The Procter & Gamble Company

Patent Trials and Appeals Board

Jan 26, 2021

2020005322 (P.T.A.B. Jan. 26, 2021)

UNITED STATES PATENT AND TRADEMARK OFFICE
UNITED STATES DEPARTMENT OF COMMERCE
United States Patent and Trademark Office
Address: COMMISSIONER FOR PATENTS
P.O. Box 1450
Alexandria, Virginia 22313-1450
www.uspto.gov
APPLICATION NO. FILING DATE FIRST NAMED INVENTOR ATTORNEY DOCKET NO. CONFIRMATION NO.
15/465,292 03/21/2017 Faiz Feisal SHERMAN 14198 6048
27752 7590 01/26/2021
THE PROCTER & GAMBLE COMPANY
GLOBAL IP SERVICES
CENTRAL BUILDING, C9
ONE PROCTER AND GAMBLE PLAZA
CINCINNATI, OH 45202
EXAMINER
GEORGALAS, ANNE MARIE
ART UNIT PAPER NUMBER
3625
NOTIFICATION DATE DELIVERY MODE
01/26/2021 ELECTRONIC
Please find below and/or attached an Office communication concerning this application or proceeding.
The time period for reply, if any, is set in the attached communication.
Notice of the Office communication was sent electronically on above-indicated "Notification Date" to the
following e-mail address(es):
centraldocket.im@pg.com
mayer.jk@pg.com
pair_pg@firsttofile.com
PTOL-90A (Rev. 04/07)
UNITED STATES PATENT AND TRADEMARK OFFICE
BEFORE THE PATENT TRIAL AND APPEAL BOARD
Ex parte FAIZ FEISAL SHERMAN,
SHANNON CHRISTINE WEITZ, and JUN XU
Appeal 2020-005322
Application 15/465,292
Technology Center 3600
Before MURRIEL E. CRAWFORD, PHILIP J. HOFFMANN, and
BRADLEY B. BAYAT, Administrative Patent Judges.
BAYAT, Administrative Patent Judge.
DECISION ON APPEAL
STATEMENT OF THE CASE
Appellant1 appeals under 35 U.S.C. § 134(a) from the Examiner’s
decision to reject claims 1–19, which constitute all pending claims in the
application. We have jurisdiction under 35 U.S.C. § 6(b).
We REVERSE.

1 We use the term “Appellant” to refer to “applicant” as defined in 37 C.F.R.
§ 1.42. Appellant identifies the real party in interest “is The Procter &
Gamble Company of Cincinnati, Ohio.” Appeal Br. 1.
Appeal 2020-005322
Application 15/465,292

2
CLAIMED SUBJECT MATTER
Appellant’s invention is “related to utilizing a convolutional neural
network [(CNN)] to identify features of an image and utilize the features to
recommend products.” Spec. 1:5–8.
Independent claim 1, reproduced below, is illustrative of the claimed
subject matter:
1. A system for providing customized product recommendations
to a user, comprising:
a) an image capture device;
b) an image of a user captured by the image capture
device; and
c) a computing device coupled to the image capture
device, wherein the computing device includes a memory
component that stores logic that causes the system to
(i) analyze the captured image of the user using a
convolutional neural network to predict a
user’s age, wherein predicting the user’s age
includes identifying a portion of skin in the
captured image that contributes to the
predicted age of the user,
(ii) utilize at least one pixel from the portion of skin
identified in the captured image to generate a
heat map that identifies a region of the image
that contributes to the user's predicted age,
(iii) display the heat map to the user on a display
device visible to the user, and
(iv) recommend a product for the user to apply to a
region of skin for achieving a target skin age.
Appeal 2020-005322
Application 15/465,292

