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 .
Status of Claims
Claims 1-20 are pending in this application.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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, 9, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Sener et al. ("Active learning for convolutional neural networks: A core-set approach." arXiv preprint arXiv:1708.00489 (2017) (hereinafter, “Sener”) in view of Balderer et al. (US 7,716,152 B2) (hereinafter, “Balderer”).
Regarding claim 1, Sener discloses a method comprising:
PNG
media_image1.png
378
664
media_image1.png
Greyscale
generating a plurality of feature vectors from a plurality of images, each feature vector summarizing an image of the plurality of images (Page 2 Second Paragraph “We further study the behavior of our proposed algorithm empirically for the problem of image classification using three different datasets.”; Page 7 Subsection 4.4 first paragraph "One of the critical design choices is the distance metric ∆(·, ·). We use the l2 distance between activations of the final fully-connected layer as the distance."; Page 8 end of last paragraph "We compute embeddings for all points using the features which are learned using the labelled examples and visualize the points"; Page 6 Algorithm 1)
;
selecting a seed vector (See Algorithm 1 Element 3) from the plurality of feature vectors and adding the seed vector to a coreset (Page 3 Section 3 second paragraph “We further consider an initial pool of data-points chosen uniformly at random as
s
0
=
s
0
j
∈
n
j
∈
[
m
]
”; Page 6 Algorithm 1: Initialize s = s0 )
computing a plurality of distance metrics, each distance metric measuring a distance between the seed vector and a given one of the plurality of feature vectors (See Algorithm 1 Element 5; Page 6 Subsection 4.3 “we assume an upper limit on the number of outliers Ξ such that our algorithm can choose not to cover at most Ξ unsupervised data points. This mixed integer program can be written as:
PNG
media_image2.png
210
780
media_image2.png
Greyscale
;
Page 7 Subsection 4.4 first paragraph "One of the critical design choices is the distance metric ∆(·, ·). We use the l2 distance between activations of the final fully-connected layer as the distance.");
adding a feature vector having a largest distance metric relative to all other ones of the plurality of distance metrics to the coreset (Page 6 Subsection 4.3 “the greedy algorithm shown in Algorithm 1 is proven to have a solution (s1) such that;
m
a
x
i
m
i
n
j
∈
s
1
∪
s
0
∆
x
i
,
x
j
≤
2
×
O
P
T
”; See Algorithm 1 Elements 5-6) ;
determining a next feature vector to use as a next seed vector [based on a nearest neighbor search] (Page 6 Subjection 4.3 first paragraph "We have so far provided an upper bound for the loss function of the core-set selection problem and showed that minimizing it is equivalent to the k-Center problem (minimax facility location (Wolf, 2011)) which can intuitively be defined as follows; choose b center points such that the largest distance between a data point and its nearest center is minimized."; See Algorithm 1 Element 4-6; Note that the algorithm is iterative as seen with the loop structure);
iteratively adding additional ones of the plurality of feature vectors to the coreset until a predefined coverage is achieved (See Algorithm 1 Elements 4-7; Note that the algorithm repeats until the selected set s reaches a desired size); and
generating a training set using images of the coreset (Page 2 second paragraph “In order to tailor an active learning method for the batch sampling case, we decided to define the active learning as core-set selection problem. Core-set selection problem aims to find a small subset given a large labeled dataset such that a model learned over the small subset is competitive over the whole dataset.”), wherein the training set is used to train a machine learning model to predict features of input images (Page 4 paragraph 3 "At each iteration, an active learning algorithm has two stages: 1. identifying a set of data-points and presenting them to an oracle to be labelled, and 2. training a classifier using both the new and the previously labeled data-points.").
However, Sener fails to teach based on a nearest neighbor search.
Balderer teaches based on a nearest neighbor search (Column 4 [lines 25-32] “The following is an algorithm in accordance with the invention for sequential nearest neighbor clustering:
for each p in S, do the following:
remove p from S
if Q is empty or the nearest neighbor distance between p
and each instance in Q>rmax, then
add p to Q.”).
Therefore, it would have been obvious to one of ordinary skill of the art before the effective filing date to modify Sener’s reference to include based on a nearest neighbor search taught by Balderer’s reference. The motivation for doing so would have been to reduce the time required to choose representative samples from a large data set as suggested by Balderer (see Balderer, Abstract).
