Stochastic Neural Learning Algorithm

Stochastic

Neural Learning

Algorithm

4.2.

Framework

of

the

algorithm

The

framework

of

the

learning

algorithm

based on

the

main

theorem

(23)

is

as

follows:

1.

Initialize

the

connection

weights

w^(t)

for

t

=

0,1,2,

.

.

.

,

T

-

1.

2.

For

learning

time

r

:=

1,2,.

. .

,

M

do

begin

Clear

the

update

increments

5w,,(t)

for

t

=

0,1,2,.

.

.

,

T

-

1.

For

running

time

r

:=

1,2,

. .

.

,

R

do

begin

Set

an

initial

state

vector

x(0)

=

xO.

Generate

the

Gaussian white

noise

vectors

at)

for

t

=

l,

2,

.

.

.

,

T.

Calculate

the

state

vectors

x(t)

for

t

=

1,2,.

. .

,

T.

Calculate

the

performance

functional

L[x(T)]

.

Calculate

new

update

increments

based on (23)

as

SwG(t)

:=

Swij(t)

-

pL[x(T)]f.[t

+

l)Sj(t)

for

t

=

0,1,2,.

. .

,T

-

1.

end;

Update

the

connection

weights

as

wij{t)

:=

W@

(t)

+

6wij{t)/R

for

t

=

0,1,2,

.

.

.

,

T

-

1.

Adjust

the

connection

weights based

on

the

given

value

range

[wmin,

wmax]

as

wij(t)

:=

max

{min

{~~(t),

wmai:}

,

Wmin}

for

t

=

0,1,2,

.

.

.

,

T

-

1.

Stop

if

the

given

terminal

condition

is

met.

end;

A set

of