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 .
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. JP2021-157331, filed on September 28, 2021.
Information Disclosure Statement
The information disclosure statement(s) (IDS) submitted on 5/28/2024 is/are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement(s) is/are being considered by the examiner.
Claim Interpretation
The following is a quotation of 35 U.S.C. 112(f):
(f) Element in Claim for a Combination. – An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph:
An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked.
As explained in MPEP § 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph:
(A) the claim limitation uses the term “means” or “step” or a term used as a substitute for “means” that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function;
(B) the term “means” or “step” or the generic placeholder is modified by functional language, typically, but not always linked by the transition word “for” (e.g., “means for”) or another linking word or phrase, such as “configured to” or “so that”; and
(C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function.
Use of the word “means” (or “step”) in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function.
Absence of the word “means” (or “step”) in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function.
Claim limitations in this application that use the word “means” (or “step”) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word “means” (or “step”) are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action.
This application includes one or more claim limitations that do not use the word “means,” but are nonetheless being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, because the claim limitation(s) uses a generic placeholder that is coupled with functional language without reciting sufficient structure to perform the recited function and the generic placeholder is not preceded by a structural modifier.
The following limitations are interpreted as invoking 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph:
Claim 1 recites “a base class feature extraction unit”, “a novel class feature extraction unit”, “a mixture feature calculation unit”, “a learning unit”, and “a weight selection unit”.
If applicant does not intend to have these limitations interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitations to avoid them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitations recite sufficient structure to perform the claimed function so as to avoid them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph.
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, and 4 are rejected under 35 U.S.C. 103 as being unpatentable over YOON ("XtarNet: Learning to Extract Task-Adaptive Representation for Incremental Few-Shot Learning") in view of SHI ("US 20210012226 A1") and BELOUADAH ("Initial Classifier Weights Replay for Memoryless Class Incremental Learning"), hereafter YOON, SHI, and BELOUADAH, respectively.
Regarding Claim 1:
YOON teaches:
A machine learning device that performs continual learning of a fewer number of novel classes than the number of base classes, comprising: (YOON [page 1, Abstract] teaches: “We propose XtarNet, which learns to extract task-adaptive representation (TAR) for facilitating incremental few-shot learning (i.e., continual learning).” YOON [page 3, Pretraining phase] teaches: “Pretraining phase: On data split
D
(
b
a
s
e
/
t
r
a
i
n
)
with a fixed set of
N
b
base classes (i.e., the number of base classes), we train the embedding network
f
θ
and the base classifier weights of W.” YOON [page 3, Meta-training phase] teaches: “For each episode,
N
novel classes are randomly chosen from training set
D
(
n
o
v
e
l
/
t
r
a
i
n
)
(i.e., a fewer number of novel classes).” YOON [page 8, section 5.2. Quality of Clustering] teaches: “For each test episode for the 5-shot 200+5 tieredImageNet classification, the SSE values for each method are averaged for base and novel categories.” YOON [page 5, section 4. Experimental Result] teaches: “Pretraining is done with
D
b
a
s
e
consisting of 64 and 200 base categories for miniImageNet and tieredImageNet experiments, respectively. In the meta-training phase, 5 novel categories are selected from
D
(
n
o
v
e
l
/
t
r
a
i
n
)
for each episode (i.e., performs continual learning of a fewer number of novel classes than the number of base classes).” Examiner’s note: Under BRI, a machine learning device that performs continual learning can be interpreted as the device or system in which XtarNet is implemented.)
a base class feature extraction unit that extracts feature vectors of the base classes; (YOON [page 2, section 2. Proposed Model] teaches: “The backbone feature extractor
f
θ
(i.e., a base class feature extraction unit) is prepared through a regular supervised learning on a training dataset for base categories. […] The backbone network output represents the base feature (i.e., that extracts feature vectors of the base classes).”)
a novel class feature extraction unit that extracts feature vectors of the novel classes; (YOON [page 2, section 2. Proposed Model] teaches: “MetaCNN
g
(i.e., a novel class feature extraction unit) is a small convolutional neural network (CNN) module that takes as input an intermediate layer output of
f
θ
. The output of
g
represents the novel feature (i.e., that extracts feature vectors of the novel classes).”)
