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 .
Response to Amendment
The office action is in response to Applicant’s amendment filed 07/15/2026 which has been entered and made of record. Claims 1, 4, 8, 11, and 15-20 have been amended. No claim has been newly added or canceled. Claims 1-20 are pending in the application.
Response to Arguments
Applicant’s arguments, see page 9, filed 07/15/2026, with respect to the rejection of under 35 U.S.C. 112(b) have been fully considered and are persuasive. The 112(b) rejection of claims 4, 11, and 18 have been withdrawn.
Applicant's arguments, see pages 9-11, filed 07/15/2026, with respect to the rejection of claims 1-20 under 35 U.S.C. 103 have been fully considered but they are not persuasive.
Applicant argues Claims 8 and 15 recite similar features to claim 1.
Examiner responds that Jabbour further teaches the amended limitation “wherein the bias evaluation result represents whether the to-be-evaluated model has a bias against an evaluation image comprising the to-be-verified factor” in Page 3, Section 2.1 “we focus on a type of bias that arises when training data are skewed with respect to a particular attribute (i.e., a characteristic that divides patients into subgroups), resulting in undesirable model performance when the model is applied to test data that do not exhibit the same kind of skew”, where test data consists of images of X-rays comprising particular attributes, corresponding to evaluation image comprising the to-be-verified factor.
Conclusions: The rejections set in the previous Office Action are shown to have been
proper, and the claims are rejected below. New citations and parenthetical remarks can be
considered new grounds of rejection and such new grounds of rejection are necessitated by the
Applicant's amendments to the claims. Therefore, the present Office Action is made final.
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Jabbour et al
(Deep Learning Applied to Chest X-Rays: Exploiting and Preventing Shortcuts, hereinafter
Jabbour) and Zhu et al (Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial
Networks, hereinafter Zhu).
Regarding claim 1, Jabbour teaches A bias evaluation method (Page 2, Par 2 “In this
paper, we examine the extent to which potential biases related to age, sex, race, body mass
index (BMI), different treatments, and image preprocessing can be exploited by CNNs. ”),
applied to an electronic device, the method comprising: obtaining a to-be-verified factor
existing in a plurality of evaluation images that are used to perform bias evaluation on a to-
be-evaluated model; (Page 6 Par 2-3 “In our analysis, we assume the spurious correlation is
modeled by a bias attribute b ∈ {0,1}”, “We consider a scenario in which in the training and
validation data associated with the target task yt and attribute b may be highly correlated”,
Page 7 Par 3 “AHRF dataset. We considered a cohort of 1,296 patients admitted to a large
acute care hospital center during 2017-2018 who developed AHRF. We analyzed chest X-rays
acquired closest to the time of AHRF onset”, Page 8 Par 2 “For each patient, we extracted
several attributes that could serve as b, including body mass index (BMI), age, sex, race, and the
presence of a pacemaker.”, where attributes that could serve as b correspond to a to-be-verified factor) classifying the plurality of evaluation images based on the to-be-verified factor to obtain a first target evaluation image set, wherein the first target evaluation image set comprises at least one of a first evaluation image set comprising the to-be-verified factor or a second evaluation image set not comprising the to-be-verified factor (Page 8, Par 2 “For each patient, we extracted several attributes that could serve as b, including body mass index (BMI), age, sex, race, and the presence of a pacemaker. Demographic variables were extracted from the electronic health record, while images were manually labeled for the presence of a pacemaker. We dichotomized each variable: i) BMI ≥ 30, ii) age ≥ 63 years (the median age of our dataset), iii) sex = female, iv) race = Black, and v) pacemaker present.”, Page 8 Par 3 “CheXpert was provided with age and sex, which we dichotomized the same way as described above. MIMIC-CXR was provided with age, sex, insurance provider, and marital status. We dichotomized each variable”, where the plurality of evaluation images corresponds to the input dataset, the first target evaluation set corresponds to the labeled images, and the first and second evaluation sets are the dichotomized sets); inputting the first target evaluation image set and the second target evaluation image set into the to-be-evaluated model for inference to obtain a target inference result (Page 10 Sect 4.3 “We initialized models on either ImageNet or by pretraining on the MIMIC-CXR and CheXpert datasets”); and outputting a bias evaluation result of the to-be- evaluated model based on the target inference result, wherein the bias evaluation result represents whether has a bias against an evaluation image comprising the to-be-verified factor (Page 3 Sect 2.1 “We assume access to a labeled training set, and a separate held-out test set. In this work, we focus on a type of bias that arises when training data are skewed with respect to a particular attribute (i.e., a characteristic that divides patients into subgroups), resulting in undesirable model performance when the model is applied to test data that do not exhibit the same kind of skew”, Page 10 Sect 4.3 AHRF dataset training setup “Validation and test images were center cropped to 512×512.”, Page 13 Par 3 “these results (for the most part) show that attributes such as demographics and treatment are easy to infer based on a chest X-ray.”, Page 20 Sect 6 “we showed that in the presence of spurious correlations, deep learning models can exploit potential shortcuts associated with biased attributes”)
