Lifelong Transfer Learning for Heterogeneous Teams of Agents in Sequential Decision Processes
Abstract
Transferring knowledge from prior experience to a new problem is a key characteristic of human intelligence, enabling us to continually build upon and refine our knowledge. Recent work on transfer in machine learning for autonomous systems has demonstrated that knowledge transfer can improve model performance and accelerate learning. However, current research tends to focus on transfer to a single new problem, showing little consideration for the challenges of transfer learning over consecutive tasks and across diverse domains. Our work develops methods that enable teams of heterogeneous agents to rapidly adapt control and coordination policies to new scenarios, combining lifelong transfer learning and autonomous instruction to support continual transfer among heterogeneous agents and across diverse tasks. We apply these methods to sequential decision-making (SDM) tasks in dynamic environments with simulated and physical robots.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2016
- Accession Number
- AD1011856
Entities
People
- Eric Eaton
- Matthew Taylor
- Paul Ruvolo
Organizations
- Washington State University