ISSN:
1433-3058
Keywords:
Sign-constrained weights
;
Storage capacity
;
Dynamics
;
Dynamic threshold
Source:
Springer Online Journal Archives 1860-2000
Topics:
Computer Science
,
Mathematics
Notes:
Abstract In this paper we begin by briefly reviewing recent interest in neural nets with sign-constrained weights, and outline recent progress in establishing the properties of these models. We consider the dynamics for these types of models, and show that uniform attracting states can dominate the dynamics if there is a substantial weight-sign bias. We then show that it is possible to define dynamic thresholds for a variety of learning rules which can eliminate uniform attracting states for any value of the weightsign bias.
Type of Medium:
Electronic Resource
URL:
http://dx.doi.org/10.1007/BF01414946
Permalink