3
REJECTIONS2
Claims 1–19 are rejected under 35 U.S.C. § 112(a) as failing to
comply with the written description requirement.
Claims 1–19 are rejected under 35 U.S.C. § 112(b) as indefinite.
Claims 1, 3, 5, and 7–9 are rejected under 35 U.S.C. § 103 as
unpatentable over Goodman (US 8,625,864 B2, issued Jan. 7, 2014) and
Rowe et al., (US 7,362,886 B2, issued Apr. 22, 2008) (“Rowe”).
Claims 4 and 6 are rejected under 35 U.S.C. § 103 as unpatentable
over Goodman, Rowe, and Yoo et al. (US 9,928,410 B2, issued Mar. 27,
2018) (“Yoo”).
Claims 10–12 and 15–19 are rejected under 35 U.S.C. § 103 as
unpatentable over Goodman, Rowe, and Peyrelevade (US 7,437,344 B2,
issued Oct. 14, 2008).
Claim 14 is rejected under 35 U.S.C. § 103 as unpatentable over
Goodman, Rowe, Peyrelevade, and Yoo.
OPINION
Rejection under § 112(a) for Lack of Written Description Support
The Examiner finds that the claim language in limitation c) (i), which
recites a system to analyze the captured image of the user using a
convolutional neural network to predict a user’s age “does not conform to
the disclosure in such a manner that one of ordinary skill in the art would
recognize as being adequately described as the invention or as subject matter
which Applicants actually had possession of at the time of the invention.”

2 The Examiner withdraws rejections under § 103 of claims 2 and 13.
Answer 3.
Appeal 2020-005322
Application 15/465,292

4
Final Act. 6. According to the Examiner, Appellant’s “disclosure does not
reveal the manner in which a portion of skin that contributes to the predicted
age of the user is identified” and “no algorithm as to how the trouble spots
are identified is provided.” Id. at 7.
Appellant argues that the “application discloses several techniques for
training a CNN to identify portions of skin that contribute to the age of the
user from a digital image” and cites pages 6 and 7 of the Specification in
support. Appeal Br. 3–4. Appellant also contends, citing Yoo at column 3,
line 39 to column 4, lines 49, that “convolutional neural networks and
methods of training them are known.”3 Id. at 4.
In response to Appellant’s arguments, the Examiner states that the
Specification is “directed to training the CNN but does not relate specifically
to the identification of ‘trouble spots’ . . . [and] does not disclose training the
CNN as to how to identify trouble spots in the captured images, using either
SGD or supervised learning.” Answer 6.
Appellant’s description as filed is considered to be adequate, unless or
until sufficient evidence or reasoning to the contrary has been presented by
the Examiner to rebut the presumption. See, e.g., In re Marzocchi, 439 F.2d
220, 224 (CCPA 1971). As such, the Examiner has the initial burden of
presenting, by a preponderance of evidence, why a person skilled in the art
would not recognize in Appellant’s disclosure a description of the invention
defined by the claims. In re Wertheim, 541 F.2d 257, 263 (CCPA 1976).
The Specification describes that a convolutional neural network
(“CNN”) “functions as an in silico skin model, to predict the skin age of a