Jabbour teaches injecting a bias feature into an image-set through an image preprocessing step (Page 11, Par 5 “in a series of follow-up experiments repeat our analyses
with a synthetic bias injected via an image preprocessing step, answering the final question”),
but fails to explicitly teach performing style conversion on the first target evaluation image set based on the to-be-verified factor to obtain a second target evaluation image set, wherein style conversion performed on the first evaluation image set is implemented by removing the to-be-verified factor, and style conversion performed on the second evaluation image set is implemented by adding the to-be- verified factor. In related field of endeavor, Zhu teaches performing style conversion on the first target evaluation image set based on the to-be-verified factor to obtain a second target evaluation image set, wherein style conversion performed on the first evaluation image set is implemented by removing the to-be-verified factor, and style conversion performed on the second evaluation image set is implemented by adding the to-be-verified factor. (Sect 5.2 Applications, Par 3 “Object transfiguration (Figure 13) The model is trained to translate one object class from ImageNet [5] to another (each class contains around 1000 training images)”, Figure 4 “The input images x, output images G(x) and the reconstructed images F(G(x)) from various experiments. From top to bottom: photo ↔ Cezanne, horses ↔ zebras, winter → summer Yosemite, aerial photos ↔ Google maps.”, where an image factor can be added to a photo using G(x), and removed using F(G(x)) )
It would have been obvious to one of ordinary skill in the art to have modified Jabbour
to include performing style conversion on the first target evaluation image set based on the to-
be-verified factor to obtain a second target evaluation image set, wherein style conversion
performed on the first evaluation image set is implemented by removing the to-be-verified
factor, and style conversion performed on the second evaluation image set is implemented by
adding the to-be-verified factor as taught by Zhu. Doing so would provide a method for the
image preprocessing step for injecting bias into an image set (Sect 2. Related Work, Neural
Style Transfer “Therefore, our method can be applied to other tasks, such as Input 𝑥 painting→
photo, object transfiguration, etc”).
Regarding claim 2, Jabbour and Zhu teach the method according to claim 1, and Jabbour further teaches wherein the bias evaluation result comprises at least one of: information about whether the to-be-verified factor is a factor that causes the bias of the to-be-evaluated model (Page 3 Sect 2.1 “We assume access to a labeled training set, and a separate held-out test set. In this work, we focus on a type of bias that arises when training data are skewed with respect to a particular attribute (i.e., a characteristic that divides patients into subgroups), resulting in undesirable model performance when the model is applied to test data that do not exhibit the same kind of skew”, Page 10 Sect 4.3 AHRF dataset training setup “Validation and test images were center cropped to 512×512.”, Page 13 Par 3 “these results (for the most part) show that attributes such as demographics and treatment are easy to infer based on a chest X-ray.”, Page 20 Sect 6 “we showed that in the presence of spurious correlations, deep learning models can exploit potential shortcuts associated with biased attributes”); a difference image in the first target evaluation image set, wherein the difference image is an evaluation image that is in the first target evaluation image set and which inference result is different from an inference result of at least one corresponding converted image that is obtained through style conversion and that is in the second target evaluation image set, and both the inference result of the difference image and the inference result of the converted image are inference results output by the to-be-evaluated model; a converted image that is obtained by performing style conversion on each difference image and that is comprised in the second target evaluation set; an inference result of the to-be-evaluated model for each difference image; an inference result of the to-be-evaluated model for each converted image; or a proportion of a difference image in the first target evaluation image set in the plurality of evaluation images.
