DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This action is in response to the amendments and arguments filed on 01/26/2026. In the current amendments, claims 1-7, 13, 19, 25 and 31-36 were amended, and no claims were added or canceled. As such, claims 1-36 are pending and have been examined.
Claim Rejections - 35 USC § 102
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 the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claims 1-2, 7-8, 13-15, 17, 19-21, 23, 25-27, 29, 31-33 and 35 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by non-patent literature Yamashita et al. (“Improving Quality of Training Samples Through Exhaustless Generation and Effective Selection for Deep Convolutional Neural Networks”, 2015, hereinafter “Yamashita”).
Regarding claim 1, Yamashita discloses the invention as claimed including one or more processors comprising: circuitry to (Sect. 5.1 MNIST, Page 234 “We implemented the methods using Theano library and train the network on a NVIDIA GT640 2GB GPU” discloses a GPU corresponding to one or more processors comprising circuitry to perform operations):
adjust an input to one or more neural networks based, at least in part, on backpropagation of output of the one or more neural networks, wherein the adjustment causes the adjusted input to represent features learned by the one or more neural networks; and adjust one or more weights of the one or more neural networks based, at least in part, on use of the adjusted input as training samples (see, e.g., Sect. 3.1 Training of Convolutional Neural Networks & Page 230 “ConvNets are trained by backpropagation just as general neural networks. Backpropagation uses the function shown in Eq.(1) to estimate the connected weights with minimized E by gradient descent in Eq.(2)… Note that {p|1,...,P} is the training sample, op is the corresponding value of training sample p in output layer and tp is the label data of p” and 4.8 Network Training & Page 233 “We use backpropagation to update the parameters Wt+1 in iteration t +1 as defined in Eq.(10)” discloses adjusting an input to neural networks based on backpropagation to update the parameters of neural network, where the adjusted weights of neural network are based on backpropagation of output, and adjusting the weights of neural networks based on the training samples).
Regarding claim 2,
As discussed above, Yamashita discloses the one or more processors of claim 1.
Yamashita further discloses wherein the input to the one or more neural networks includes a set of noise images modified during a number of passes through the one or more neural networks (see, e.g., Sect. 4.1 Exhaustless Sample Generation & Page 231 “After applying elastic distortion on both the sample images and the binary labeled data, the synthetic images are then synthesized with background images while applying scaling, rotation and translation…Sample generation results in a package that includes a set amount of synthetic images” teaches one or more neural networks includes a synthetic images corresponds to noise image).
Independent claim 7:
With respect to independent claim 7, claim 7 is substantially similar to claim 1 and therefore is rejected on the same grounds as claim 1, discussed above. In particular, claim 7 is a system claim with operations that correspond to the operations performed by the “One or more processors” of claim 1.
Yamashita further discloses “A system comprising: one or more processors” (see, e.g., FIG. 1 – depicting a “training system” and page 231, Sect. 4: “We illustrated the entire training system in Fig.1. The system consists of two processes, data augmentation and ConvNets training.” and Sect. 5.1, p. 234: “We implemented the methods using Theano library and train the network on a NVIDIA GT640 2GB GPU.” Discloses a system comprising one or more processors).
Claim 8:
Regarding claim 8, as discussed above, Yamashita discloses the system of claim 7.
Yamashita further discloses wherein the input to the one or more neural networks includes a set of noise images modified during a number of passes through the one or more neural networks (see, e.g., Sect. 4.1 Exhaustless Sample Generation & Page 231 “After applying elastic distortion on both the sample images and the binary labeled data, the synthetic images are then synthesized with background images while applying scaling, rotation and translation…Sample generation results in a package that includes a set amount of synthetic images” teaches one or more neural networks includes a synthetic images corresponds to noise image).
