Neurocomputing in Complex Domain - High Dimensional Neurocomputing: Growth, Appraisal and Applications - page 65

(Huber 1981) has good effects of quadratic and absolute EFs. The Huber EF is given
by

e i

E =

2 ;

if | e i | < c (3.19)

n

c

c
2

= | e i | − ; if | e i | ≥ c (3.20)

n

The complex Huber EF is defined to be

ʵ i ʵ i

| ʵ i | <

E =

2 ;

if c (3.21)

n

c

c
2

= | ʵ i | − ; if | ʵ i | ≥ c (3.22)

n

where n is the number of outputs and c is the tuning constant. A typical value for
c is 1.345. The Huber EF is defined piece-wise. The characteristic feature of the
functions involve the quadratic error on the one hand and an absolute error on the
other. The parameter, c in the definition is the point of demarcation to assign a domain
of operation for each EF (Fig. 3.13 ).

Fig. 3.13 a Plane and b Surface graphs for Huber error function