Regarding claim 3, Jabbour and Zhu teach the method according to claim 2, and Jabbour further teaches wherein the to-be-verified factor is determined based on a background and a foreground of each original image in a verification dataset, and the to-be-verified factor corresponds to an image feature in the background. (Jabbour Page 13 Figure 3, “(a) Given a chest X-ray of patient with age≥60, the model localizes around joints and bones, suggesting that it is using bone density in its prediction. (b) When tasked to predict insurance, the model focuses around bone edges and joints, further suggesting that it may be using age as an indicator of insurance status.”, Jabbour Page 8 Par 2 “For each patient, we extracted several attributes that could serve as b, including body mass index (BMI), age, sex, race, and the presence of a pacemaker. Demographic variables were extracted from the electronic health record, while images were manually labeled for the presence of a pacemaker. We dichotomized each variable: i) BMI ≥ 30, ii) age ≥ 63 years (the median age of our dataset), iii) sex = female, iv) race = Black, and v) pacemaker present.”, where a to-be-verified factor corresponding to an image feature in the background is age, where the model localizes on joints and bones.)
Regarding claim 4, Jabbour and Zhu teach the method according to claim 3. Zhu further
teaches wherein style conversion of an image is implemented by using an image style
conversion model, and the image style conversion model is obtained through training based
on the first evaluation image set and the second evaluation image set (5.2 Applications Object
Transfiguration “The model is trained to translate one object class from ImageNet [5] to
another (each class contains around 1000 training images).”, Figure 4 “The input images x,
output images G(x) and the reconstructed images F(G(x)) from various experiments. From top
to bottom: photo↔Cezanne, horses↔zebras, winter→summer Yosemite, aerial
photos↔Google maps.”, Figure 13 Description: “In the top two rows, we show results on
object transfiguration between horses and zebras, trained on 939 images from the wild horse
class and 1177 images from the zebra class in Imagenet [5].”, where the first evaluation image
set corresponds to the 939 images, and the second evaluation image set corresponds to the
1177 images); and wherein the image style conversion model is configured to remove the to-
be-verified factor from an image comprising the to-be-verified factor, and add the to-be-verified factor to an image not comprising the to-be-verified factor, and wherein the to-be-verified factor is an image feature (Figure 4 “The input images x, output images G(x) and the reconstructed images F(G(x)) from various experiments. From top to bottom: photo↔Cezanne, horses↔zebras, winter→summer Yosemite, aerial photos↔Google maps.”, where an image factor can be added to a photo using G(x), and removed using F(G(x)) ).
It would have been obvious to one of ordinary skill in the art prior to the time of filing to
have modified Jabbour to include wherein style conversion of an image is implemented by
using an image style conversion model, and the image style conversion model is obtained
through training based on the first evaluation image set and the second evaluation image set ;
and wherein the image style conversion model is configured to remove the to-be-verified factor
from an image comprising the verification factor, and add the verification factor to an image
not comprising the verification factor, and wherein the verification factor is an image feature as
taught by Zhu. Doing so would provide a method for the image preprocessing step for injecting
bias into an image set (2. Related Work, Neural Style Transfer “Therefore, our method can be
applied to other tasks, such as Input 𝑥 painting→ photo, object transfiguration, etc”).