a mixture feature calculation unit that mixes the feature vectors of the base classes and the feature vectors of the novel classes and calculates a mixture feature vector of the base classes and the novel classes; (YOON [page 2, section 2. Proposed Model] teaches: “MergeNet
r
(i.e., a mixture feature calculation unit), consisting of two relatively small fully-connected networks, creates a mixture of the base and novel features (i.e., that mixes the feature vectors of the base classes and the feature vectors of the novel classes). This mixture is what we call the task-adaptive representation (TAR) (i.e., and calculates a mixture feature vector of the base classes and the novel classes) of the given input image.”)
a learning unit that classifies a query sample of a query set based on the distance between the position of a mixture feature vector of the query sample of the query set and the position of a classification weight vector of each class in a projection space and learns classification weight vectors of the novel classes so as to minimize classification loss; (YOON [page 2, section 2. Proposed Model] teaches: “Based on the support set, the model (i.e., a learning unit) aims to classify query samples from both base and novel categories. Our XtarNet utilizes three different meta-learnable modules in addition to a backbone network.” YOON [page 6, section 4.1. Evaluation Metrics] teaches: “Individual accuracy for base (or novel) classes are obtained by classifying the
Q
b
a
s
e
( or
Q
n
o
v
e
l
) query set (i.e., classifies a query sample of a query set) using only the base ( or novel) classifier.” YOON [page 14, Algorithm 1] teaches, for
i
=
1
,
…
,
N
b
+
N
(i.e., each class in a projection space) and for each
x
,
y
=
i
∈
Q
(i.e., query sample of the query set), computing the Euclidean distance between (i.e., distance between)
ω
p
r
e
⨀
f
θ
x
+
ω
m
e
t
a
⨀
g
a
θ
x
M
(i.e., a mixture feature vector of the query sample of the query set) and
w
i
*
M
(i.e., and the position of a classification weight vector of each class in a projection space). Further, YOON [page 4, TconNet for Conditioning Classifiers] teaches: “For a given query
x
∈
Q
, the Euclidean distance
D
i
x
between the projected feature of query and the projected classifier weights is used in classification:
D
i
x
=
d
w
i
*
M
,
ω
p
r
e
⨀
f
θ
x
+
ω
m
e
t
a
⨀
g
a
θ
x
M
,
(
5
)
Finally, classification of the query x is carried out based on the posterior probabilities (i.e., classifies a query sample of a query set based on the distance between the position of a mixture feature vector of the query sample of the query set and the position of a classification weight vector of each class in a projection space):
p
y
=
i
|
x
=
exp
-
D
i
x
∑
l
exp
-
D
l
x
,
(
6
)
for all
i
=
1
,
…
,
N
b
+
N
.” YOON [page 14, Algorithm 1] teaches that various parameters are updated in line 33 by minimizing the training loss via the optimizer (i.e., and learns classification weight vectors of the novel classes so as to minimize classification loss).)
[…] base classes […] for the query set in the projection space at the time of learning the query set in units of episodes. (YOON [page 2, section 2. Proposed Model] teaches: “The classification weight vector set W consists of base classifier weight vectors (i.e., base classes […] for the query set in the projection space), which are also pretrained on base categories, as well as novel classifier weight vectors, which vary with each training episode (i.e., at the time of learning the query set in units of episodes).” Examiner’s note: Under BRI, training in units of episodes can be interpreted as “for
t
=
1
…
T
d
o
” during the training of the model using query set
Q
(i.e., learning the query set), as shown in YOON [page 14, Algorithm 1].)
YOON is not relied upon for teaching:
a weight selection unit that sequentially adds classification weight vectors of […] selected for the query set […]
However, SHI teaches: […] base classes selected for the query set […] (SHI [0111] teaches: “In more detail, a base classifier may be constructed by selecting one or more base classes and one or more novel classes from the multiple classes of training data TD.” SHI [0113] teaches: “For example, a fixed number of query instances (i.e., for the query set) may be sampled for each class (i.e., base classes selected), e.g., randomly, from the training data TD. […] Interestingly, both base class instances and novel class instances may be used as query instances […].” SHI [0119] teaches: “Thereby, fast learning for novel few-shot classes may be combined with slow learning for the classes of training dataset TD, e.g., the information from training dataset TD may be incorporated in the classifier instead of being forgotten.”)
Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of YOON and SHI before them, to include SHI’s selection of a subset of base classes for each episode in YOON’s incremental few-shot learning method. One would have been motivated to make such a combination in order to obtain a more accurate classification of the query instance (SHI [0036]).