Further, one skilled in the art could have combined the elements described above by known methods with no change to the respective functions, and the combination would have yielded nothing more that predictable results. Therefore, it would have been obvious to combine Balderer with Sener to obtain the invention specified in claim 1.
Regarding claim 9, Sener discloses a non-transitory computer-readable medium comprising memory with instructions encoded thereon, the instructions, when executed by one or more processors, causing the one or more processors to perform operations instructions (Page 4 third paragraph “At each iteration, an active learning algorithm has two stages: 1. identifying a set of data-points and presenting them to an oracle to be labelled, and 2. training a classifier using both the new and the previously labeled data-points. The second stage (training the classifier) can be done in a fully or weakly-supervised manner.” Examiner interprets the disclosed algorithm to be implemented through a stored instruction, as such algorithms require software instructions for training and execution.), the instructions comprising instructions to:
generate a plurality of feature vectors from a plurality of images, each feature vector summarizing an image of the plurality of images (Page 2 Second Paragraph “We further study the behavior of our proposed algorithm empirically for the problem of image classification using three different datasets.”; Page 7 Subsection 4.4 first paragraph "One of the critical design choices is the distance metric ∆(·, ·). We use the l2 distance between activations of the final fully-connected layer as the distance."; Page 8 end of last paragraph "We compute embeddings for all points using the features which are learned using the labelled examples and visualize the points"; Page 6 Algorithm 1)
PNG
media_image1.png
378
664
media_image1.png
Greyscale
;
select a seed vector (See Algorithm 1 Element 3) from the plurality of feature vectors and adding the seed vector to a coreset (Page 3 Section 3 second paragraph “We further consider an initial pool of data-points chosen uniformly at random as
s
0
=
s
0
j
∈
n
j
∈
[
m
]
”; Page 6 Algorithm 1: Initialize s = s0 );
compute a plurality of distance metrics, each distance metric measuring a distance between the seed vector and a given one of the plurality of feature vectors (See Algorithm 1 Element 5; Page 6 Subsection 4.3 “we assume an upper limit on the number of outliers Ξ such that our algorithm can choose not to cover at most Ξ unsupervised data points. This mixed integer program can be written as:
PNG
media_image2.png
210
780
media_image2.png
Greyscale
;
Page 7 Subsection 4.4 first paragraph "One of the critical design choices is the distance metric ∆(·, ·). We use the l2 distance between activations of the final fully-connected layer as the distance.");
add a feature vector having a largest distance metric relative to all other ones of the plurality of distance metrics to the coreset (Page 6 Subsection 4.3 “the greedy algorithm shown in Algorithm 1 is proven to have a solution (s1) such that;
m
a
x
i
m
i
n
j
∈
s
1
∪
s
0
∆
x
i
,
x
j
≤
2
×
O
P
T
”; See Algorithm 1 Elements 5-6);
determine a next feature vector to use as a next seed vector [based on a nearest neighbor search] (Page 6 Subjection 4.3 first paragraph "We have so far provided an upper bound for the loss function of the core-set selection problem and showed that minimizing it is equivalent to the k-Center problem (minimax facility location (Wolf, 2011)) which can intuitively be defined as follows; choose b center points such that the largest distance between a data point and its nearest center is minimized."; See Algorithm 1 Element 4-6; Note that the algorithm is iterative as seen with the loop structure);
iteratively add additional ones of the plurality of feature vectors to the coreset until a predefined coverage is achieved (See Algorithm 1 Elements 4-7; Note that the algorithm repeats until the selected set s reaches a desired size); and
generate a training set using images of the coreset (Page 2 second paragraph “In order to tailor an active learning method for the batch sampling case, we decided to define the active learning as core-set selection problem. Core-set selection problem aims to find a small subset given a large labeled dataset such that a model learned over the small subset is competitive over the whole dataset.”), wherein the training set is used to train a machine learning model to predict features of input images (Page 4 paragraph 3 "At each iteration, an active learning algorithm has two stages: 1. identifying a set of data-points and presenting them to an oracle to be labelled, and 2. training a classifier using both the new and the previously labeled data-points.").
However, Sener fails to teach based on a nearest neighbor search.
Balderer teaches based on a nearest neighbor search (Column 4 [lines 25-32] “The following is an algorithm in accordance with the invention for sequential nearest neighbor clustering:
for each p in S, do the following:
remove p from S
if Q is empty or the nearest neighbor distance between p
and each instance in Q>rmax, then
add p to Q.”).