Regarding claim 5, Jabbour and Zhu teach the method according to claim 4. Zhu further
teaches wherein the first evaluation image set corresponds to a first classification label, the
second evaluation image set corresponds to a second classification label different from the
first classification label (Figure 13 Description: “In the top two rows, we show results on object
transfiguration between horses and zebras, trained on 939 images from the wild horse class
and 1177 images from the zebra class in Imagenet [5].”, where the first evaluation image set corresponds to the 939 images, the first classification label corresponds to the wild horse class,
the second evaluation image set corresponds to the 1177 images, and the second classification
label corresponds to the zebra class), and the image style conversion model is obtained
through training based on an image in the first evaluation image set, the first classification
label, an image in the second evaluation image set, and the second classification label (5.2
Applications, Object transfiguration: “The model is trained to translate one object class from
ImageNet [5] to another (each class contains around 1000 training images).”).
It would have been obvious to one of ordinary skill in the art prior to the time of filing to
have modified Jabbour to include wherein the first evaluation image set corresponds to a first
classification label, the second evaluation image set corresponds to a second classification label
different from the first classification label and the image style conversion model is obtained
through training based on an image in the first evaluation image set, the first classification
label, an image in the second evaluation image set, and the second classification label as taught
by Zhu. Doing so would provide a method for the image preprocessing step for injecting bias
into an image set (2. Related Work, Neural Style Transfer “Therefore, our method can be
applied to other tasks, such as Input 𝑥 painting→ photo, object transfiguration, etc”).
Regarding claim 6, Jabbour and Zhu teach the method according to claim 3, and Jabbour further teaches wherein the method further comprises: receiving the verification dataset and the to-be-evaluated model that are input by a user (Jabbour Page 3 Par 3 “We apply a transfer learning approach in which we first train a deep learning model to learn pneumonia from an unskewed dataset. Then, we use the model features to predict congestive heart failure in a skewed dataset.”, Jabbour Page 7 Par 4 “We considered a cohort of 1,296 patients admitted to a large acute care hospital center during 2017-2018 who developed AHRF. We analyzed chest X-rays acquired closest to the time of AHRF onset. Each patient had a corresponding study, containing one or more chest X-rays taken at the same time. Within each study, we considered only frontal images”).
Regarding claim 7, Jabbour and Zhu teach the method according to claim 6, and Jabbour further teaches wherein the method further comprises: receiving the to-be-verified factor that is input by the user, wherein the to-be-verified factor is an image feature or an identifier indicating an image feature. (Page 8 Par 2 “For each patient, we extracted several attributes that could serve as b, including body mass index (BMI), age, sex, race, and the presence of a pacemaker. Demographic variables were extracted from the electronic health record, while images were manually labeled for the presence of a pacemaker. We dichotomized each variable: i) BMI ≥ 30, ii) age ≥ 63 years (the median age of our dataset), iii) sex = female, iv) race = Black, and v) pacemaker present”).
