DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This action is in response to the original application filed on May 28th, 2024.
Priority
Acknowledgment is made of applicant’s claim for foreign priority under 35 U.S.C. 119 (a)-(d). The certified copy has been filed in parent Application No. FR 21306638.4, filed on Nov. 25th, 2021.
Specification
The title of the invention is not descriptive. A new title is required that is clearly indicative of the invention to which the claims are directed.
Claim Objections
Claim 3 is objected to because of the following informalities: Claim 3 recites, “The method of of claim 1, …” (Emphasis added). The word “of” is repeated two times in the claim. This appears to be a grammar error and should be corrected. Appropriate action is required.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claim 9 is rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 9 recites the limitation, “wherein the training is performed to minimize the maximum likelihood-based loss function defined as:
L
θ
,
π
=
-
∑
i
=
1
N
L
o
g
∑
m
=
1
M
e
x
p
[
l
o
g
π
m
+
log
p
θ
σ
m
y
i
x
i
]
” (Emphasis added). The claim discloses a mathematical equation and fails to explicitly disclose the variables in the equation. The variables:
L
,
θ
,
π
,
N
,
M
,
σ
,
y
a
n
d
x
are all not defined in the claims making the broadest reasonable interpretation of these variables to be any undefined value. Therefore, since these variables are undefined, this claim is rejected under 35 U.S.C. 112(b) for being indefinite. For examination purposes the claimed variables will be interpreted to be real values consistent with the specification’s description of the loss function. Appropriate correction is required.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claim 14 is rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter. The claim does not fall within at least one of the four categories of patent eligible subject matter.
Claim 14 recites:
"A computer program product comprising:”
“code instructions that cause a computer system to perform the method as defined in claim 1 when the program is run on the computer system."
The submitted specification recites: “Another aspect of the invention relates to a computer program product comprising code instructions that cause a computer system to perform the method as defined above when the program is run on the computer system.” (Specification, pp. 13, Ln. 4-6). The specification states the proposed system could be in a non-physical, or non-tangible form, such as a computer program. Therefore, these claims would not fall under the four categories of patent eligible subject matter per MPEP 2106.03(I) which states, “The other three categories (machines, manufactures and compositions of matter) define the types of physical or tangible "things" or "products" that Congress deemed appropriate to patent. […] Thus, when determining whether a claimed invention falls within one of these three categories, examiners should verify that the invention is to at least one of the following categories and is claimed in a physical or tangible form. […] Non-limiting examples of claims that are not directed to any of the statutory categories include: Products that do not have a physical or tangible form, such as information (often referred to as "data per se") or a computer program per se (often referred to as "software per se") when claimed as a product without any structural recitations;”. Appropriate action is required.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-14 are rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The analysis of the claims will follow the 2019 Revised Patent Subject Matter Eligibility Guidance, 84 Fed. Reg. 50 (“2019 PEG”).
Claim 1
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
Claim 1 recites, "A computer-implemented method for training a multi-task network with at least one recurrent network having:" therefore it is directed to the statutory category of a process.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites, inter alia:
“a differentiable order selector for determining a convex combination of a number M of different possible task orders, respectively blocks orders, for processing an input, by allocating a selector order coefficient
π
i
to each task order, respectively block order, and,” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses mathematical concept of utilizing a mathematical formula to perform calculations. This claim recites a process of performing mathematical functions to process data or tasks. A human is able to perform computations using known methods and functions. This claim also discloses a math operation and therefore is ineligible.
“a merging module for computing the weighted average of the outputs given by the recurrent network for the M orders using as weights the order selector coefficients (
π
1
,
π
2
,
…
,
π
M
), the method comprising:” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses mathematical concept of utilizing a mathematical formula to perform calculations. This claim recites a process of performing mathematical functions to process data or tasks. A human is able to perform computations using known methods and functions. This claim also discloses a math operation and therefore is ineligible.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “A computer-implemented method for training a multi-task network with at least one recurrent network having:” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
“task-specific cells, respectively blocks of tasks,” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
“training jointly the task-specific cells and the order selector to minimize a loss function.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “A computer-implemented method for training a multi-task network with at least one recurrent network having:” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
“task-specific cells, respectively blocks of tasks,” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
“training jointly the task-specific cells and the order selector to minimize a loss function.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 2
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites, inter alia:
“wherein the order selector determines the convex combination of different possible orders based on a soft order modelling inside Birkhoff's polytope.” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to evaluate data using known methods and/or algorithms. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
This claim does not recite any additional limitations which integrate the abstract idea into a practical application.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea and thus the claim is subject-matter ineligible.
