Baidu USA, LLC

Patent Trials and Appeals Board

Apr 21, 2021

2020004875 (P.T.A.B. Apr. 21, 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.
15/226,249 08/02/2016 Yi Zhen 28888-2017
(BN160513USN7)
1668
119276 7590 04/21/2021
BAIDU USA LLC
c/o NORTH WEBER & BAUGH LLP
3260 Hillview Avenue
Palo Alto, CA 94304
EXAMINER
ARAQUE JR, GERARDO
ART UNIT PAPER NUMBER
3689
NOTIFICATION DATE DELIVERY MODE
04/21/2021 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):
bbaugh@northweber.com
docket1@northweber.com
docket2@northweber.com
PTOL-90A (Rev. 04/07)
UNITED STATES PATENT AND TRADEMARK OFFICE
___________

BEFORE THE PATENT TRIAL AND APPEAL BOARD
___________

Ex parte YI ZHEN, HONGLIANG FEI,
SHULONG TAN, and WEI FAN
____________

Appeal 2020-004875
Application 15/226,249
Technology Center 3600
____________

Before ANTON W. FETTING, JOSEPH A. FISCHETTI, and
NINA L. MEDLOCK, Administrative Patent Judges.

FETTING, Administrative Patent Judge.

DECISION ON REQUEST FOR REHEARING

Appeal 2020-004875
Application 15/226,249

2

STATEMENT OF CASE
This is a decision on rehearing in Appeal No. 2020-004875. We have
jurisdiction under 35 U.S.C. § 6(b).
Requests for Rehearing are limited to matters misapprehended or
overlooked by the Board in rendering the original decision, or to responses
to a new ground of rejection designated pursuant to § 41.50(b). 37 C.F.R.
§ 41.52.
ISSUES ON REHEARING
Appellant raises the sole issue of the new ground we raised of patent
eligibility in the Request for Rehearing.
ANALYSIS
We found in our decision that the rejection of claims 1–11 and 16–20
over art was improper, but we introduced a new ground of rejection of those
same claims under 35 U.S.C. § 101 as directed to a judicial exception
without significantly more. Decision 18–19.
We are not persuaded by Appellant’s argument that “for a machine
learning model to be trained, vast amounts of good quality training datasets
must be generated.” Request 1. The claims recite no lower limit on the
amount of training data. The Specification describes no such lower limit.
We are not persuaded by Appellant’s argument that “[g]enerating
such information is a complex technical problem that has plagued machine
learning from its inception and continues to limit the development of
machine learning applications in some fields.” Id. The claims recite no
technological implementation details for such generation. As we
determined, claim 1 at issue amounts to nothing significantly more than an
Appeal 2020-004875
Application 15/226,249

3

instruction to apply estimating demand by selecting a model based on
evaluations of entity representations using some unspecified, generic
computer. Decision 14.
We are not persuaded by Appellant’s argument that
[o]btaining datasets is only the start of the machine learning
model development pipeline. The correct elements of the data,
or features, must be selected and converted into a format that is
useable as an input for a model. Again, significant portions of
the current Application deals with extracting features and
converting them into representations that retain semantic
meaning but are digitally usable by models.
Request 2. The claims recite no technological implementation details for
selecting and converting into a format that is useable as an input for a model.
Appellant invites us to look to Figures 5 and 6 for support. Id. Figure 5 is a
flow diagram containing 4 blocks of highly general functional descriptions
that mirror claim 1 limitations 4–6. Figure 6 is a simple data flow diagram
that labels the various data used in the claims. Neither figure nor any of its
supporting text provides technological implementation details.
We are not persuaded by Appellant’s argument that
aspects of the current Application disclose processes by way a
set of models are trained. Following training, each trained
model may then be evaluated using an evaluation dataset. The
evaluation dataset represents tests for the model to measure
how well it correctly predicts outputs—each model operates on
inputs selected from the evaluation set, the model performs its
complex computations and outputs corresponding predictions.
Based upon the accuracy of the outputted predictions, a model
may be selected as being the best one and may be deployed for
its intended inference purpose.
Appeal 2020-004875
Application 15/226,249