Regarding claim 8, Jabbour teaches an electronic device, comprising: at least one
processor; and at least one memory coupled to the at least one processor and storing
programming instructions for execution by the at least one processor to perform operations
comprising (Page 5 Par 4 “Given a chest X-ray image X ∈ Rd×d, we consider the task of
predicting some target diagnosis yt ∈ {0,1}. We assume a convolutional neural network (CNN)
f(e(X;θ);w) is the composition of two functions:”, where a convolutional neural network isstored in memory and executed by a processor): obtaining a to-be-verified factor existing in a
plurality of evaluation images that are used to perform bias evaluation on a to-be-evaluated
model; (Page 6 Par 2-3 “In our analysis, we assume the spurious correlation is modeled by a
bias attribute b ∈ {0,1}”, “We consider a scenario in which in the training and validation data
associated with the target task yt and attribute b may be highly correlated”, Page 7 Par 3 “AHRF
dataset. We considered a cohort of 1,296 patients admitted to a large acute care hospital
center during 2017-2018 who developed AHRF. We analyzed chest X-rays acquired closest to
the time of AHRF onset”, Page 8 Par 2 “For each patient, we extracted several attributes that
could serve as b, including body mass index (BMI), age, sex, race, and the presence of a
pacemaker.” ) classifying the plurality of evaluation images based on the to-be-verified factor
to obtain a first target evaluation image set, wherein the first target evaluation image set
comprises at least one of a first evaluation image set comprising the to-be-verified factor or a
second evaluation image set not comprising the to-be-verified factor (Page 8, Par 2 “For each
patient, we extracted several attributes that could serve as b, including body mass index (BMI),
age, sex, race, and the presence of a pacemaker. Demographic variables were extracted from
the electronic health record, while images were manually labeled for the presence of a
pacemaker. We dichotomized each variable: i) BMI ≥ 30, ii) age ≥ 63 years (the median age of
our dataset), iii) sex = female, iv) race = Black, and v) pacemaker present.”, where the plurality
of evaluation images corresponds to the input dataset, the first target evaluation set
corresponds to the labeled images, and the first and second evaluation set are the set
dichotomized by the pacemaker variable); inputting the first target evaluation image set and
the second target evaluation image set into the to-be-evaluated model for inference to obtain a target inference result (Page 10 Sect 4.3 “We initialized models on either ImageNet or by pretraining on the MIMIC-CXR and CheXpert datasets”, Page 14, Par 4 “To measure the extent to which the injected bias affects performance, we compare the learned model’s performance, Skewed Training Data, to the performance of a model trained on data that more closely mimics the test set, Unskewed Training Data. To generate the unskewed training dataset, we subsampled the data so that b was uncorrelated with yt. Again, to facilitate comparisons, we kept the size of the training data similar”); and outputting a bias evaluation result of the to-be-evaluated model based on the target inference result, wherein the bias evaluation result represents whether has a bias against an evaluation image comprising the to-be-verified factor (Page 3 Sect 2.1 “We assume access to a labeled training set, and a separate held-out test set. In this work, we focus on a type of bias that arises when training data are skewed with respect to a particular attribute (i.e., a characteristic that divides patients into subgroups), resulting in undesirable model performance when the model is applied to test data that do not exhibit the same kind of skew”, Page 10 Sect 4.3 AHRF dataset training setup “Validation and test images were center cropped to 512×512.”, Page 13 Par 3 “these results (for the most part) show that attributes such as demographics and treatment are easy to infer based on a chest X-ray.”, Page 20 Sect 6 “we showed that in the presence of spurious correlations, deep learning models can exploit potential shortcuts associated with biased attributes”)
Jabbour teaches injecting a bias feature into an image-set through an image preprocessing step (Page 11, Par 5 “in a series of follow-up experiments repeat our analyses
with a synthetic bias injected via an image preprocessing step, answering the final question”),
but fails to explicitly teach performing style conversion on the first target evaluation image set
based on the to-be-verified factor to obtain a second target evaluation image set, wherein
style conversion performed on the first evaluation image set is implemented by removing the
to-be-verified factor, and style conversion performed on the second evaluation image set is
implemented by adding the to-be-verified factor. In related field of endeavor, Zhu teaches
performing style conversion on the first target evaluation image set based on the to-be-verified factor to obtain a second target evaluation image set, wherein style conversion performed on
the first evaluation image set is implemented by removing the to-be-verified factor, and style
conversion performed on the second evaluation image set is implemented by adding the to-be-
verified factor. (Sect 5.2 Applications, Par 3 “Object transfiguration (Figure 13) The model is
trained to translate one object class from ImageNet [5] to another (each class contains around
1000 training images)”, Figure 4 “The input images x, output images G(x) and the reconstructed
images F(G(x)) from various experiments. From top to bottom: photo ↔ Cezanne, horses ↔
zebras, winter → summer Yosemite, aerial photos ↔ Google maps.”, where an image factor
can be added to a photo using G(x), and removed using F(G(x)))
It would have been obvious to one of ordinary skill in the art to have modified Jabbour
to include performing style conversion on the first target evaluation image set based on the to-
be-verified factor to obtain a second target evaluation image set, wherein style conversion
performed on the first evaluation image set is implemented by removing the to-be-verified
factor, and style conversion performed on the second evaluation image set is implemented by
adding the to-be-verified factor as taught by Zhu. Doing so would provide a method for the
image preprocessing step for injecting bias into an image set (Sect 2. Related Work, Neural Style
Transfer “Therefore, our method can be applied to other tasks, such as Input 𝑥 painting→
photo, object transfiguration, etc”).
