An Instructable Connectionist/Control Architecture: Using Rule-Based Instructions to Accomplish Connectionist Learning in a Human Time Scale
Abstract
We describe a hybrid cognitive architecture that combines connectionist and controlled processing. The connectionist/control architecture (CAP2) uses instructions to decompose cognitive tasks into subtasks that can be learned in a human time scale. A CAP2 simulation model that uses the same task decomposition used by human subjects learns a logic task ten times faster than a standard connectionist model that does not use task decomposition. Rules for carrying out tasks are stored in a sequential network (Elman, 1988; Jordan, 1986) that controls the flow of information through a modular connectionist network. We argue that the CAP2 architecture better matches the human cognitive architectures than purely symbolic or purely connectionist architecture. Keywords: Human learning, Connectionist models, Automatic processing, Controlled processing, Cognitive architecture.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1989
- Accession Number
- ADA219274
Entities
People
- Walter Schneider
- William L. Oliver
Organizations
- Carnegie Mellon University