Claim 8 recites a system using a processor to perform the operations recited in claim 2. As Yamashita discloses performing the operations of claim 2 using one or more processors (see, above rejection of claim 2), Yamashita, 5.1 MNIST & Page 234) in a system (see, e.g., FIG. 1 – depicting a “training system” and p. 231, Sect. 4: “We illustrated the entire training system in Fig.1. The system consists of two processes, data augmentation and ConvNets training.” and Sect. 5.1, p. 234: “We implemented the methods using Theano library and train the network on a NVIDIA GT640 2GB GPU.”), claim 8 is rejected for reasons set forth above in the rejections of claim 2.
Independent claim 13:
With respect to independent claim 13, claim 13 is substantially similar to claim 1 and therefore is rejected on the same grounds as claim 1, discussed above. In particular, claim 13 is a method claim with operations that correspond to the operations performed by the “One or more processors” of claim 1.
Yamashita further discloses “A method” (see, e.g., Sect. 4 PROPOSED METHOD, p. 231 “we propose here a method for an asynchronous exhaustless sample generation coupled with an effective sample selection during parameter updating” Discloses a method).
Claim 14:
Regarding claim 14, as discussed above, Yamashita discloses the method of claim 13.
Yamashita further discloses wherein the input to the one or more neural networks includes a set of noise images modified during a number of passes through the one or more neural networks (see, e.g., Sect. 4.1 Exhaustless Sample Generation & Page 231 “After applying elastic distortion on both the sample images and the binary labeled data, the synthetic images are then synthesized with background images while applying scaling, rotation and translation…Sample generation results in a package that includes a set amount of synthetic images” teaches one or more neural networks includes a synthetic images corresponds to noise image).
Claim 14 recites a method to perform the operations recited in claim 2. As Yamashita discloses performing the operations of claim 2 using one or more processors (see, above rejections of claim 2), Yamashita, 5.1 MNIST & Page 234) in a method (see, e.g., 4 PROPOSED METHOD, p. 231 “we propose here a method for an asynchronous exhaustless sample generation coupled with an effective sample selection during parameter updating” Discloses a method), claim 14 is rejected for reasons set forth above in the rejection of claim 2.
Independent claim 19:
With respect to independent claim 19, claim 19 is substantially similar to claim 1 and therefore is rejected on the same grounds as claim 1, discussed above. In particular, claim 19 is a non-transitory machine-readable medium claim with operations that correspond to the operations performed by the “One or more processors” of claim 1.
Yamashita further discloses “a non-transitory machine-readable medium having stored thereon a set of instructions” (see, e.g., Sect. 4 PROPOSED METHOD, p. 231 “MNIST contains 50000 images of each class for training and 10000 images for testing”, Sects. 5.1-5.2, p. 234, “We implemented the methods using Theano library and train the network on a NVIDIA GT640 2GB GPU” and “We stored a set amount of augmented samples in the package, here 1000 images.” discloses a NVIDIA GT640 2GB GPU is part of a computer system that can store and process data and the MNIST dataset is a structured set of digit images and labels stored in a digital format in a machine-readable medium, which the computer/machine can read and use for training).
Claim 20:
Regarding claim 20, as discussed above, Yamashita discloses the non-transitory machine-readable medium of claim 19.
Yamashita further discloses wherein the input to the one or more neural networks includes a set of noise images modified during a number of passes through the one or more neural networks (see, e.g., Sect. 4.1 Exhaustless Sample Generation & Page 231 “After applying elastic distortion on both the sample images and the binary labeled data, the synthetic images are then synthesized with background images while applying scaling, rotation and translation…Sample generation results in a package that includes a set amount of synthetic images” teaches one or more neural networks includes a synthetic images corresponds to noise image).
Claim 20 recites a non-transitory machine-readable medium to perform the operations recited in claim 2. As Yamashita discloses performing the operations of claim 2 using one or more processors (see, above rejections of claim 2 and Yamashita, 5.1 MNIST & Page 234) and a machine-readable medium (see, e.g., Sect. 4 PROPOSED METHOD, p. 231 “MNIST contains 50000 images of each class for training and 10000 images for testing”, Sects. 5.1-5.2, p. 234, “We implemented the methods using Theano library and train the network on a NVIDIA GT640 2GB GPU” and “We stored a set amount of augmented samples in the package, here 1000 images.” discloses a NVIDIA GT640 2GB GPU is part of a computer system that can store and process data and the MNIST dataset is a structured set of digit images and labels stored in a digital format in a machine-readable medium, which the computer/machine can read and use for training), claim 20 is rejected for reasons set forth in the rejection of claim 2.
