Information Technology Reference
In-Depth Information
sign of the weight increment in eq. (3.29):
x
(
t
)
−
i
α (
t
)
y
j
(
t
)
w
w
j
(
t
+
1
)
=
w
j
(
t
)
+
y
i
(
t
) w
i
(
t
)
T
(
t
) w (
t
)
>
j
x
(
t
)
−
i
T
j
w
(
t
)
j
y
i
(
t
) w
i
(
t
)
w
j
(
t
)
>
−
(3.39)
w
T
(
t
) w (
t
)
This network has not been used in this topic and will be the subject of future work.
Because of its similarity to the PSA LUO learning law, the same convergence
theorem applies, but we think that this convergence theorem can be extended to a
more general constraint for the weight moduli because it is based only on upper
bounds for the averaging ODE and subsequent formulas.
Search WWH ::
Custom Search