Therefore, it would have been obvious to one of ordinary skill of the art before the effective filing date to modify Sener’s reference to include based on a nearest neighbor search taught by Balderer’s reference. The motivation for doing so would have been to reduce the time required to choose representative samples from a large data set as suggested by Balderer (see Balderer, Abstract).
Further, one skilled in the art could have combined the elements described above by known methods with no change to the respective functions, and the combination would have yielded nothing more that predictable results. Therefore, it would have been obvious to combine Balderer with Sener to obtain the invention specified in claim 9.
Regarding claim 17, Sener discloses a system comprising:
memory with instructions encoded thereon; and one or more processors that, when executing the instructions (Page 4 third paragraph “At each iteration, an active learning algorithm has two stages: 1. identifying a set of data-points and presenting them to an oracle to be labelled, and 2. training a classifier using both the new and the previously labeled data-points. The second stage (training the classifier) can be done in a fully or weakly-supervised manner.” Examiner interprets the disclosed algorithm to be implemented through a stored instruction, as such algorithms require software instructions for training and execution.), are caused to perform operations comprising:
generating a plurality of feature vectors from a plurality of images, each feature vector summarizing an image of the plurality of images (Page 2 Second Paragraph “We further study the behavior of our proposed algorithm empirically for the problem of image classification using three different datasets.”; Page 7 Subsection 4.4 first paragraph "One of the critical design choices is the distance metric ∆(·, ·). We use the l2 distance between activations of the final fully-connected layer as the distance."; Page 8 end of last paragraph "We compute embeddings for all points using the features which are learned using the labelled examples and visualize the points"; Page 6 Algorithm 1)
PNG
media_image1.png
378
664
media_image1.png
Greyscale
;
selecting a seed vector (See Algorithm 1 Element 3) from the plurality of feature vectors and adding the seed vector to a coreset (Page 3 Section 3 second paragraph “We further consider an initial pool of data-points chosen uniformly at random as
s
0
=
s
0
j
∈
n
j
∈
[
m
]
”; Page 6 Algorithm 1: Initialize s = s0 );
computing a plurality of distance metrics, each distance metric measuring a distance between the seed vector and a given one of the plurality of feature vectors (See Algorithm 1 Element 5; Page 6 Subsection 4.3 “we assume an upper limit on the number of outliers Ξ such that our algorithm can choose not to cover at most Ξ unsupervised data points. This mixed integer program can be written as:
PNG
media_image2.png
210
780
media_image2.png
Greyscale
;
Page 7 Subsection 4.4 first paragraph "One of the critical design choices is the distance metric ∆(·, ·). We use the l2 distance between activations of the final fully-connected layer as the distance.");
adding a feature vector having a largest distance metric relative to all other ones of the plurality of distance metrics to the coreset (Page 6 Subsection 4.3 “the greedy algorithm shown in Algorithm 1 is proven to have a solution (s1) such that;
m
a
x
i
m
i
n
j
∈
s
1
∪
s
0
∆
x
i
,
x
j
≤
2
×
O
P
T
”; See Algorithm 1 Elements 5-6);
determining a next feature vector to use as a next seed vector [based on a nearest neighbor search] (Page 6 Subjection 4.3 first paragraph "We have so far provided an upper bound for the loss function of the core-set selection problem and showed that minimizing it is equivalent to the k-Center problem (minimax facility location (Wolf, 2011)) which can intuitively be defined as follows; choose b center points such that the largest distance between a data point and its nearest center is minimized."; See Algorithm 1 Element 4-6; Note that the algorithm is iterative as seen with the loop structure);
iteratively adding additional ones of the plurality of feature vectors to the coreset until a predefined coverage is achieved (See Algorithm 1 Elements 4-7; Note that the algorithm repeats until the selected set s reaches a desired size); and
generating a training set using images of the coreset (Page 2 second paragraph “In order to tailor an active learning method for the batch sampling case, we decided to define the active learning as core-set selection problem. Core-set selection problem aims to find a small subset given a large labeled dataset such that a model learned over the small subset is competitive over the whole dataset.”), wherein the training set is used to train a machine learning model to predict features of input images (Page 4 paragraph 3 "At each iteration, an active learning algorithm has two stages: 1. identifying a set of data-points and presenting them to an oracle to be labelled, and 2. training a classifier using both the new and the previously labeled data-points.").
