Weighted pattern learning for neural networks
A method of training a neural net includes receiving a plurality of sets of data, each set representative of a plurality of inputs to the neural net and a resulting at least one output from the neural net and calculating a plurality of network weights for the neural network based on the received plurality of sets of data. Calculating the plurality of network weights including attributing greater weight in the calculation to at least one set of the plurality of sets of data than at least one other set of the plurality of sets of data.
1. A computer system comprising:
a processor; and
a program stored in computer-readable media and executable by the processor, the computer program operable to:
receive a plurality of sets of data, each set representative of a plurality of inputs to a neural net and a resulting at least one output from the neural net;
calculate a plurality of network weights for the neural net based on the received plurality of sets of data, including attributing greater weight in the calculation to at least one set of the plurality of sets of data than at least one other set of the plurality of sets of data; and
train the neural net using the calculated plurality of network weights.
2. The computer system of claim 1 , wherein the computer program is operable to calculate a plurality of network weights for the neural net by calculating a least squares approximation to a plurality of simultaneous equations based on the plurality of sets of data.
3. The computer system of claim 2 , wherein the computer program is operable to attribute a pattern weight to each set of the plurality of sets of data.
4. The computer system of claim 3 , wherein the computer program is further operable to attribute greater weight in the calculation of the plurality of network weights to at least one set of the plurality of sets of data by calculating a least squares approximation to the following equations:
[
u
1
y
1
u
2
y
2
⋮
u
P
y
P
]
=
y
u
=
F
u
w
=
[
u
1
F
1
u
2
F
2
⋮
u
P
F
P
]
w
=
[
u
1
f
1
(
x
1
)
u
1
f
2
(
x
1
)
⋯
u
1
f
J
(
x
1
)
u
2
f
1
(
x
2
)
u
2
f
2
(
x
2
)
⋯
u
2
f
J
(
x
2
)
⋮
⋮
⋰
⋮
u
P
f
1
(
x
P
)
u
p
f
2
(
x
P
)
⋯
u
P
f
J
(
x
P
)
]
w
where
w=the plurality of network weights,
x 1 through x p =the plurality of inputs in the plurality of sets of data,
u 1 through u p =a plurality of pattern weights each associated with a respective one of the plurality of sets of data,
f 1 through f p =a plurality of functions, and
y 1 through y p =the at least one output in the plurality of sets of data.
5. The computer system of claim 4 , wherein the computer program is further capable of calculating a least squares approximation by solving the following system of equations
∑
p
u
p
2
F
p
T
y
p
=
F
u
T
y
u
=
F
u
T
F
u
w
=
(
∑
p
u
p
2
F
p
T
F
p
)
w
where
[
F
u
T
y
u
]
j
=
∑
p
=
1
P
u
p
2
f
j
(
x
p
)
y
p
,
where
j
=
1
,
…
,
J
and
[
F
u
T
F
u
]
ij
=
∑
p
=
1
P
u
p
2
f
i
(
x
p
)
f
j
(
x
p
)
,
where
both
i
and
j
=
1
,
…
,
J
.
6. The computer system of claim 3 , wherein the computer program is further operable to calculate the pattern weights.
7. The computer system of claim 2 , wherein the computer program is further operable to calculate a least squares approximation by calculating an orthogonal least squares approximation.
8. The computer system of claim 1 , wherein the computer program is further operable to store the trained neural net.
9. A system comprising:
means for receiving a plurality of sets of data, each set representative of a plurality of inputs to the neural net and a resulting at least one output from the neural net;
means for calculating a plurality of network weights for the neural net based on the received plurality of sets of data, including means for attributing greater weight in the calculation to at least one set of the plurality of sets of data than at least one other set of the plurality of sets of data; and
means for training the neural net using the calculated plurality of network weights.
10. The system of claim 9 , wherein the means for calculating a plurality of network weights comprises means for calculating a least squares approximation to a plurality of simultaneous equations based on the plurality of sets of data.
11. The system of claim 10 , wherein the means for attributing greater weight in the calculation comprises means for attributing a pattern weight to each set of the plurality of sets of data.
12. The system of claim 9 , further comprising means for storing the trained neural net.