3 The Examiner cites Yoo in the rejection of claims 4, 6, and 14 under § 103.
Appeal 2020-005322
Application 15/465,292

5
user by analyzing a captured image of the skin of the user (e.g., facial skin).”
Spec. 5:24–26. The Specification describes that pre-processing can be used
before training the CNN, including having images “aligned,” “cropped,” and
normalized/standardized, such as to contrast, and “masked . . . to minimize
the influence of other features like hair, neck and other undesired objects in
the image.” Id. at 6:1–10. According to the Specification, the CNN may be
“trained using a deep learning technique . . . much in the same way as a
mammalian visual cortex learns to recognize important features in an
image.” Id. at 6:24–26. A CNN may be “trained by analyzing images in
which the age of the person in the image is predetermined.” Id. at 6:33–7:1.
The “CNN analyzes the image and identifies portions of skin in the image
that contribute to the predicted age of the user (‘trouble spots’). The CNN
then uses the trouble spots to predict the age of the user.” Id. at 7:8–10.
Moreover, “FIG. 16 depicts a flowchart for training a convolutional neural
network for identifying a feature from an image.” Id. at 4:9–10.
Given the description of training a CNN to recognize “portions of skin
in the image that contribute to the predicted age” and the flowcharts in
Figures 16–18, which are not fully considered and adequately evaluated, we
find the Examiner has failed to meet the initial burden by simply asserting
that “no algorithm as to how the trouble spots are identified is provided”
(Final Act. 7); indeed, an algorithm may be expressed in any understandable
terms (e.g., a mathematical formula, in prose, or as a flowchart). The
Examiner provides no explanation or reasoning as to why those disclosures
do not convey enough detail to show Appellant had possession of the
invention at the time of filing.
Appeal 2020-005322
Application 15/465,292

6
Furthermore, “[t]he ‘written description’ requirement must be applied
in the context of the particular invention and the state of the knowledge.”
Capon v. Eshhar, 418 F.3d 1349, 1358 (Fed. Cir. 2005). “Since the law is
applied to each invention in view of the state of relevant knowledge, its
application will vary with differences in the state of knowledge in the field
and differences in the predictability of the science.” Id. at 1357. Here,
Appellant acknowledges that “convolutional neural networks and methods
of training them are known” and cites Yoo’s description of “training the one
or more CNNs to perform the recognition with respect to both an image
identification (ID) and at least one image attribute. The at least one attribute
may include at least one of: . . . an age corresponding to the face region.”
Yoo 3:47–55; see also Yoo at 3:55–4:39. The Examiner, however,
dismisses Appellant’s acknowledgment of the state of relevant knowledge in
the field and asserts that “[w]hether the Yoo reference discloses methods of
training a CNN is irrelevant to the inquiry of whether Appellant had
possession.” Answer 6. We disagree because a specification does not need
to reiterate information that is “already known in the field” to satisfy the
written description requirement. Capon, 418 F.3d at 1361. Because the
Examiner has not met the burden to demonstrate a lack of possession, we do
not sustain the rejection under § 112(a) for lacking written description.
Rejection under § 112(b) as Indefinite
Claims 1–19
The Examiner determines “the disclosure does not disclose any
meaningful structure/algorithm as to how the portion of skin that contributes
to the predicted age of the user is identified.” Final Act. 8. This, according
to the Examiner, is “because the particular method of identifying the portion
Appeal 2020-005322
Application 15/465,292

7
of skin that contributes to the predicted age of the user is not disclosed.”
Answer 7. Without providing further explanation or reasoning, the
Examiner “refers Appellants to the above discussion regarding the rejection
under 35 USC 112(a).” Id. at 8.
Appellant essentially argues that the claim language is definite
because the Specification indicates, through description and flow charts,
how a CNN is trained to perform the claimed function, thus appraising of the
scope of the claims. Appeal Br. 4–5; see also Reply Br. 3 (“The
specification discloses the use of a convolutional neural network, which is a
well-known algorithm. Accordingly, the skilled artisan, when reading the
claims in light of the specification and the prosecution history, would be
informed, with reasonable certainty, about the scope of the invention”).
The test for definiteness under 35 U.S.C. § 112(b) is whether “those
skilled in the art would understand what is claimed when the claim is read in
light of the specification.” Orthokinetics, Inc. v. Safety Travel Chairs, Inc.,
806 F.2d 1565, 1576 (Fed. Cir. 1986) (citations omitted). “[A] claim is
indefinite when it contains words or phrases whose meaning is unclear.” In
re Packard, 751 F.3d 1307, 1322 (Fed. Cir. 2014).
Here, in response to Appellant’s arguments, the Examiner refers only
to the analysis under § 112(a) (Ans. 8), and fails to provide reasoning as to
why the claim language is rendered indefinite under § 112(b). The written
description requirement is separate and distinct from indefiniteness because
a claim satisfies the definiteness requirement of § 112 “[i]f one skilled in the
art would understand the bounds of the claim when read in light of the
specification.” Exxon Research & Eng’g Co. v. United States, 265 F.3d
1371, 1375 (Fed. Cir. 2001). The Examiner fails to demonstrate that an
Appeal 2020-005322
Application 15/465,292

