Computation and Learning in Neural Networks With Binary Weights
Abstract
Under the aegis of the AFOSR grant they have been investigating computational learning attributes of networks of formal neurons. The formal neurons considered are linear threshold elements which produce binary outputs based on the sign of a linear form of a set of inputs. The researchers have been interested in (1) exploring the theoretical limitations on what can be computed or learnt in neural network architectures, and (2) developing and analyzing learning algorithms which specify weights as a function of a set of examples of a computation.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 28, 1992
- Accession Number
- ADA261182
Entities
People
- Santosh S. Venkatesh
Organizations
- Moore School of Electrical Engineering