Transfer Learning in Integrated Cognitive Systems
Abstract
The ISLE Transfer Learning project was a three year DARPA-funded research effort whose goal was to enable machines to acquire knowledge in one domain and use it to improve performance in another. This capacity is a first step of a longer term vision that employs transfer to reduce the cost of developing intelligent software, hereby replacing the construction of multiple, one-off systems, with a methodology based on the automated transfer and augmentation of capabilities encoded in an initial source. The project emphasized machine learning and artificial intelligence (AI) techniques, and focused on the acquisition and transfer of behavioral knowledge (meaning strategies for how to act together with the supporting domain knowledge) from a source task to a target task drawn from a different domain. The work addressed three component problems within transfer: knowledge acquisition in a source task, knowledge transfer to a target task, and knowledge augmentation in the target domain given the transferred wisdom.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2010
- Accession Number
- ADA530009
Entities
People
- Alan Fern
- Dana S. Nau
- Daniel Mark Shapiro
- David J. Stracuzzi
- Pat Langley
- Pedro Domingos
- Peter Stone
- Ray Mooney