YOON in view of SHI is not relied upon for teaching, but BELOUADAH teaches: a weight selection unit that sequentially adds classification weight vectors of […] classes […] (BELOUADAH [page 5, section 3.1 Motivation] teaches: “New classifiers from previous states of Figure 2(a) are aggregated to represent past classes in each current state of Figure 2(b).” BELOUADAH [page 2, Figure 1] teaches a method that accumulates weight vectors (i.e., sequentially adds classification weight vectors of […] classes) in each incremental state to counter catastrophic forgetting. Examiner’s note: Under BRI, a weight selection unit can be interpreted as the component that executes the accumulation of weight vectors in a classification layer
W
t
, as shown in BELOUADAH [page 2, Figure 1].)
Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of YOON, SHI, and BELOUADAH before them, to include BELOUADAH’s reuse of initial classifier weights in YOON and SHI’s incremental few-shot learning method. One would have been motivated to make such a combination in order to reuse normalized initial classifier weights to mitigate the effect of catastrophic forgetting (BELOUADAH [page 10, section 5 conclusion]).
Regarding Claim 3:
The claim recites similar limitations as corresponding claim 1 and is rejected for similar reasons as claim 1 using similar teachings and rationale.
Regarding Claim 4:
YOON teaches:
A non-transitory computer-readable recording medium embodied thereon a machine learning program that performs continual learning of a fewer number of novel classes than the number of base classes, the program comprising computer-implemented modules including: (YOON [page 5, footer] teaches source code for XtarNet. One of ordinary skill in the art would recognize that running XtarNet continual learning method requires computer hardware, processor, and memory storing the instructions for executing XtarNet’s code (i.e., A non-transitory computer-readable recording medium embodied thereon a machine learning program that performs continual learning). YOON [page 1, Abstract] teaches: “We propose XtarNet, which learns to extract task-adaptive representation (TAR) for facilitating incremental few-shot learning (i.e., continual learning).” YOON [page 3, Pretraining phase] teaches: “Pretraining phase: On data split
D
(
b
a
s
e
/
t
r
a
i
n
)
with a fixed set of
N
b
base classes (i.e., the number of base classes), we train the embedding network
f
θ
and the base classifier weights of W.” YOON [page 3, Meta-training phase] teaches: “For each episode,
N
novel classes are randomly chosen from training set
D
(
n
o
v
e
l
/
t
r
a
i
n
)
(i.e., a fewer number of novel classes).” YOON [page 8, section 5.2. Quality of Clustering] teaches: “For each test episode for the 5-shot 200+5 tieredImageNet classification, the SSE values for each method are averaged for base and novel categories.” YOON [page 5, section 4. Experimental Result] teaches: “Pretraining is done with
D
b
a
s
e
consisting of 64 and 200 base categories for miniImageNet and tieredImageNet experiments, respectively. In the meta-training phase, 5 novel categories are selected from
D
(
n
o
v
e
l
/
t
r
a
i
n
)
for each episode (i.e., performs continual learning of a fewer number of novel classes than the number of base classes).” Examiner’s note: Under BRI, a machine learning device that performs continual learning can be interpreted as the device or system in which XtarNet is implemented.)
a base class feature extraction module that extracts feature vectors of the base classes; (YOON [page 2, section 2. Proposed Model] teaches: “The backbone feature extractor
f
θ
(i.e., a base class feature extraction module) is prepared through a regular supervised learning on a training dataset for base categories. […] The backbone network output represents the base feature (i.e., that extracts feature vectors of the base classes).”)
a novel class feature extraction module that extracts feature vectors of the novel classes; (YOON [page 2, section 2. Proposed Model] teaches: “MetaCNN
g
(i.e., a novel class feature extraction module) is a small convolutional neural network (CNN) module that takes as input an intermediate layer output of
f
θ
. The output of
g
represents the novel feature (i.e., that extracts feature vectors of the novel class).”)
a mixture feature calculation module that mixes the feature vectors of the base classes and the feature vectors of the novel classes and calculates a mixture feature vector of the base classes and the novel classes; (YOON [page 2, section 2. Proposed Model] teaches: “MergeNet
r
(i.e., a mixture feature calculation module), consisting of two relatively small fully-connected networks, creates a mixture of the base and novel features (i.e., that mixes the feature vectors of the base classes and the feature vectors of the novel classes). This mixture is what we call the task-adaptive representation (TAR) (i.e., and calculates a mixture feature vector of the base classes and the novel classes) of the given input image.”)
