A Workshop on High-Level Connectionist Models
Abstract
The difficulties that continually show up in connectionist modelling attempts directed towards high-level cognitive processing and knowledge representation include the inter-related problems of generative capacity, representational adequacy, variable binding, multiple instantiation of schemata (concepts, frames, etc.), rapid construction and modification of information structures, task control, and recursive processing. These problems are of more direct concern to computer scientists than to psychologists, philosophers or neuroscientists. Accordingly, the purpose of the workshop is to bring together computer-science oriented connectionist researchers who have addressed the problems, so as to understand the issue of achieving high-level cognitive processing in connectionist systems more clearly. The following is a list of preferred topics for presentations and discussions at the workshop. The topics are not orthogonal. 1. Connectionist/neural implementation of (aspects of) commonsense reasoning, planning, natural language understanding and rapid, 'one-shot' learning. 2. Ways of coping with the productivity of natural language. 3. Ways of encoding complex information in connectionist/neural models. 4. Capabilities and limitations of coarse-coded and distributed representations. 6. The relative advantages of localist and distributed representations. 7. Connectionist implementation of temporary, dynamic, complex data structures (such as frame instantiation, network fragments, stacks, trees). 8. Multiple simultaneous instantiation of rules, concept structures, schemata, frames, etc. 9. Variable binding. 10. Task control, including sequencing, iteration and recursion. 11. Real and apparent rule following.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 11, 1988
- Accession Number
- ADA198759
Entities
People
- John A. Barnden
- Jordan B. Pollack
Organizations
- New Mexico State University