Ex parte Toomarian et al.

Board of Patent Appeals and Interferences

Oct 29, 1998

07969868 (B.P.A.I. Oct. 29, 1998)

Application for patent filed October 27, 1992. 1
THIS OPINION WAS NOT WRITTEN FOR PUBLICATION
The opinion in support of the decision being entered
today (1) was not written for publication in a law
journal and (2) is not binding precedent of the
Board.
Paper No. 16
UNITED STATES PATENT AND TRADEMARK OFFICE
____________
BEFORE THE BOARD OF PATENT APPEALS
AND INTERFERENCES
____________
Ex parte NIKZAD TOOMARIAN and JACOB BARHEN
____________
Appeal No. 96-0065
Application No. 07/969,8681
____________
ON BRIEF
____________
Before KRASS, MARTIN, and CARMICHAEL, Administrative Patent
Judges.
KRASS, Administrative Patent Judge.
DECISION ON APPEAL
This is a decision on appeal from the final rejection of
claims 1 through 20, all of the claims in the application.
Appeal No. 96-0065
Application No. 07/969,868
2
The invention is directed to a method and apparatus for
neural network training.
Representative independent claim 1 is reproduced as
follows:
1. A method of training a neural network so that a
neuron output state vector thereof obeys a set of forward
sensitivity equations over a finite learning period, said
method comprising:
defining first and auxiliary adjoint systems of equations
governing an adjoint function and an auxiliary adjoint
function, respectively, of said neural network;
setting said adjoint function to zero at the beginning of
said learning period and integrating said adjoint system of
equations forward in time over said learning period to produce
a first term of an indirect effect of a sensitivity gradient
of said neural network;
setting said auxiliary adjoint function to zero at the
end of said learning period and integrating said auxiliary
adjoint system of equations forward in time over said learning
period to produce a remaining term of said indirect effect;
computing a sum of said first and remaining terms, and
multiplying said sum by a learning rate; and
subtracting the product thereof from a current neuron
parameter vector to produce an updated neuron parameter
vector.
No references are relied on by the examiner.
Appeal No. 96-0065
Application No. 07/969,868
3
Claims 1 through 20 stand rejected under 35 U.S.C. § 101
as being directed to nonstatutory subject matter.
Reference is made to the brief and answer for the
respective positions of appellants and the examiner.
OPINION
We reverse.
On its face, the instant claimed invention is directed to
an operation of an algorithm for producing an updated, or
revised version of, a neuron parameter vector which is useful
for training a neural network. The instant claimed subject
matter therefore has practical utility. Even if the
mathematical algorithm by which the final result is reached
may be considered an abstract idea, that abstract idea is
clearly employed in a useful way. The transformation of data
through a series of mathematical calculations to produce the
updated version of the neuron parameter vector for training a
neural network constitutes a practical application of the
abstract idea or mathematical algorithm because it produces a
Appeal No. 96-0065
Application No. 07/969,868
4
useful, concrete and tangible result - the updated version of
the neuron parameter vector. See State Street Bank & Trust
Co., v. Signature Financial Group, Inc., 149 F.3d 1368, 47
USPQ2d 1596 (Fed. Cir. 1998).
The examiner even admits [answer, paper no. 15, paragraph
(9)] that the “invention is a neural network training
technique for solving problems to save computational time. It
has a practical application.” Clearly, the saving of
computational time is a useful, concrete and tangible result
and there is a practical application of the subject matter
recited in the instant claims. Thus, the claimed subject
matter constitutes patentable subject matter within the
meaning of 35 U.S.C. § 101.
The examiner appears to disregard the practical
application of the claimed subject matter because “all the
activity takes place inside the computer. The computer merely
provides an ‘indication’ when it has finished its computation”
[answer, paper no. 15, paragraph (9)]. Merely because the
claimed subject matter is practiced via a computer is not an
adequate basis for holding the subject matter to be
Appeal No. 96-0065
Application No. 07/969,868
5
nonstatutory under 35 U.S.C. § 101. Further, the
“indication” provided by the computer is the updated version
of the neuron parameter vector to be used for training the
neural network, a practical application. Thus, it is more
than merely an indication that the computation has been
completed.
The examiner’s decision is reversed.
Appeal No. 96-0065
Application No. 07/969,868
6
REVERSED
ERROL A. KRASS )
Administrative Patent Judge )
)
)
)
) BOARD OF PATENT
JOHN C. MARTIN ) APPEALS
Administrative Patent Judge ) AND
) INTERFERENCES
)
)
)
JAMES T. CARMICHAEL )
Administrative Patent Judge )
EAK/jlb
Appeal No. 96-0065
Application No. 07/969,868
7
National Aeronautics and
Space Administration
NASA Resident Office
4800 Oak Grove Drive
Pasadena, CA 91109
JoAnne
Appeal No. 96-0065
Application No. 07/969,868
APJ KRASS
APJ CARMICHAEL
APJ MARTIN
REVERSED
Prepared: August 13, 1999