Claim 3
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites the abstract ideas of the preceding claims from which it depends.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “wherein the order selector performs an order dropout during at least part of the training.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “wherein the order selector performs an order dropout during at least part of the training.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 4
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites the abstract ideas of the preceding claims from which it depends.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “wherein the dropout comprises training each example on a random subset of k permutations by zeroing-out order selector coefficients.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “wherein the dropout comprises training each example on a random subset of k permutations by zeroing-out order selector coefficients.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 5
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites, inter alia:
“comprising freezing the order selector during a warm-up phase so that all M task orders are given an identical weight.” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human can evaluate data and make a determination to halt/continue operations of computing system. The limitation recites a mental process including observation, evaluation, judgment and/or opinion, see MPEP 2106.04(a).
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
This claim does not recite any additional limitations which integrate the abstract idea into a practical application.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea and thus the claim is subject-matter ineligible.
Claim 6
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites the abstract ideas of the preceding claims from which it depends.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “wherein the training comprises joint training of a shared encoder with the training of the order selector and task-specific cells of the recurrent network.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “wherein the training comprises joint training of a shared encoder with the training of the order selector and task-specific cells of the recurrent network.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 7
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites the abstract ideas of the preceding claims from which it depends.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “wherein the tasks are selected among one of face attribution classification or facial action detection.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “wherein the tasks are selected among one of face attribution classification or facial action detection.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 8
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites the abstract ideas of the preceding claims from which it depends.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “wherein the cells are recurrent cells, in particular of GRU type.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “wherein the cells are recurrent cells, in particular of GRU type.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 9
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites, inter alia:
"wherein the training is performed to minimize the maximum likelihood-based loss function defined as:
L
θ
,
π
=
-
∑
i
=
1
N
L
o
g
∑
m
=
1
M
e
x
p
[
l
o
g
π
m
+
log
p
θ
σ
m
y
i
x
i
]
” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses mathematical concept of utilizing a mathematical formula to perform calculations. This claim discloses a mathematical formula that a human is able to evaluate and execute on a computing system. This claim discloses a math operation and therefore is ineligible.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
This claim does not recite any additional limitations which integrate the abstract idea into a practical application.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea and thus the claim is subject-matter ineligible.
Claim 10
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites the abstract ideas of the preceding claims from which it depends.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “wherein the order selector comprises a softmax layer over logits u.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “wherein the order selector comprises a softmax layer over logits u.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 11
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites, inter alia:
“sampling R orders from the order selector, and L trajectories for each recurrent cell,” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to evaluate data and select tasks from an organized list with given parameters. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
“generating a global network prediction by averaging the predictions of the L *R samples.” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses mathematical concept of utilizing a mathematical formula to perform calculations. A human is able to perform a simple calculation such as an average of multiple values. This claim discloses a math operation and therefore is ineligible.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “A computer-implemented method for performing multiple prediction tasks using a multi-task network trained according to the method claim 1, the method comprising:” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “A computer-implemented method for performing multiple prediction tasks using a multi-task network trained according to the method claim 1, the method comprising:” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 12
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites, inter alia:
“wherein the sampling is done using Monte-Carlo sampling estimation.” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to evaluate data with given methods and functions and execute them on a generic computing systems. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
This claim does not recite any additional limitations which integrate the abstract idea into a practical application.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea and thus the claim is subject-matter ineligible.