4

Request 2–3. The claims recite no technological implementation details for
training a model. The claims recite no more than the conceptual ideas
presented in the above argument.
At that level of generality, the claims do no more than
describe a desired function or outcome, without providing any
limiting detail that confines the claim to a particular solution
to an identified problem. The purely functional nature of the
claim confirms that it is directed to an abstract idea, not to a
concrete embodiment of that idea.
Affinity Labs of Texas, LLC v. Amazon.com Inc., 838 F.3d 1266, 1269 (Fed.
Cir. 2016).
We are not persuaded by Appellant’s argument that
the Board Decision requires that a person can, as a mental
process, practically perform the following. First, the person
must identify one or more entities within each labeled record by
applying to a labeled record one or more techniques correlating
to that label for identifying an entity or entities. Then, while
keeping track of all of this information, add more complex
information by generating, from the labeled records, entity
features and relationships between two or more entities. And
then, for each entity, the person must convert the entity features
and relationships between two or more entities into a vectorized
representation of the entity. These processes alone are beyond
the practical capabilities of any person, even with paper and
pen. But, added to these processes, the person must use the
vectorized data of entities and resource data to train not one but
a set of models using the dataset. Training even one model
requires vast numbers of complex calculations to be performed.
If even a single computational error is made, that error will
propagate through the system and through multiple iterations,
rendering the training process futile. And, the heavy
computation load and the requirement to keep in memory
practically innumerable parameter values and intermediate
values does not end there. Claim 1 continues with the
requirement of evaluating each model of the set of models
using an evaluation set of data.
Appeal 2020-004875
Application 15/226,249

5

Request 4. Appellant presents no evidence that “[t]hese processes alone are
beyond the practical capabilities of any person, even with paper and pen.”
Id. Speed and data volume are insufficient to confer eligibility.
[W]e do not rely on the pen and paper test to reach our holding
of patent eligibility in this case. At the same time, we note that,
in viewing the facts in FairWarning’s favor, the inability for the
human mind to perform each claim step does not alone confer
patentability. As we have explained, “the fact that the required
calculations could be performed more efficiently via a computer
does not materially alter the patent eligibility of the claimed
subject matter.”
FairWarning IP, LLC v. Iatric Sys., Inc., 839 F.3d 1089, 1098 (Fed. Cir.
2016) (citations omitted). Further, the claims recite no minimum data set
size. The argued vast numbers of complex calculations are not recited in the
claims. The claims also recite no technological implementation details for
the evaluation step.
Appellant next argues that the present claims are patent eligible
because they are similar to the USPTO’s Example 39 in the “Subject Matter
Eligibility Examples: Abstract Ideas,” published January 7, 2019 (“2019
Eligibility Examples”). Request 5–6. More particularly, Appellant argues
[b]oth claims discuss creating dataset and both claims discuss
training models. Some of the key differences is that claims of
the current Application involve training a set of models, not just
a single neural network, and involve evaluating the set of
models using an evaluation dataset. Each of those claim
elements require even more computation processing than
Example 39 that cannot practically be performed in the human
mind.
Request 5–6.
Appeal 2020-004875
Application 15/226,249

6

The difficulty with Appellant’s argument is that Example 39 was
deemed patent eligible because it provided training of a neural network for
facial detection. See 2019 Eligibility Examples, 8–9. The guidelines said
expanded training set is developed by applying mathematical
transformation functions on an acquired set of facial images.
These transformations can include affine transformations, for
example, rotating, shifting, or mirroring or filtering
transformations, for example, smoothing or contrast reduction.
The neural networks are then trained with this expanded
training set using stochastic learning with backpropagation
which is a type of machine learning algorithm that uses the
gradient of a mathematical loss function to adjust the weights of
the network. Unfortunately, the introduction of an expanded
training set increases false positives when classifying non-facial
images. Accordingly, the second feature of applicant’s
invention is the minimization of these false positives by
performing an iterative training algorithm, in which the system
is retrained with an updated training set containing the false
positives produced after face detection has been performed on a
set of non-facial images. This combination of features provides
a robust face detection model that can detect faces in distorted
images while limiting the number of false positives.
Thus, Example 39 addressed technological difficulties related to
analyzing graphic images to identify and analyze facial images within them.
Appellant has neither identified nor demonstrated that the present claims
provide such image analysis. Instead, the Specification says the claims are
directed to using information handling systems to estimate demand for
healthcare resources (Specification para. 1), which would include
conventional textual data. Such analysis of textual data is the bread and
butter of mental processes.
Appellant also attempts to analogize the claims to those involved in
McRO, Inc. v. Bandai Namco Games America Inc., 837 F.3d 1299 (Fed. Cir.
Appeal 2020-004875
Application 15/226,249