However, Sener fails to teach based on a nearest neighbor search.
Balderer teaches based on a nearest neighbor search (Column 4 [lines 25-32] “The following is an algorithm in accordance with the invention for sequential nearest neighbor clustering:
for each p in S, do the following:
remove p from S
if Q is empty or the nearest neighbor distance between p
and each instance in Q>rmax, then
add p to Q.”).
Therefore, it would have been obvious to one of ordinary skill of the art before the effective filing date to modify Sener’s reference to include based on a nearest neighbor search taught by Balderer’s reference. The motivation for doing so would have been to reduce the time required to choose representative samples from a large data set as suggested by Balderer (see Balderer, Abstract).
Further, one skilled in the art could have combined the elements described above by known methods with no change to the respective functions, and the combination would have yielded nothing more that predictable results. Therefore, it would have been obvious to combine Balderer with Sener to obtain the invention specified in claim 17.
Claims 2-3, 10-11, and 18-19 are rejected under 35 U.S.C. 103 as being unpatentable over Sener et al. ("Active learning for convolutional neural networks: A core-set approach." arXiv preprint arXiv:1708.00489 (2017) (hereinafter, “Sener”) in view of Balderer et al. (US 7,716,152 B2) (hereinafter, “Balderer”) as applied to claims 1, 9, and 17 above; and further in view of Guo et al. (US 2021/0125001 A1) (hereinafter, “Guo”).
Regarding claim 2, which claim 1 is incorporated, Sener and Balderer both fail to teach wherein the plurality of images comprises a first set of images formed from a same task as a task represented by the input images, a second set of images formed by a different task as the task represented by the input images but having at least a threshold similarity metric to the same task, and a third set of images having a high diversity and taken from a public data set.
Guo teaches wherein the plurality of images comprises a first set of images (images included in subset 412 in Paragraph [0068] equate to first set of images) formed from a same task as a task represented by the input images (Paragraph [0068] “subset 412 is defined as a clean subset with respect to the label of “cat.” Insert 410 shows a portion of the images included in subset 412. The images within insert 410 appear to cleanly depict images of a house cat. Subsets with a smaller local density metric include images having a larger diversity in visual appearance, which may include more irrelevant images with incorrect labels.”), a second set of images (images included in subset 422 in Paragraph [0068] equate to second set of images) formed by a different task as the task represented by the input images (Paragraph [0068] “Thus subset 422 is considered as a “noisy” subset, as indicated in insert 420, displaying images included in subset 422. The images displayed in insert 420 do include a greater variance in the depictions of a cat.”) but having at least a threshold similarity metric to the same task (Paragraph [0062] “For each training image, a local density metric (ρi) is determined based on the distance norms between each pair of training images.”; Paragraph [0063] “ For the ith image, the density metric indicates a number of images that are within a threshold distance of dc from the vector embedding of the ith training image in the feature space.”), and a third set of images (random images 430 in Paragraph [0068] equate to third set of images) having a high diversity and taken from a public data set (Paragraph [0068] “ As seen in FIG. 4A, images 410 from the clean subset 421 have very close visual appearance, while the noisiest subset 432 contains a number of random images 430 which are completely different from those images 410 in the clean subset 412.”).
Therefore, it would have been obvious to one of ordinary skill of the art before the effective filing date to modify Sener in view of Balderer to include wherein the plurality of images comprises a first set of images formed from a same task as a task represented by the input images, a second set of images formed by a different task as the task represented by the input images but having at least a threshold similarity metric to the same task, and a third set of images having a high diversity and taken from a public data set taught by Guo’s reference. The motivation for doing so would have been to accumulate knowledge and enable the capture of the main underlying structure of the image features as suggested by Guo (see Guo, Paragraph [0072]).
Further, one skilled in the art could have combined the elements described above by known methods with no change to the respective functions, and the combination would have yielded nothing more that predictable results. Therefore, it would have been obvious to combine Guo with Sener and Balderer to obtain the invention specified in claim 2.
Regarding claim 3, which claim 2 is incorporated, Sener and Balderer both fail to teach wherein separate coresets are formed for each of the first set of images, the second set of images, and the third set of images, the first set of images corresponding to a first coreset, the second set of images corresponding to a second coreset, and the third set of images corresponding to a third coreset.