Claim 13
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites the abstract ideas of the preceding claims from which it depends.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “wherein said method includes performing at least one of: human face and body analysis, scene analysis, speech recognition, image classification.” amounts to no more than generally linking the use of a judicial exception to a particular technology environment or field of use (see MPEP 2106.05(h)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “wherein said method includes performing at least one of: human face and body analysis, scene analysis, speech recognition, image classification.” amounts to no more than generally linking the use of a judicial exception to a particular technology environment or field of use (see MPEP 2106.05(h)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 14
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
This claim is not directed to one of the four categories of statutory subject matter per MPEP 2106.03(I). However, for compact prosecution, the examiner will interpret this claim as falling under one of the four categories to further evaluate the claim using the Alice/Mayo test.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites, inter alia:
“code instructions that cause a computer system to perform the method as defined in claim 1 when the program is run on the computer system.” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. As stated above, claim 1 is interpreted to recite an abstract idea. Claim 14 is claiming the method of claim 1 and is merely applying an abstract idea, claim 1, on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
This claim does not recite any additional limitations which integrate the abstract idea into a practical application.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea and thus the claim is subject-matter ineligible.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1, 3, 5-8, 11, and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Meyerson et al, (Meyerson et al, “BEYOND SHARED HIERARCHIES: DEEP MULTITASK LEARNING THROUGH SOFT LAYER ORDERING”, US 20190130257 A1, Filed 2018, hereinafter “Meyerson”) in view of Sun et al, (Sun et al, “AdaShare: Learning What To Share For Efficient Deep Multi-Task Learning”, 2020, hereinafter “Sun”).
Regarding claim 1, Meyerson discloses, “A computer-implemented method for training a multi-task network with at least one recurrent network having:” (Some Alternative Implementations, pp. 6, [0088]; “Eq. 7 is defined recursively with respect to the learned layers shared across tasks. Thus, the soft-ordering architecture can be viewed as a new type of recurrent architecture designed specifically for MTL. From this perspective, FIG. 4 shows an unrolling of a soft layer module: different scaling parameters are applied at different depths when unrolled for different tasks.” Meyerson discloses a multitask network that is able to use recurrent methods and systems to make predictions.)
“task-specific cells, respectively blocks of tasks,” (The Parallel Ordering Assumption, pp. 3, [0046];; “Consider T tasks
t
1
,
…
,
t
T
to be learned jointly, with each
t
i
associated with a model
y
i
=
F
i
(
x
i
)
. Suppose sharing across tasks occurs at D consecutive depths. Let
ε
i
(
D
i
) be
t
i
'
s
task-specific encoder (decoder) to (from) the core sharable portion of the network from its inputs (to its outputs).” This article discloses the use of different modules used to evaluate tasks. The tasks are encoded into the system and placed in the task specific cells or modules. See fig. 2 for the architecture of the proposed model.)
“a merging module for computing the weighted average of the outputs given by the recurrent network for the M orders using as weights the order selector coefficients (
π
1
,
π
2
,
…
,
π
M
), the method comprising:” (Visualization the Behavior of Soft Order Layers, pp. 6, [0084]; “Ten tasks are trained using soft ordering with four shared dense ReLU layers of 100 units each.
ε
i
is a linear encoder that is shared across tasks, and
D
i
is a global average pooling decoder. Thus, task models are distinguished completely by their learned soft ordering scaling parameters
s
i
. To visualize the behavior of layer I at depth d for task t, the predicted image for task t is generated across varying magnitudes of
s
(
t
,
l
,
d
)
· The results for the first two tasks and the first layer are shown in Table 1.” This model uses many different processing layers depending on the tasks input. This model can use an average pooling layer which will take the output from the previous layer and average the outputs. This averaging is after inference and the values are altered by the previous layers weights.)
Meyerson fails to explicitly disclose:
“a differentiable order selector for determining a convex combination of a number M of different possible task orders, respectively blocks orders, for processing an input, by allocating a selector order coefficient
π
i
to each task order, respectively block order, and,”
“training jointly the task-specific cells and the order selector to minimize a loss function.”
