  • Giedrius Burachas
  • Meng Ye
  • Nikoletta Basiou
  • Ray Kolczynski
  • Xiao Lin
  • Yi Yao

Organizations

  • SRI International

Tags

Communities of Interest

  • Energy and Power Technologies

DTIC Thesaurus Topics

  • Accuracy
  • Air Force
  • Artificial Intelligence Software
  • Birds
  • Computational Science
  • Computer Vision
  • Cross Domain
  • Detection
  • Dimensionality Reduction
  • Image Classification
  • Information Science
  • Information Systems
  • Machine Learning
  • Neural Networks
  • Ontologies
  • Semi-Supervised Learning
  • Supervised Machine Learning

Fields of Study

  • Computer science

Readers

  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • AI & ML - Neural Networks