Regarding claim 9, the device claim 9 is similar in scope to the method claim 2, and is
rejected under similar rationale.
Regarding claim 10, the device claim 10 is similar in scope to the method claim 3, and is
rejected under similar rationale.
Regarding claim 11, the device claim 11 is similar in scope to the method claim 4, and is
rejected under similar rationale.
Regarding claim 12, the device claim 12 is similar in scope to the method claim 5, and is
rejected under similar rationale.
Regarding claim 13, the device claim 13 is similar in scope to the method claim 6, and is
rejected under similar rationale.
Regarding claim 14, the device claim 14 is similar in scope to the method claim 7, and is
rejected under similar rationale.
Regarding claim 15, Jabbour teaches a non-transitory computer-readable storage cause the one or more processors to perform operations comprising (Page 5 Par 4 “Given a chest X-ray image X ∈ Rd×d, we consider the task of predicting some target diagnosis yt ∈ {0,1}. We assume a convolutional neural network (CNN) f(e(X;θ);w) is the composition of two functions:”, where a convolutional neural network is stored in computer-readable storage media): obtaining a to-be-verified factor existing in a plurality of evaluation images that are used to perform bias evaluation on a to-be-evaluated model; (Page 6 Par 2-3 “In our analysis, we assume the spurious correlation is modeled by a bias attribute b ∈ {0,1}”, “We consider a scenario in which in the training and validation data associated with the target task yt and attribute b may be highly correlated”, Page 7 Par 3 “AHRF dataset. We considered a cohort of 1,296 patients admitted to a large acute care hospital center during 2017-2018 who developed AHRF. We analyzed chest X-rays acquired closest to the time of AHRF onset”, Page 8 Par 2 “For each patient, we extracted several attributes that could serve as b, including body mass index (BMI), age, sex, race, and the presence of a pacemaker.”) classifying the plurality of evaluation
images based on the to-be-verified factor to obtain a first target evaluation image set,
wherein the first target evaluation image set comprises at least one of a first evaluation
image set comprising the to-be-verified factor or a second evaluation image set not
comprising the to-be-verified factor (Page 8, Par 2 “For each patient, we extracted several
attributes that could serve as b, including body mass index (BMI) , age, sex, race, and the
presence of a pacemaker. Demographic variables were extracted from the electronic health
record, while images were manually labeled for the presence of a pacemaker. We dichotomized
each variable: i) BMI ≥ 30, ii) age ≥ 63 years (the median age of our dataset), iii) sex = female, iv)
race = Black, and v) pacemaker present.”, where the plurality of evaluation images corresponds
to the input dataset, the first target evaluation set corresponds to the labeled images, and the
first and second evaluation set are the set dichotomized by the pacemaker variable); inputting
the first target evaluation image set and the second target evaluation image set into the to-
be-evaluated model for inference to obtain a target inference result (Page 10 Sect 4.3 “We initialized models on either ImageNet or by pretraining on the MIMIC-CXR and CheXpert datasets”, Page 14, Par 4 “To measure the extent to which the injected bias affects performance, we compare the learned model’s performance, Skewed Training Data, to the performance of a model trained on data that more closely mimics the test set, Unskewed Training Data. To generate the unskewed training dataset, we subsampled the data so that b was uncorrelated with yt. Again, to facilitate comparisons, we kept the size of the training data similar”); and outputting a bias evaluation result of the to-be-evaluated model based on the target inference result, wherein the bias evaluation result represents whether has a bias against an evaluation image comprising the to-be-verified factor. (Page 3 Sect 2.1 “We assume access to a labeled training set, and a separate held-out test set. In this work, we focus on a type of bias that arises when training data are skewed with respect to a particular attribute (i.e., a characteristic that divides patients into subgroups), resulting in undesirable model performance when the model is applied to test data that do not exhibit the same kind of skew”, Page 10 Sect 4.3 AHRF dataset training setup “Validation and test images were center cropped to 512×512.”, Page 13 Par 3 “these results (for the most part) show that attributes such as demographics and treatment are easy to infer based on a chest X-ray.”, Page 20 Sect 6 “we showed that in the presence of spurious correlations, deep learning models can exploit potential shortcuts associated with biased attributes”)