However, Sun discloses, “a differentiable order selector for determining a convex combination of a number M of different possible task orders, respectively blocks orders, for processing an input, by allocating a selector order coefficient
π
i
to each task order, respectively block order, and,” (Fig. 2 Illustration of our proposed approach, pp. 4; “AdaShare learns the layer sharing pattern among multiple tasks through predicting a select-or-skip policy decision sampled from the learned task-specific policy distribution (logits). These select-or-skip vectors define which blocks should be executed in different tasks. A block is said to be shared across two tasks if it is being used by both of them or task-specific if it is being used by only one task for predicting the output. During training, both policy logits and network parameters are jointly learned using standard back-propagation through Gumbel-Softmax Sampling. We use task-specific losses and policy regularizations (to encourage sparsity and sharing) in training. Best viewed in color.” Sun discloses multitask learning network that is able to determine the execution order of the tasks with different layers or modules. This system will evaluate each task and determine if a layer or module would be beneficial for the specific task and adjust the module execution order. During this evaluation, each task is uses specific values to determine the model layout.)
“training jointly the task-specific cells and the order selector to minimize a loss function.” (Learning a Task-Specific Policy, pp. 4; “In AdaShare, we learn the select-or-skip policy U and network weights W jointly through standard back-propagation from our designed loss functions. However, each select-or-skip policy
u
l
,
k
is discrete and non-differentiable and this makes direct optimization difficult. Therefore, we adopt Gumbel-Softmax Sampling [25] to resolve this non-differentiability and enable direct optimization of the discrete policy
u
l
,
k
using back-propagation.” This model disclose the use of a loss function to train the different modules of the model.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Meyerson and Sun. Meyerson teaches a multitask learning model that is able to evaluate the order of the leaning modules depending on the task. Sun teaches a multitask learning model that is able to evaluate a task and determine an execution order and suggested models for the specified tasks. One of ordinary skill would have motivation to combine different ways to implement a multitask learning network and different ways to select and order internal modules based on input tasks to improve existing MTL, “AdaShare requires much less computation (FLOPs) as compared to existing MTL methods. E.g., in Cityscapes 2-task, Cross-stitch/Sluice, NDDR, MTAN, DEN, and AdaShare use 37.06G, 38.32G, 44.31G, 39.18G and 33.35G FLOPs and in NYU v2 3-task, they use 55.59G, 57.21G, 58.43G, 57.71G and 50.13G FLOPs, respectively. Overall, AdaShare offers on average about 7.67%-18.71% computational savings compared to state-of-the-art methods over all the tasks while achieving better recognition accuracy with about 50%-80% less parameters.” (Sun, Computation Cost (FLOPs), pp. 8).
Regarding claim 3, Meyerson discloses, “wherein the order selector performs an order dropout during at least part of the training.” (UCI Experiments, pp. 7, [0100]; “For all tasks, each input feature is scaled to be between O and 1. For each task, training and validation data were created by a random 80-20 split. This split is fixed across trials. A dropout rate of0.8 is applied at the output of each core layer.” This model discloses the use of a dropout rate which would indicate the use of performing dropout during training.)
Regarding claim 5, Meyerson discloses, “comprising freezing the order selector during a warm-up phase so that all M task orders are given an identical weight.” (Parallel Ordering of Layers in Deep Multitask Learning (MTL), pp. 5, [0068]; “In training, all
s
(
i
,
j
,
k
)
are initialized with equal values, to reduce initial bias of layer function across tasks. It is also helpful to apply dropout after each shared layer.” This systems is able to adjust the weights during a training phase to give them equal values. This occurs at the start of the training process where the model is paused, or frozen, waiting to be trained.)
Regarding claim 6, Meyerson discloses, “wherein the training comprises joint training of a shared encoder with the training of the order selector and task-specific cells of the recurrent network.” (System, pp. 10, [0129]; “In implementations, encoder 102 comprises learnable components, parameters, and hyperparameters that can be trained by backpropagating errors using an optimization algorithm. The optimization algorithm can be based on stochastic gradient descent (or other variations of gradient descent like batch gradient descent and mini-batch gradient descent). Some examples of optimization algorithms that can be used to train the encoder 102 are Momentum, Nesterov accelerated gradient, Adagrad, Adadelta, RMSprop, and Adam.” This system uses a shared encoder to encode all the tasks to the appropriate module. This system will train the modules as well using backpropagation and a loss function.)