a learning module that classifies a query sample based on the distance between the position of a mixture feature vector of the query sample of a query set and the position of a classification weight vector of each class in a projection space and learns classification weight vectors of the novel classes so as to minimize classification loss; and (YOON [page 2, section 2. Proposed Model] teaches: “Based on the support set, the model (i.e., a learning module) aims to classify query samples from both base and novel categories. Our XtarNet utilizes three different meta-learnable modules in addition to a backbone network.” YOON [page 6, section 4.1. Evaluation Metrics] teaches: “Individual accuracy for base (or novel) classes are obtained by classifying the
Q
b
a
s
e
( or
Q
n
o
v
e
l
) query set (i.e., classifies a query sample of a query set) using only the base ( or novel) classifier.” YOON [page 14, Algorithm 1] teaches, for
i
=
1
,
…
,
N
b
+
N
(i.e., each class in a projection space) and for each
x
,
y
=
i
∈
Q
(i.e., query sample of the query set), computing the Euclidean distance between (i.e., distance between)
ω
p
r
e
⨀
f
θ
x
+
ω
m
e
t
a
⨀
g
a
θ
x
M
(i.e., a mixture feature vector of the query sample of the query set) and
w
i
*
M
(i.e., and the position of a classification weight vector of each class in a projection space). Further, YOON [page 4, TconNet for Conditioning Classifiers] teaches: “For a given query
x
∈
Q
, the Euclidean distance
D
i
x
between the projected feature of query and the projected classifier weights is used in classification:
D
i
x
=
d
w
i
*
M
,
ω
p
r
e
⨀
f
θ
x
+
ω
m
e
t
a
⨀
g
a
θ
x
M
,
(
5
)
Finally, classification of the query x is carried out based on the posterior probabilities (i.e., classifies a query sample of a query set based on the distance between the position of a mixture feature vector of the query sample of the query set and the position of a classification weight vector of each class in a projection space):
p
y
=
i
|
x
=
exp
-
D
i
x
∑
l
exp
-
D
l
x
,
(
6
)
for all
i
=
1
,
…
,
N
b
+
N
.” YOON [page 14, Algorithm 1] teaches that various parameters are updated in line 33 by minimizing the training loss via the optimizer (i.e., and learns classification weight vectors of the novel classes so as to minimize classification loss).)
[…] base classes […] for the query set in the projection space at the time of learning the query set in units of episodes. (YOON [page 2, section 2. Proposed Model] teaches: “The classification weight vector set W consists of base classifier weight vectors (i.e., base classes […] for the query set in the projection space), which are also pretrained on base categories, as well as novel classifier weight vectors, which vary with each training episode (i.e., at the time of learning the query set in units of episodes).” Examiner’s note: Under BRI, training in units of episodes can be interpreted as “for
t
=
1
…
T
d
o
” during the training of the model using query set
Q
(i.e., learning the query set), as shown in YOON [page 14, Algorithm 1].)
YOON is not relied upon for teaching:
a weight selection module that sequentially adds classification weight vectors of […] selected for the query set […]
However, SHI teaches: […] base classes selected for the query set […] (SHI [0111] teaches: “In more detail, a base classifier may be constructed by selecting one or more base classes and one or more novel classes from the multiple classes of training data TD.” SHI [0113] teaches: “For example, a fixed number of query instances (i.e., for the query set) may be sampled for each class (i.e., base classes selected), e.g., randomly, from the training data TD. […] Interestingly, both base class instances and novel class instances may be used as query instances […].” SHI [0119] teaches: “Thereby, fast learning for novel few-shot classes may be combined with slow learning for the classes of training dataset TD, e.g., the information from training dataset TD may be incorporated in the classifier instead of being forgotten.”)
Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of YOON and SHI before them, to include SHI’s selection of a subset of base classes for each episode in YOON’s incremental few-shot learning method. One would have been motivated to make such a combination in order to obtain a more accurate classification of the query instance (SHI [0036]).