8
artisan would not understand the scope of what is claimed upon reviewing
pages 6–7 of the Specification and the flowcharts of Figures 16–18.
Accordingly, we do not sustain the rejection of claims 1–19 as
indefinite.
Claim 6
Dependent claim 6 recites:
wherein training the convolutional neural network includes data
augmentation that is utilized to create additional samples from
the training image, wherein the data augmentation includes at
least one of the following: randomly zoom in on the training
image, zoom out of the training image, perform a random
rotation of the image in a clockwise direction, perform a random
rotation of the image and in a counter clockwise direction,
randomly crop the image, randomly change saturation of the
image, randomly change exposure of the training image, and
utilize vertical dropout to randomly dropout a column of pixels
of the training image.

The Examiner rejects claim 6 as indefinite because:
The use of the word “randomly” is unclear. Computers do not
“randomly” do anything. A computer is controlled by
algorithms so that even a seemingly “random” act is actually
the result of an algorithm that starts with a seed value. The seed
value determines the result of the “random number generating”
algorithm. This is called a pseudo-random number generator.

Final Act. 8–9; Answer 8.
Appellant asserts the Examiner acknowledges “that it is known that a
seemingly random process in a computer, such as a random number
generator, is controlled by an algorithm to create a ‘pseudo-random’ effect.”
Appeal Br. 5. Appellant argues that “since the skilled artisan would
recognize what is known in the art, they would recognize the metes and
Appeal 2020-005322
Application 15/465,292

9
bounds of claim 6 with regard to the use of the word ‘random.’” Id. In other
words, “the skilled artisan would also construe the term ‘random,’ as recited
in the computer-implemented system of pending claim 6, to mean the so-
called ‘pseudo-random’ effect” as interpreted by the Examiner. Reply Br. 4.
The Examiner responds that “the word ‘random’ has an ordinary and
customary meaning in the art, namely to refer to an occurrence in which
each possible outcome has an equal probability of occurring, e.g., the
outcome of a fair coin toss, the outcome of a fair dice roll.” Answer 10
(“The fields of probability and statistics, for example, rely on this definition
of random.”). According to the Examiner: “If Appellants wish to be their
own lexicographer and define ‘random’ as meaning ‘pseudo-random’, per
the MPEP, Appellants ‘must clearly set forth a special definition of a claim
term in the specification.’” Answer 10.
We are persuaded of Examiner error.
The Specification discloses that “data augmentation may be
performed to create additional samples from an inputted image. The
additional samples are used to train the CNN to tolerate variation in input
images.” Spec. 6:11–12. “The additional samples generated by data
augmentation can also force the CNN to learn to rely on a variety of features
for skin age prediction, rather than one particular feature, and may prevent
over-training of the CNN.” Id. at 6:16–18. The Examiner interprets that
what Appellant meant by “random” is “pseudo-random” in the context of the
invention but contends that the meaning of “random” is limited to the “fields
of probability and statistics” and improper in the context of claim 6.
However, the Examiner has not shown why, in the context of providing
additional images to train a CNN, any difference between random and
Appeal 2020-005322
Application 15/465,292