Regarding claim 7, Meyerson discloses, “wherein the tasks are selected among one of face attribution classification or facial action detection.” (Empirical Evaluation of Soft Layer Ordering, pp. 5, [0069]; “These experiments evaluate soft ordering against fixed ordering MTL and single-task learning. The first experiment applies them to closely related MNIST tasks, the second to " superficially unrelated" UCI tasks, the third to the real-world problem of Omniglot character recognition, and the fourth to large-scale facial attribute recognition. In each experiment, single task, parallel ordering (Eq. 2), permuted ordering (Eq. 3), and soft ordering (Eq. 7) train an equivalent set of core layers. In permuted ordering, the order of layers is randomly generated for each task in each trial. See Section 7 for additional details specific to each experiment.” This system is able to evaluate image data for facial attribute recognition.)
Regarding claim 8, “wherein the cells are recurrent cells, in particular of GRU type.” (System, pp. 10, [0127]; “Encoder 102 is a processor that receives information characterizing input data and generates an alternative representation and/or characterization of the input data, such as an encoding. In particular, encoder 102 is a neural network such as a convolutional neural network (CNN), a multilayer perceptron, a feed-forward neural network, a recursive neural network, a recurrent neural network (RNN), a deep neural network, a shallow neural network, a fully-connected neural network, a sparsely-connected neural network, a convolutional neural network that comprises a fully-connected neural network (FCNN), a fully convolutional network without a fully-connected neural network, a deep stacking neural network, a deep belief network, a residual network, echo state network, liquid state machine, highway network, maxout network, long short-term memory (LSTM) network, recursive neural network grammar (RNNG), gated recurrent unit (GRU), pre-trained and frozen neural networks, and so on.” This system discloses the use of different types of ML models, and includes using recurrent modules and gated recurrent units.)
Regarding claim 11, Meyerson discloses, “A computer-implemented method for performing multiple prediction tasks using a multi-task network trained according to the method claim 1, the method comprising:” (Some Alternative Implementations, pp. 6, [0088]; “Eq. 7 is defined recursively with respect to the learned layers shared across tasks. Thus, the soft-ordering architecture can be viewed as a new type of recurrent architecture designed specifically for MTL. From this perspective, FIG. 4 shows an unrolling of a soft layer module: different scaling parameters are applied at different depths when unrolled for different tasks.” Meyerson discloses a multitask learning network able to evaluate tasks and produce predictions.)
Meyerson fails to explicitly disclose:
“sampling R orders from the order selector, and L trajectories for each recurrent cell,”
“generating a global network prediction by averaging the predictions of the L *R samples.”
However, Sun discloses, “sampling R orders from the order selector, and L trajectories for each recurrent cell,” (Proposed Method, pp. 4; “After the training finishes, we sample the binary decision
u
l
,
k
for each block l from
u
l
,
k
to decide what blocks to select or skip in the task
T
k
. Specifically, with the help of the select-or-skip decisions, we form a novel and non-trivial network architecture for MTL parameter-sharing, and share knowledge at different levels across all tasks in a flexible and efficient way. At test time, when a novel input is presented to the multi-task network, the optimal policy is followed, selectively choosing what blocks to compute for each task. Our proposed approach not only encourages positive sharing among tasks via shared blocks but also minimizes negative interference by using task-specific blocks when necessary.” This system is able to evaluate a set of tasks and their input data and evaluate the inputs. This system will evaluate the tasks and determine an order which to execute these tasks by enabling and disabling layers of the model depending on the specified task. This model will also use a sampling process to sample tasks.)