YOON in view of SHI is not relied upon for teaching, but BELOUADAH teaches: a weight selection module that sequentially adds classification weight vectors of […] classes […] (BELOUADAH [page 5, section ] teaches: “New classifiers from previous states of Figure 2(a) are aggregated to represent past classes in each current state of Figure 2(b).” BELOUADAH [page 2, Figure 1] teaches a method that accumulates weight vectors (i.e., sequentially adds classification weight vectors of […] classes) in each incremental state to counter catastrophic forgetting. Examiner’s note: Under BRI, a weight selection module can be interpreted as the component that executes the accumulation of weight vectors in a classification layer
W
t
, as shown in BELOUADAH [page 2, Figure 1].)
Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of YOON, SHI, and BELOUADAH before them, to include BELOUADAH reuse of initial classifier weights in YOON and SHI’s incremental few-shot learning method. One would have been motivated to make such a combination in order to reuse normalized initial classifier weights to mitigate the effect of catastrophic forgetting (BELOUADAH [page 10, section 5 conclusion]).
Claim 2 is rejected under 35 U.S.C. 103 as being unpatentable over YOON in view of SHI and BELOUADAH as applied to claim 1 above, and further in view of ZHU ("Fewer is More: A Deep Graph Metric Learning Perspective Using Fewer Proxies") and ZHANG ("A Positive-biased Nearest Neighbour Algorithm for Imbalanced Classification"), hereafter ZHU and ZHANG, respectively.
Regarding Claim 2:
YOON in view of SHI and BELOUADAH teaches the elements of claim 1 as outlined above. YOON further teaches:
[…] the mixture feature vector of the query sample […] (YOON [page 2, section 2. Proposed Model] teaches: “MergeNet
r
, consisting of two relatively small fully-connected networks, creates a mixture of the base and novel features (i.e., the mixture feature vector). This mixture is what we call the task-adaptive representation (TAR) of the given input image (i.e., the mixture feature vector of the query sample).”)
wherein the learning unit classifies the query sample of the query set based on the distance between the position of the mixture feature vector of the query sample and the position of classification weight vectors […] classes in the projection space[…] (Finally, classification of the query x is carried out based on the posterior probabilities (i.e., classifies the query sample of the query set based on the distance between the position of the mixture feature vector of the query sample of the query set and the position of classification weight vectors […] classes in the projection space):
p
y
=
i
|
x
=
exp
-
D
i
x
∑
l
exp
-
D
l
x
,
(
6
)
for all
i
=
1
,
…
,
N
b
+
N
.” YOON [page 14, Algorithm 1] teaches computing a distance between
D
i
x
=
d
w
i
*
M
,
ω
p
r
e
⨀
f
θ
x
+
ω
m
e
t
a
⨀
g
a
θ
x
M
(i.e., distance between) between
ω
p
r
e
⨀
f
θ
x
+
ω
m
e
t
a
⨀
g
a
θ
x
M
(i.e., a mixture feature vector of the query sample) and
w
i
*
M
(i.e., and the position of classification weight vectors […] classes in the projection space). Examiner’s note: Under BRI, the learning unit can be interpreted as the component that carries out the classification using YOON’s equation (6).)
[…] and learns classification weight vectors of the novel classes so as to minimize classification loss. (YOON [page 14, Algorithm 1] computes a cross-entropy loss (i.e., classification loss) in line 30 by using equation (6) in YOON [page 4] to perform classification, and then updates all parameters by minimizing
L
t
r
a
i
n
via the optimizer (i.e., learns classification weight vectors of the novel classes so as to minimize classification loss).)
YOON in view of SHI and BELOUADAH is not relied upon for teaching:
[…] vectors of the selected predetermined number of neighboring classes […]
a neighbor selection unit that selects a predetermined number of classes located within a predetermined distance from the position of the […] vector of the query sample as neighboring classes in the projection space,
wherein the neighbor selection unit expands a target range until classes with correct labels are included and selects neighboring classes when the classes located within the predetermined distance from the position of the […] feature vector of the query sample do not include classes with correct labels in the projection space, and
However, ZHU teaches: […] vectors of the selected predetermined number of neighboring classes […] (ZHU [page 1, section 1 Introduction] teaches: “The core idea of DML is to learn an embedding space, where the embedded vectors (i.e., vectors) of similar samples are close to each other while those of dissimilar ones are far apart from each other.” ZHU [page 4, Generating A Directed Similarity Graph] teaches: “we directly use cosine similarity to efficiently capture the relationship between sample
x
i
s
and proxy
x
j
p
:
S
i
j
=
x
i
s
T
x
j
p
,
(
2
)
ZHU [page 4, Constructing k-NN subgraphs] teaches: “A common way is keeping the k-max values (i.e., selected predetermined number of neighboring classes) in each row of S to construct k-nearest neighbor (k-NN) subgraphs.”)