Guo teaches wherein separate coresets are formed for each of the first set of images, the second set of images, and the third set of images (Paragraph [0067] “The clusters of images are indicated as 412, 422, and 432. The images within a particular cluster are grouped into a corresponding subset of training images.”), the first set of images corresponding to a first coreset (subset 412 in Paragraph [0068] equates to first coreset) (Paragraph [0068] “subset 412 is defined as a clean subset with respect to the label of “cat.” Insert 410 shows a portion of the images included in subset 412. The images within insert 410 appear to cleanly depict images of a house cat. Subsets with a smaller local density metric include images having a larger diversity in visual appearance, which may include more irrelevant images with incorrect labels.”), the second set of images corresponding to a second coreset (subset 422 in Paragraph [0068] equates to second coreset) (Paragraph [0068] “Thus subset 422 is considered as a “noisy” subset, as indicated in insert 420, displaying images included in subset 422. The images displayed in insert 420 do include a greater variance in the depictions of a cat.”), and the third set of images corresponding to a third coreset (subset 432 in Paragraph [0068] equates to third coreset) (Paragraph [0068] “ As seen in FIG. 4A, images 410 from the clean subset 421 have very close visual appearance, while the noisiest subset 432 contains a number of random images 430 which are completely different from those images 410 in the clean subset 412.”).
Therefore, it would have been obvious to one of ordinary skill of the art before the effective filing date to modify Sener in view of Balderer to include wherein separate coresets are formed for each of the first set of images, the second set of images, and the third set of images, the first set of images corresponding to a first coreset, the second set of images corresponding to a second coreset, and the third set of images corresponding to a third coreset taught by Guo’s reference. The motivation for doing so would have been to accumulate knowledge and enable the capture of the main underlying structure of the image features as suggested by Guo (see Guo, Paragraph [0072]).
Further, one skilled in the art could have combined the elements described above by known methods with no change to the respective functions, and the combination would have yielded nothing more that predictable results. Therefore, it would have been obvious to combine Guo with Sener and Balderer to obtain the invention specified in claim 3.
Regarding claim 10 (drawn to a non-transitory storage medium), claim 10 is rejected the same as claim 2 and the arguments similar to that presented above for claim 2 are equally applicable to the claim 10, and all the other limitations similar to claim 2 are not repeated herein, but incorporated by reference.
Regarding claim 11 (drawn to a non-transitory storage medium), claim 11 is rejected the same as claim 3 and the arguments similar to that presented above for claim 3 are equally applicable to the claim 11, and all the other limitations similar to claim 3 are not repeated herein, but incorporated by reference.
Regarding claim 18 (drawn to a system), claim 18 is rejected the same as claim 2 and the arguments similar to that presented above for claim 2 are equally applicable to the claim 18, and all the other limitations similar to claim 2 are not repeated herein, but incorporated by reference.
Regarding claim 19 (drawn to a system), claim 19 is rejected the same as claim 3 and the arguments similar to that presented above for claim 3 are equally applicable to the claim 19, and all the other limitations similar to claim 3 are not repeated herein, but incorporated by reference.
Claims 4, 12, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Sener et al. ("Active learning for convolutional neural networks: A core-set approach." arXiv preprint arXiv:1708.00489 (2017) (hereinafter, “Sener”) in view of Balderer et al. (US 7,716,152 B2) (hereinafter, “Balderer”); Guo et al. (US 2021/0125001 A1) (hereinafter, “Guo”) as applied to claims 3, 11, and 19 above and further in view of Narayanan et al. (US 11,087,883 B1) (hereinafter, “Narayanan”)
Regarding claim 4, which claim 3 is incorporated, Sener and Balderer both fail to teach wherein the training set comprises all of the images in the first coreset and the second coreset, and wherein the training set comprises a truncated amount of images in the third coreset based on a size of the first coreset and the second coreset.
Guo teaches wherein the training set comprises all of the images in the first coreset and the second coreset (Paragraph [0071] “MVM_v1 482 is trained, via supervised learning methods, by employing the combination of the clean subset of training images 452 and the noisy subset of training images 454.”).
Therefore, it would have been obvious to one of ordinary skill of the art before the effective filing date to modify Sener in view of Balderer to include wherein the training set comprises all of the images in the first coreset and the second coreset taught by Guo’s reference. The motivation for doing so would have been to accumulate knowledge and enable the capture of the main underlying structure of the image features as suggested by Guo (see Guo, Paragraph [0072]).