Jabbour teaches injecting a bias feature into an image-set through an image
preprocessing step (Page 11, Par 5 “in a series of follow-up experiments repeat our analyses
with a synthetic bias injected via an image preprocessing step, answering the final question”),
but fails to explicitly teach performing style conversion on the first target evaluation image set
based on the to-be-verified factor to obtain a second target evaluation image set, wherein
style conversion performed on the first evaluation image set is implemented by removing the
to-be-verified factor, and style conversion performed on the second evaluation image set is
implemented by adding the to-be-verified factor. In related field of endeavor, Zhu teaches
performing style conversion on the first target evaluation image set based on the to-be-verified
factor to obtain a second target evaluation image set, wherein style conversion performed on
the first evaluation image set is implemented by removing the to-be-verified factor, and style
conversion performed on the second evaluation image set is implemented by adding the to-be-
verified factor. (Sect 5.2 Applications, Par 3 “Object transfiguration (Figure 13) The model is
trained to translate one object class from ImageNet [5] to another (each class contains around
1000 training images)”, Figure 4 “The input images x, output images G(x) and the reconstructed
images F(G(x)) from various experiments. From top to bottom: photo ↔ Cezanne, horses ↔zebras, winter → summer Yosemite, aerial photos ↔ Google maps.”, where an image factor
can be added to a photo using G(x), and removed using F(G(x)))
It would have been obvious to one of ordinary skill in the art to have modified Jabbour
to include performing style conversion on the first target evaluation image set based on the to-
be-verified factor to obtain a second target evaluation image set, wherein style conversion
performed on the first evaluation image set is implemented by removing the to-be-verified
factor, and style conversion performed on the second evaluation image set is implemented by
adding the to-be-verified factor as taught by Zhu. Doing so would provide a method for the
image preprocessing step for injecting bias into an image set (Sect 2. Related Work, Neural Style
Transfer “Therefore, our method can be applied to other tasks, such as Input 𝑥 painting→
photo, object transfiguration, etc”).
Regarding claim 16, the non-transitory computer-readable medium claim 16 is similar in scope to the method claim 2, and is rejected under similar rationale.
Regarding claim 17, the non-transitory computer-readable medium claim 17 is similar in scope to the method claim 3, and is rejected under similar rationale.
Regarding claim 18, the non-transitory computer-readable medium claim 18 is similar in scope to the method claim 4, and is rejected under similar rationale.
Regarding claim 19, the non-transitory computer-readable medium claim 19 is similar in scope to the method claim 5, and is rejected under similar rationale.
Regarding claim 20, the non-transitory computer-readable medium claim 20 is similar in scope to the method claim 6, and is rejected under similar rationale.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JOHN PATRICK GOCO whose telephone number is (571)272-5872. The examiner can normally be reached M-Th, 7:00 am - 5:00 pm.
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, Kee Tung can be reached at (571)272-7794. 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.
/J.P.G./ Examiner, Art Unit 2611
/KEE M TUNG/ Supervisory Patent Examiner, Art Unit 2611
Prosecution Insights