7

2016). Request 7. In McRO, the court held that, although the processes
were previously performed by humans, “the traditional process and newly
claimed method . . . produced . . . results in fundamentally different ways.”
FairWarning, 839 F.3d at 1094 (differentiating the claims at issue from
those in McRO). In McRO, “it was the incorporation of the claimed rules
not the use of the computer, that improved the existing technology process,”
because the prior process performed by humans “was driven by subjective
determinations rather than specific, limited mathematical rules.” 837 F.3d at
1314 (internal quotation marks, citation, and alterations omitted). In
contrast, the claims of the instant application merely implement an old
practice of using decision criteria in making decisions in a new environment.
Merely pigeon holing the objects of decision making in tiers to aid decision
making is both old and itself abstract.
The claims in McRO were not directed to an abstract idea, but
instead were directed to “a specific asserted improvement in
computer animation, i.e., the automatic use of rules of a
particular type.” We explained that “the claimed improvement
[was] allowing computers to produce ‘accurate and realistic lip
synchronization and facial expressions in animated characters’
that previously could only be produced by human animators.”
The claimed rules in McRO transformed a traditionally
subjective process performed by human artists into a
mathematically automated process executed on computers.
FairWarning, 839 F.3d at 1094.
We are not persuaded by Appellant’s argument that “the analysis [in
Step 2B] fails to apprehend the claim elements individually and the claims
as a whole.” Request 8. Our analysis explicitly bifurcated into both
individual elements and the claim as a whole. Decision 16–17.
Appeal 2020-004875
Application 15/226,249

8

We are not persuaded by Appellant’s argument that the Decision
“fails to appreciate the uniqueness of claim elements and the claims as a
whole. Proof of significance of claim elements is demonstrably shown in
the record of the current Application in which at least two claim elements
were not found following searches.” Request 8. Novelty is not at issue.
“[A] claim for a new abstract idea is still an abstract idea. The search for a §
101 inventive concept is thus distinct from demonstrating § 102 novelty.”
Synopsys, Inc. v. Mentor Graphics Corp., 839 F.3d 1138, 1151 (Fed. Cir.
2016).
We are not persuaded by Appellant’s argument that “it invented at
least the above-identified claim elements, among other claim elements, in
the current Application.” Request 9. This is in response to our
determination that
[u]sing a computer for modifying, generating, analyzing, and
processing data amounts to electronic data query and
retrieval—one of the most basic functions of a computer. All
of these computer functions are generic, routine, conventional
computer activities that are performed only for their
conventional uses. See Elec. Power Grp. LLC v. Alstom S.A.,
830 F.3d 1350, 1353 (Fed. Cir. 2016). See also In re Katz
Interactive Call Processing Patent Litig., 639 F.3d 1303, 1316
(Fed. Cir. 2011) (“Absent a possible narrower construction of
the terms ‘processing,’ ‘receiving,’ and ‘storing,’ . . . those
functions can be achieved by any general purpose computer
without special programming”). None of these activities is
used in some unconventional manner nor does any produce
some unexpected result. Appellant does not contend it
invented any of these activities. In short, each step does no
more than require a generic computer to perform generic
computer functions. Even data vectorization is simply
application of a mathematical algorithm, itself an abstract idea
Appeal 2020-004875
Application 15/226,249

9

manifestation. Further, such vectorization is no more than
formatting data as appropriate for a machine learning operation
Decision 16 (emphasis added). Appellant tellingly contends it invented the
limitation, which are the activities applied to the particular data, and not the
activities themselves, which was what we determined. But as we also
determined, the source of the data cannot confer eligibility. Id.
We are not persuaded by Appellant’s argument that “a vast gulf exists
between the general concept of using vectors and the claim element of: for
each entity, converting the entity features and relationships between two or
more entities into a vectorized representation of the entity.” Request 10.
The limitation recited is itself no more than the conceptual application of
vectorization to particular data. No technological implementation details are
recited. Thus, the recitation is simply that of changing the manner of
representation. Representational change is a hallmark of mental analysis.
CONCLUSION
Nothing in Appellant’s request has convinced us that the new ground
of patent eligibility is improper as argued by Appellant. Accordingly, we
DENY the request.
Appeal 2020-004875
Application 15/226,249

10

To summarize, our decision is as follows:
 We have considered the REQUEST FOR REHEARING
 We DENY the request that we reverse the Examiner as to claims 1–11
and 16–20.

Outcome of Decision on Rehearing:
Claim(s) 35 U.S.C. § Basis Granted Denied
1–11, 16–20 101 Eligibility 1–11, 16–20

Final Outcome of Appeal after Rehearing:
Claims 35 U.S.C. § Basis Affirmed Reversed
1–11, 16–20 101 Eligibility 1–11, 16–20
1, 2, 6–8,
16, 19, 20
102(a)(2) Laster 1, 2, 6–8, 16,
19, 20
3, 4, 10, 11,
17, 18
103 Laster, Li 3, 4, 10, 11,
17, 18
10, 11 103 Laster,
Trouillon
10, 11
5 103 Laster,
Karpistsenko
5
9 103 Laster,
Danner
9
Overall Outcome 1–11, 16–20

REHEARING DENIED