“generating a global network prediction by averaging the predictions of the L *R samples.” (Figure 2, pp. 4; This figure discloses the architecture of the model used in Sun. As seen in the image there are multiple different layers that can be activated or deactivated based on the input task. This teaches the generation of a global network which is able to produce a prediction. This will take each task and apply specified functions to the tasks to produce a prediction) and (Evaluation Metrics, pp. 6; “For image classification and text recognition, we report classification accuracy for each domain/dataset. Instead of reporting the absolute task performance with multiple metrics for each task Ti, we follow [37] and report a single relative performance
∆
T
i
with respect to the Single-Task baseline to clearly show the positive/negative transfer in different baselines: [see equation (6)] where
l
j
=
1
if a lower value represents better for the metric
M
j
and 0 otherwise. Finally, we average over all tasks to get overall performance
∆
T
=
1
T
∑
i
=
1
K
∆
T
i
.” Sun further discloses a method for image classification and text generation. This method will evaluate the tasks with the trained machine learning network and average tasks in order to get an overall performance metric.)
PNG
media_image1.png
323
1088
media_image1.png
Greyscale
Regarding claim 14, Meyerson discloses, “code instructions that cause a computer system to perform the method as defined in claim 1 when the program is run on the computer system.” (Computer system, pp. 12-13, [0188]; “FIG. 9 is a simplified block diagram of a computer system 900 that can be used to implement the technology disclosed. Computer system 900 includes at least one central processing unit (CPU) 972 that communicates with a number of peripheral devices via bus subsystem 955. These peripheral devices can include a storage subsystem 910 including, for example, memory devices and a file storage subsystem 936, user interface input devices 938, user interface output devices 976, and a network interface subsystem 974. The input and output devices allow user interaction with computer system 900. Network interface subsystem 974 provides an interface to outside networks, including an interface to corresponding interface devices in other computer systems.” Meyerson discloses a multitask learning network and is able to execute the proposed methods on a computing system.)
Claim 2 is rejected under 35 U.S.C. 103 as being unpatentable over Meyerson and Sun in view of Mena et al, (Mena et al, “LEARNING LATENT PERMUTATIONS WITH GUMBELSINKHORN NETWORKS”, 2018, hereinafter “Mena”).
Regarding claim 2, Mena discloses, “wherein the order selector determines the convex combination of different possible orders based on a soft order modelling inside Birkhoff's polytope.” (An Approximation Theorem for the Matching Problem, pp. 15-16; “We aim to maximize a linear functional (in the sense of the Frobenius norm) in the space of permutation matrices. In this context, let’s define the matching operator
M
(
∙
)
as the one that returns the solution of the assignment problem: [See equation (8)] Likewise, we define
M
~
(
∙
)
as a related operator, but changing the feasible space by the Birkhoff polytope: [See equation (9)] Notice that in general
M
~
X
,
M
(
X
)
might not be unique matrices, but a face of the Birkhoff polytope, or a set of permutations, respectively (see Lemma 2 for details). In any case, the relation
M
(
X
)
⊆
M
~
(
X
)
holds by virtue of Birkhoff’s theorem, and the fundamental theorem of linear programming.” Mena discloses the use of determining an order of vectors based on a shape similar to Birkhoffs polytope. This article uses this polytope in a matching setting however the concepts of evaluating each task, or input, and then ordering them applies similar concepts to the claimed invention.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Meyerson, Sun, and Mena. Meyerson teaches a multitask learning model that is able to evaluate the order of the leaning modules depending on the task. Sun teaches a multitask learning model that is able to evaluate a task and determine an execution order and suggested models for the specified tasks. Mena teaches different methods to sort and organize data and permutations. One of ordinary skill would have motivation to combine different ways to implement a multitask learning network and different ways to select and order internal modules based on input tasks to improve existing MTL and would research subjects on ordering and sorting data to further improve task sorting and execution orders, “In Table 3 (and also in Table 7 of appendix C.4) we show results for this task, using accuracy in matching as the performance measure. These are broken down by relevant experimental covariates (Linderman et al., 2017): different proportion of neurons known beforehand, and by task difficulty. As baselines, we include i) a simple MCMC sampler that proposes local swipes on permutations ii) the rounding method presented in Linderman et al. (2017), iii) our method, where we also consider the absence of regularization. Results show our method outperforms the alternatives in most cases. MCMC fails because mixing is poor, but differences are much subtler with the other baselines.” (Posterior Inference Over Permutations with the Gumbel-Sinkhorn Estimator, pp. 9)
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Meyerson and Sun in view of Devin et al, (Devin et al, “Learning Modular Neural Network Policies for Multi-Task and Multi-Robot Transfer”, 2016, hereinafter “Devin”).
