This paper describes the SC-net system that has been developed to prov
ide expert systems capability augmented with learning in a hybrid conn
ectionist/symbolic approach. A distributed connectionist representatio
n of cells connected by links is used to represent symbolic knowledge.
Rules may be directly encoded in the connectionist network or learned
from examples. The learning method is a form of instance-based learni
ng in which some of the individual instances in the training set are e
ncoded by adding structure to the network and others cause modificatio
ns to biases in the network. Both continuous and nominal attributes ar
e directly represented in the network structure. A limited form of var
iables in the form of attribute value bindings on the right-hand side
of rules is supported. Relational comparators in the form of cell grou
ps are also supported. Relational comparators and attribute value stru
ctures are represented by groups of connected cells in the network. Th
e learning algorithm is presented and methods for providing generaliza
tion in an instance-based connectionist environment are presented. Emp
irical results are presented, which include learning in domains (fever
s and gems) that contain uncertainty and the well-known iris, and soyb
ean data sets together with a real world domain for semiconductor wafe
r fault diagnosis. The generalization ability of the learned network i
s shown to be good in several domains including iris. The system is sh
own to compare favorably with a nonneural instance-based learning algo
rithm IBL.