Rough and neo-fuzzy neurons are two different ways of introducing semantic
structures in a neural network. Both have been shown to be useful in practi
cal applications. A rough neuron consists of an upper and a lower neuron. R
ough neurons can be used to effectively represent an interval or a set of v
alues. The rough neural networks provide more flexible architectures than e
xclusive interval-based neural networks. Neofuzzy neurons are used to elabo
rate on a value by partitioning the crisp value into fuzzy segments for pro
cessing by a neural network. Previous work has shown that fuzzy values can
be used to describe the difference in output of a rough neuron. This paper
provides a more comprehensive introduction to serial combinations of rough
and neofuzzy neurons in neurocomputing. The neofuzzy neurons are used to au
gment output of a rough neuron. On the other hand, rough neurons are shown
to be useful for extending the expressive powers of a neofuzzy neuron. The
first type of serial combination is termed a rough-fuzzy subnet. The latter
serial combination is called a fuzzy-rough subnet. This paper describes th
e architectures of fuzzy-rough and rough-fuzzy subnets. A discussion on pot
ential applications of the subnets is also provided along with examples. (C
) 2001 Elsevier Science B.V. All rights reserved.