Regarding claim 4, Devin discloses, “wherein the dropout comprises training each example on a random subset of k permutations by zeroing-out order selector coefficients.” (Regularization, pp. 5; “If, for example, a robot module overfits to the robot task combinations seen during training, it might partition itself into different “receptors” for different tasks, instead of acquiring a task-invariant interface. With only a few robots and tasks (e.g. 3 robots and 3 tasks), we have found overfitting to be problematic. To mitigate this effect, we regularize our modules in two ways: by limiting the number of hidden units in the module interface, and by applying the dropout method, described below. … Dropout is a neural network regularization method that sets a random subset of the activations to 0 at each minibatch [29]. This prevents the network from depending too heavily on any particular hidden unit and instead builds redundancy into the weights. This limits the information flow between the task and robot modules, reducing their ability to overspecialize to the training conditions.” Devin discloses a dropout method used in their multitask networks for regularization. Devin discloses a method where a subset of tasks, i.e. permutations, activation functions, i.e. coefficients, are zeroed out.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Meyerson and Sun. Meyerson teaches a multitask learning model that is able to evaluate the order of the leaning modules depending on the task. Sun teaches a multitask learning model that is able to evaluate a task and determine an execution order and suggested models for the specified tasks. Devin teaches a system that is able to use reinforcement learning and multi-task learning to improve transfer learning of tasks for robotic systems. One of ordinary skill would have motivation to combine different ways to implement training methods in multitask learning networks using a dropout method to increase generalization and avoid overfitting, “In order to obtain zero-shot performance on unseen robot task combinations, the modules must learn standardized interfaces. If, for example, a robot module overfits to the robot task combinations seen during training, it might partition itself into different “receptors” for different tasks, instead of acquiring a task-invariant interface. With only a few robots and tasks (e.g. 3 robots and 3 tasks), we have found overfitting to be problematic. To mitigate this effect, we regularize our modules in two ways: by limiting the number of hidden units in the module interface, and by applying the dropout method, described below.”.
Claims 10 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Meyerson and Sun in view of Ranjan et al, (Ranjan et al, “HyperFace: A Deep Multi-Task Learning Framework for Face Detection, Landmark Localization, Pose Estimation, and Gender Recognition”, 2019, hereinafter “Ranjan”).
Regarding claim 10, Ranjan discloses, “wherein the order selector comprises a softmax layer over logits u.” (Training, pp. 124; “The candidate regions with IOU overlap less than 0.35 are treated as negative instances (l = 0). All the other regions are ignored. We use the softmax loss function given by (1) for training the face detection task.” Ranjan discloses a multitask learning system which uses a CNN and other ML models to evaluate image data. This system does utilize a softmax layer.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Meyerson, Sun, and Ranjan. Meyerson teaches a multitask learning model that is able to evaluate the order of the leaning modules depending on the task. Sun teaches a multitask learning model that is able to evaluate a task and determine an execution order and suggested models for the specified tasks. Ranjan teaches a multitask learning network able to evaluate image data using CNNs. One of ordinary skill would be motivated to research similar multitask learning methods which are able to process and order incoming tasks and further modify system architecture to better process tasks. Further, one would be motivated to combine different task selection methods with a system able to evaluate images using a CNN model which contains sofmax layers, “The proposed algorithm called HyperFace consists of three modules. The first one generates class independent region proposals from the given image and scales them to 227 x 227 pixels. The second module is a CNN which takes in the resized candidate regions and classifies them as face or nonface. If a region gets classified as a face, the network additionally provides facial landmarks locations, estimated head pose and gender information. The third module is a postprocessing step which involves Iterative Region Proposals and Landmarks-based Non-Maximum Suppression (L-NMS) to boost the face detection score and improve the performance of individual tasks.” (Ranjan, HyperFace, pp. 123)
