Extending Semantic and Episodic Memory to Support Robust Decision Making
Abstract
Our research focused on developing and evaluating general, effective, and efficient algorithms for learning of long-term knowledge in autonomous agents, as well as developing cognitive capabilities that exploit that learning. Our work has covered episodic memory, semantic memory, and procedural memory, integrated within a general cognitive architecture (Soar). For episodic memory, our research has led to significant improvements in the efficiency of storage (memory) and retrieval (time) through the exploitation of temporal contiguity, structural regularity, high cue structural selectivity, high temporal selectivity, low cue feature co-occurrence, resulting in no significant slowdown with experience: runs for days of real time (tens of millions of episodes), faster than real time. We evaluated our approach on multiple tasks (including mobile robotics, games, planning problems, linguistics) and for multiple cognitive capabilities.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 13, 2013
- Accession Number
- ADA586671
Entities
People
- John E. Laird
Organizations
- University of Michigan