Neurocomputing in Complex Domain - High Dimensional Neurocomputing: Growth, Appraisal and Applications - page 74

if ∂

E ( t − 1 ) )

∂ (w) <

∂ E ( t

× 0 and E ( t )> E ( t − 1 )

∂ (w)

then (w( t + 1 )) = (w( t )) − (w( t − 1 )) (3.30)

if ∂

( − ) )

∂ (w) <

∂ (

E t 1 E t

× ( )> ( − )

0 and E t E t 1

∂ (w)

then (w( t + 1 )) = (w( t )) − (w( t − 1 )) (3.31)

The initialization parameters are same as in case of conventional C RPROP and
the change is only in the case where partial derivative changes its size. The resulting
algorithm shows better performance as compare to C RPROP. Let

w lm be the weight
from l th neuron in a layer to m th neuron in next layer in a neural network. Let t = 1
and 0

<μ − <μ + < .

1 2. The pseudo code for this C - i RPROP algorithm is given

below:

)) = 0 and ∂

)
∂ (w lm )

E ( t − 1

= ∂

)
∂ (w lm )

E ( t − 1

∀ l , m : ( lm ( t )) = ( lm ( t = 0

( max ) = ( max ) = max and ( min ) = ( min ) = min

Repeat {
calculate

)
∂ (w lm )

∂ E ( t

and ∂

)
∂ (w lm )

E ( t

for all weights and biases

For real part of weight:

if ∂ 0 then

)
∂ (w lm )

E ( t − 1 )

∂ (w lm ) >

∂ E ( t

×

)) × μ + , ( max ))

{ ( lm ( t )) = min ( ( lm ( t − 1

sign

)
∂ (w lm )

∂ E ( t

(w lm ( t )) =− ( ij ( t ))

(w lm ( t + 1 )) = (w lm ( t )) + (w lm ( t )) }

if ∂ 0 then

)
∂ (w lm )

E ( t − 1 )

∂ (w lm ) <

∂ E ( t

×

)) × μ − , ( min )) { ( lm ( t )) = max ( ( lm ( t − 1

( ( )> ( − ))

if E t E t 1 then

)
∂ (w lm ) =

∂ E ( t

(w lm ( t + 1 )) = (w lm ( t )) − (w lm ( t − 1 )) and 0 }