a neighbor selection unit that selects a predetermined number of classes located within a predetermined distance from the position of the […] vector of the query sample as neighboring classes in the projection space (ZHU [page 1, section 1 Introduction] teaches: “The core idea of DML is to learn an embedding space, where the embedded vectors of similar samples are close to each other while those of dissimilar ones are far apart from each other.” ZHU [page 4, Generating A Directed Similarity Graph] teaches: “we directly use cosine similarity to efficiently capture the relationship between sample
x
i
s
and proxy
x
j
p
:
S
i
j
=
x
i
s
T
x
j
p
,
(
2
)
ZHU [page 4, Constructing k-NN subgraphs] teaches: “A common way is keeping the k-max values in each row of S (i.e., that selects a predetermined number of classes) to construct k-nearest neighbor (k-NN) (i.e., located within a predetermined distance from the position of the […] vector of the query sample as neighboring classes in the projection space) subgraphs.” Examiner’s note: Under BRI, a neighbor selection unit can be interpreted as the component that computes the cosine similarity to capture the relationship, as shown in ZHU’s equation (2). Further, specification paragraphs [0074] and [0075] are the only identified locations where a predetermined distance is discussed. Moreover, Figures 11A, 11B, and 11C provide examples of how the neighboring classes of TAR are selected, and none uses a predetermined distance, bur rather a count (e.g., target classes), which corresponds to the predetermined number of classes. Specifically, the speechification does not provide examples of definition for a predetermined distance and uses neighbor counts to select the classes. Therefore, under BRI, a predetermined distance is taught by the k-max values used in ZHU to construct the k-nearest neighbor subgraphs.)
Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of YOON, SHI, BELOUADAH, and ZHU before them, to include ZHU’s use of k-max values for constructing k-nearest neighbor subgraphs in YOON, SHI, and BELOUADAH’s incremental few-shot learning method. One would have been motivated to make such a combination in order to better capture fine-grained neighborhood structures (ZHU [page 4, Constructing k-NN subgraphs]).
YOON in view of SHI, BELOUADAH, and ZHU is not relied upon for teaching, but ZHANG teaches: wherein the neighbor selection unit expands a target range until classes with correct labels are included and selects neighboring classes when the classes located within the predetermined distance from the position of the […] feature vector of the query sample do not include classes with correct labels in the projection space, and (ZHANG [page 2, section 1 Introduction] teaches: “Given a query instance and parameter k, if positive instances are scarce in the local neighbourhood of the query instance, we enlarge (i.e., expands a target range […] when the classes located within the predetermined distance from the position of the […] feature vector of the query sample do not include classes with correct labels in the projection space) the neighbourhood for classification decision.” ZHANG [page 4, section 3.1 Positive nearest neighbors] teaches: “Given query instance
t
and parameter
k
, to increase classification sensitivity to the positive class, we adjust the neighbourhood for classifying
t
so that it contains (i.e., and selects neighboring classes)
k
/
2
positive nearest neighbours of
t
– namely the
k
/
2
-PNN region of
t
.” Examiner’s note: Under BRI, until classes with correct labels are included can be interpreted as a condition in which no positive instances are present in the local neighborhood and positive instances are therefore scarce. Further, the neighbor selection unit can be interpreted as the component that enlarges the neighborhood when positive instances are scarce.)
Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of YOON, SHI, BELOUADAH, ZHU, and ZHANG before them, to include ZHANG’s neighborhood enlarging in YOON, SHI, BELOUADAH, and ZHU’s incremental few-shot learning method. One would have been motivated to make such a combination in order to increase classification sensitivity to the positive class (ZHANG [page 4, section 3.1 Positive nearest neighbors]).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Alvaro S Laham Bauzo whose telephone number is (571)272-5650. The examiner can normally be reached Mon-Fri 7:30 AM - 11:00 AM | 1:00 PM - 5:30 PM ET.
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, Usmaan Saeed can be reached on (571) 272-4046. 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.
/A.S.L./Examiner, Art Unit 2146
/USMAAN SAEED/Supervisory Patent Examiner, Art Unit 2146
Prosecution Insights