However, Sener, Balderer, and Guo fail to teach wherein the training set comprises a truncated amount of images in the third coreset based on a size of the first coreset and the second coreset.
Narayanan teaches wherein the training set comprises a truncated amount of images in the third coreset based on a size of the first coreset and the second coreset (Abstract “allocating an equal number of the training data samples of the first subset into each of the plurality of mini-corpora of training data samples”; Column 8 [lines 19-34] “the number of generated mini-corpora may be determined based on dividing the number of majority data sample images in the imbalanced training data corpus by the number of minority class data sample images in the imbalanced training data corpus. For example, if there are 1000 majority class sample images and only 10 minority class images in the imbalanced training data corpus… S222 may function to divide the 1000 majority class images by the 10 minority class images creating a total of 100 mini-corpora.”).
Therefore, it would have been obvious to one of ordinary skill of the art before the effective filing date to modify Sener in view of Balderer; and further in view of Guo to include wherein the training set comprises a truncated amount of images in the third coreset based on a size of the first coreset and the second coreset taught by Narayanan’s reference. The motivation for doing so would have been to convert imbalanced training data to class-balanced training data samples as suggested by Guo (see Narayanan, Column 8 [lines 5-7]).
Further, one skilled in the art could have combined the elements described above by known methods with no change to the respective functions, and the combination would have yielded nothing more that predictable results. Therefore, it would have been obvious to combine Narayanan with Sener, Balderer, and Guo to obtain the invention specified in claim 4.
Regarding claim 12 (drawn to a non-transitory storage medium), claim 12 is rejected the same as claim 4 and the arguments similar to that presented above for claim 4 are equally applicable to the claim 12, and all the other limitations similar to claim 4 are not repeated herein, but incorporated by reference.
Regarding claim 20 (drawn to a system), claim 20 is rejected the same as claim 4 and the arguments similar to that presented above for claim 4 are equally applicable to the claim 20, and all the other limitations similar to claim 4 are not repeated herein, but incorporated by reference.
Claims 6-8 and 14-16 are rejected under 35 U.S.C. 103 as being unpatentable over Sener et al. ("Active learning for convolutional neural networks: A core-set approach." arXiv preprint arXiv:1708.00489 (2017) (hereinafter, “Sener”) in view of Balderer et al. (US 7,716,152 B2) (hereinafter, “Balderer”) as applied to claims 1, 9, and 17 above; and further in view of Perera et al. (US 11,514,691 B2) (hereinafter, “Perera”).
Regarding claim 6, which claim 1 is incorporated, Sener and Balderer both fail to teach wherein the predefined coverage is specified by a user.
Perera teaches wherein the predefined coverage is specified by a user (Column 6 [lines 15-16] “The size of a training set may be based on input by a user and/or may be predetermined according to a particular machine learning task and goal.”).
Therefore, it would have been obvious to one of ordinary skill of the art before the effective filing date to modify Sener in view of Balderer to include wherein the predefined coverage is specified by a user taught by Perera’s reference. The motivation for doing so would have been to enable a user to manage the generation of training sets and the training of machine learning models as suggested by Perera (see Perera, Column 3 [lines 31-34]).
Further, one skilled in the art could have combined the elements described above by known methods with no change to the respective functions, and the combination would have yielded nothing more that predictable results. Therefore, it would have been obvious to combine Perera with Sener and Balderer to obtain the invention specified in claim 6.
Regarding claim 7, which claim 1 is incorporated, Sener and Balderer both fail to teach wherein as the additional ones of the plurality of feature vectors are added to the coreset, an order in which they are added is stored to memory.
Perera teaches wherein as the additional ones of the plurality of feature vectors are added to the coreset, an order in which they are added is stored to memory (End of column 5 lines 67 continuing to Column 6 line 8 “selection module 165 orders documents of a topic based on the distance of each document's vector representation to the topic, which may be defined as a centroid of a cluster or a predetermined point that is based on provided scalar values. A position vector may then be generated for each document based on the position of the document in the order for a topic (e.g., distance of a document to a topic), and documents may then be selected based on the ordering of position vectors.”; Column 6 [lines [25-30] “in one iteration, a document may be randomly selected from each cluster, in a next iteration, a document closest to the centroid of the document's cluster may be selected, and in a next iteration, a document farthest from the centroid of the document's cluster may be selected.”).