10
pseudo-random would matter, or that the skilled artisan would be confused
as to the use of the term “random” in light of the teachings of the
Specification. We find the Examiner’s interpretation of the term “random”
in the “field of probability and statistics” is unreasonable and inconsistent
with Appellant’s disclosure in the context of the invention. Therefore, we
do not sustain the separate indefiniteness rejection of claim 6.
Rejection under § 103 over Goodman and Rowe
Claims 1–3, 5, and 7–9
In rejecting claim 1, the Examiner finds Goodman discloses the
limitation “(i) analyze the captured image of the user,” but does not disclose
“using a convolutional neural network to predict a user's age,” which
Examiner finds in Rowe, at column 6, lines 12–40. Final Act. 11–12.
Appellant argues that “there is no disclosure in Rowe of using a
trained neural network to predict a user’s age” because “the system of Rowe
determines a user’s age by simply calculating the difference between two
dates.” Appeal Br. 8.
The Examiner indicates that, based on the claim’s recitation of
“wherein predicting the user’s age includes identifying a portion of skin in
the captured image that contributes to the predicted age of the user,” Rowe
discloses that the age difference calculator reads the pixel data of
each image identified by the face recognizer. The age difference
calculator reads the pixel data of each image (i.e., the pixels in
an image of a face represent a portion of skin in the captured
image) which includes the date the image was recorded overlaid
on each pixel. The age difference calculator than calculates
(predicts/determines) the age of the subject person by subtracting
the date of birth of the subject person from the date the image
was recorded (included in the pixel data).
Appeal 2020-005322
Application 15/465,292

11
Answer 11–12. We are persuaded by Appellant’s argument.
The Examiner’s finding in Rowe does not correspond to the disputed
claimed subject matter. Claim 1 recites that the system is caused to “analyze
the captured image of the user using a convolutional neural network to
predict a user’s age.” The Specification defines a convolutional neural
network as “a type of feed-forward artificial neural network where the
individual neurons are tiled in such a way that they respond to overlapping
regions in the visual field.” Spec. 5:8–10. The rejection appears to ignore
the requirement of using a CNN because Rowe utilizes an age difference
calculator for predicting age “by subtracting the date of birth of the subject
person from the image recording date.” Rowe 6:26–36. Due to this
shortcoming in Rowe, we do not sustain the rejection of independent claim
1, and claims 3, 5, and 7–9 dependent thereon.
Rejection under § 103 over Goodman, Rowe, and Yoo
Claims 4 and 6
As for dependent claims 4 and 6, the Examiner does not establish that
the cited portions of Yoo remedy the deficiency of Rowe as to independent
claim 1 discussed above. Therefore, we do not sustain the rejection of
claims 4 and 6.
Rejection under § 103 over Goodman, Rowe, and Peyrelevade
Claims 10–12 and 15–19
The Examiner does not establish that Peyrelevade remedies the
deficiency of Rowe to disclose using a convolutional neural network to
predict a user’s age. As to independent claim 11, the Examiner specifically
relies on the same portions of Goodman and Rowe for using a CNN to
determine a user’s age. Final Act. 22–23. The rejection suffers from the
Appeal 2020-005322
Application 15/465,292

12
same shortcomings as the rejection of claim 1. For these reasons, we do not
sustain the obviousness rejection of claims 10–12 and 15–19.
Rejection under § 103 over Goodman, Rowe, Peyrelevade, and Yoo
Claim 14
The Examiner does not establish that either Peyrelevade or Yoo
remedies the deficiency of Rowe discussed above to disclose using a
convolutional neural network to predict a user’s age. For this reason, we do
not sustain the obviousness rejection of claim 14.
CONCLUSION
The Examiner’s decision to reject claims 1–19 is reversed.
DECISION SUMMARY
Claim(s)
Rejected
35 U.S.C. § References/Basis Affirmed Reversed
1–19 112(a) Written Description 1–19
1–19 112(b) Indefiniteness 1–19
1, 3, 5, 7–9 103 Goodman, Rowe 1–3, 5, 7–9
4, 6 103 Goodman, Rowe, Yoo 4, 6
10–12, 15–
19
103 Goodman, Rowe,
Peyrelevade
10–12, 15–
19
14 103 Goodman, Rowe,
Peyrelevade, Yoo
14
Overall Outcome 1–19

REVERSED