Regarding claim 13, Ranjan discloses, “wherein said method includes performing at least one of: human face and body analysis, scene analysis, speech recognition, image classification.” (Fig. 2, pp. 123; “The architecture of the proposed HyperFace. The network is able to classify a given image region as face or non-face, estimate the head pose, locate face landmarks and recognize gender.” Ranjan discloses a multitask learning system able to evaluate images for facial recognition or analysis tasks.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Meyerson, Sun, and Ranjan. Meyerson teaches a multitask learning model that is able to evaluate the order of the leaning modules depending on the task. Sun teaches a multitask learning model that is able to evaluate a task and determine an execution order and suggested models for the specified tasks. Ranjan teaches a multitask learning network able to evaluate image data using CNNs. One of ordinary skill would have motivation to combine different ways to implement a multitask learning network as well as different ways to select and order internal modules based on input tasks to improve existing MTL’s. Further, one would be motivated to combine the disclosed arts with another MTL model that is able to evaluate image data and produce predictions using averaging layers, “The precision-recall curves of different detectors corresponding to AFW and PASCAL faces datasets are shown in Figs. 7a and 7b, respectively. Fig. 8 compares the performance of different detectors using the Receiver Operating Characteristic (ROC) curves on the FDDB dataset. As can be seen from these figures, both HyperFace and HF-ResNet outperform all the reported academic and commercial detectors on the AFW and PASCAL datasets. HyperFace achieves a high mean average precision (mAP) of 97.9 and 92.46 percent, for AFWand PASCAL datasets respectively. HF-ResNet further improves themAP to 99.4 and 96.2 percent respectively.” (Ranjan, Face Detection, pp. 128).
Claim 12 is rejected under 35 U.S.C. 103 as being unpatentable over Meyerson and Sun in view of Lin et al, (Lin et al, “Controllable Pareto Multi-Task Learning”, Feb. 15th, 2021, hereinafter “Lin”).
Regarding claim 12, Lin discloses, “wherein the sampling is done using Monte-Carlo sampling estimation.” (Optimization: Learning the Generator, pp. 5; “Suppose we have a probability distribution
P
p
for all valid preference vector p, a general goal would be: [See Equation (6)] However, it is hard to optimize the trainable parameters ϕ within the expectation directly. We use Monte Carlo method to sample the preference vectors, and use the stochastic gradient descent algorithm to train the hypernetwork-based MTL model.” Lin discloses a multitask learning network which is able to use Monte- Carlo sampling.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Meyerson, Sun, and Lin. Meyerson teaches a multitask learning model that is able to evaluate the order of the leaning modules depending on the task. Sun teaches a multitask learning model that is able to evaluate a task and determine an execution order and suggested models for the specified tasks. Lin discloses a MTL network that is able to use different sampling techniques to optimize training. One of ordinary skill would have motivation to combine different ways to implement a multitask learning network and different ways to select and order internal modules based on input tasks to improve existing MTL, “In this paper, we proposed a novel controllable Pareto multitask learning framework for solving MTL problems. With a preference-based hypernetwork, our method can learn the whole trade-off curve for all tasks with a single model. It allows practitioners to easily make real-time trade-off adjustment among tasks at the inference time. Experimental results on various MTL applications demonstrated the usefulness and efficiency of the proposed method.” (Lin, Conclusion, pp. 8)
Allowable Subject Matter
Claim 9 would be allowable if rewritten to overcome the rejection(s) under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), 2nd paragraph, set forth in this Office action and to include all of the limitations of the base claim and any intervening claims.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to PAUL MICHAEL GALVIN-SIEBENALER whose telephone number is (571)272-1257. The examiner can normally be reached Monday - Friday 8AM to 5PM.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Viker Lamardo can be reached at (571) 270-5871. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/PAUL M GALVIN-SIEBENALER/Examiner, Art Unit 2147 /MARC S SOMERS/Primary Examiner, Art Unit 2159
Prosecution Insights