Therefore, it would have been obvious to one of ordinary skill of the art before the effective filing date to modify Sener in view of Balderer to include wherein as the additional ones of the plurality of feature vectors are added to the coreset, an order in which they are added is stored to memory taught by Perera’s reference. The motivation for doing so would have been to select samples from clusters to improve the semantic diversity of the sampled document as suggested by Perera (see Perera, Column 2 [lines 25-27]).
Further, one skilled in the art could have combined the elements described above by known methods with no change to the respective functions, and the combination would have yielded nothing more that predictable results. Therefore, it would have been obvious to combine Perera with Sener and Balderer to obtain the invention specified in claim 7.
Regarding claim 8, which claim 7 is incorporated, Sener and Balderer both fail to teach wherein the training set is at least partially generated based on the order.
Perera teaches wherein the training set is at least partially generated based on the order (Column 6 [lines 4-11] “A position vector may then be generated for each document based on the position of the document in the order for a topic (e.g., distance of a document to a topic), and documents may then be selected based on the ordering of position vectors. Documents that are selected by selection module 165 may be assigned to a training set of documents that can be used to train a machine learning model.”).
Therefore, it would have been obvious to one of ordinary skill of the art before the effective filing date to modify Sener in view of Balderer to include wherein the training set is at least partially generated based on the order taught by Perera’s reference. The motivation for doing so would have been to select samples from clusters to improve the semantic diversity of the sampled document as suggested by Perera (see Perera, Column 2 [lines 25-27]).
Further, one skilled in the art could have combined the elements described above by known methods with no change to the respective functions, and the combination would have yielded nothing more that predictable results. Therefore, it would have been obvious to combine Perera with Sener and Balderer to obtain the invention specified in claim 8.
Regarding claim 14 (drawn to a non-transitory storage medium), claim 14 is rejected the same as claim 6 and the arguments similar to that presented above for claim 6 are equally applicable to the claim 14, and all the other limitations similar to claim 6 are not repeated herein, but incorporated by reference.
Regarding claim 15 (drawn to a non-transitory storage medium), claim 15 is rejected the same as claim 4 and the arguments similar to that presented above for claim 7 are equally applicable to the claim 15, and all the other limitations similar to claim 7 are not repeated herein, but incorporated by reference.
Regarding claim 16 (drawn to a non-transitory storage medium), claim 16 is rejected the same as claim 8 and the arguments similar to that presented above for claim 8 are equally applicable to the claim 16, and all the other limitations similar to claim 8 are not repeated herein, but incorporated by reference.
Allowable Subject Matter
Claims 5 and 13 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Claims 5 and 13 contain subject matter that is not disclosed or made obvious in the cited art.
In regards to claim 5, when considering claim 5 as a whole, the feature highlighted below are considered an improvement over the prior art and have not been found to be anticipated or rendered obvious by a combination of the prior art:
“[…] performing the nearest neighbor search with respect to the feature vector having the largest distance metric relative to all other ones of the plurality of distance metrics to the coreset; and
determining the next feature vector to be, as indicated by the nearest neighbor search, one having a lowest cosine similarity to the feature vector having the largest distance metric relative to all other ones of the plurality of distance metrics to the coreset.”.
In regards to claim 13, when considering claim 13 as a whole, the feature highlighted below are considered an improvement over the prior art and have not been found to be anticipated or rendered obvious by a combination of the prior art:
“[…] perform the nearest neighbor search with respect to the feature vector having the largest distance metric relative to all other ones of the plurality of distance metrics to the coreset; and
determine the next feature vector to be, as indicated by the nearest neighbor search, one having a lowest cosine similarity to the feature vector having the largest distance metric relative to all other ones of the plurality of distance metrics to the coreset.”.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Kakizaki et al. (US 2025/0166356 A1) discloses training a feature amount extractor such that distances between images in a feature space become close to a target distance.
Saraee et al. (US 2022/0335256 A1) discloses training a model extracting feature from training images, calculating feature weights, and generating a candidate image performance score.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to UROOJ FATIMA whose telephone number is (571)272-2096. The examiner can normally be reached M-F 8:00-5:00.
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, Henok Shiferaw can be reached at (571) 272-4637. 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.
/UROOJ FATIMA/Examiner, Art Unit 2676
/Henok Shiferaw/Supervisory Patent Examiner, Art Unit 2676
Prosecution Insights