Independent claim 25:
With respect to independent claim 25, claim 25 is substantially similar to claim 1 and therefore is rejected on the same grounds as claim 1, discussed above. In particular, claim 25 is a training system claim with operations that correspond to the operations performed by the “One or more processors” of claim 1.
Yamashita further discloses “a training system” (see, e.g., FIG. 1 – depicting a “training system” and 4 PROPOSEDMETHOD & p. 231: “We illustrated the entire training system in Fig.1. The system consists of two processes, data augmentation and ConvNets training.” And 5.1 MNIST, p. 234: “We implemented the methods using Theano library and train the network on a NVIDIA GT640 2GB GPU.” discloses a training system).
Claim 26:
Regarding claim 26, as discussed above, Yamashita discloses training system of claim 25.
Yamashita further discloses wherein the input to the one or more neural networks includes a set of noise images modified during a number of passes through the one or more neural networks (see, e.g., Sect. 4.1 Exhaustless Sample Generation & Page 231 “After applying elastic distortion on both the sample images and the binary labeled data, the synthetic images are then synthesized with background images while applying scaling, rotation and translation…Sample generation results in a package that includes a set amount of synthetic images” teaches one or more neural networks includes a synthetic images corresponds to noise image).
Claim 26 recites a training system to perform the operations recited in claim 2. As Yamashita discloses performing the operations of claim 2 using one or more processors (see, above rejections of claim 2 and Yamashita, 5.1 MNIST & Page 234) in a training system (see, e.g., FIG. 1 – depicting a “training system” and 4 PROPOSEDMETHOD & p. 231: “We illustrated the entire training system in Fig.1. The system consists of two processes, data augmentation and ConvNets training.” and Sect. 5.1 MNIST, p. 234: “We implemented the methods using Theano library and train the network on a NVIDIA GT640 2GB GPU.” discloses a training system), claim 26 is rejected for reasons set forth above in the rejection of claim 2.
Independent claim 31:
With respect to independent claim 31, claim 31 is substantially similar to claim 1 and therefore is rejected on the same grounds as claim 1, discussed above. In particular, claim 31 is a one or more processors claim with operations that correspond to the operations performed by the “One or more processors” of claim 1.
Yamashita further discloses “One or more processors, comprising: one or more arithmetic logic units (ALUs) to cause one or more neural networks to classify image data, wherein the one or more neural networks were trained by at least” (Sect. 2, p. 229 “to reduce learning time, vast parallel computing
with GPU is often used. … GPU parallelization is used when updating parameters. … GPU helps updating the parameters faster … implemented simple data augmentation in Python on CPU while parameter updating is computed on GPU” and Sect. 5.1, p. 234, “We implemented the methods using Theano library and train the network on a NVIDIA GT640 2GB GPU.” discloses arithmetic logic operations performed by ALUs of a GPU).
Claim 32:
Regarding claim 32, as discussed above, Yamashita discloses one or more processors of claim 31.
Yamashita further discloses wherein the input to the one or more neural networks includes a set of noise images modified during a number of passes through the one or more neural networks (see, e.g., Sect. 4.1 Exhaustless Sample Generation & Page 231 “After applying elastic distortion on both the sample images and the binary labeled data, the synthetic images are then synthesized with background images while applying scaling, rotation and translation…Sample generation results in a package that includes a set amount of synthetic images” teaches one or more neural networks includes a synthetic images corresponds to noise image).
Claim 32 recites a training system to perform the operations recited in claim 2. As Yamashita discloses performing the operations of claim 2 using one or more processors (see, above rejections of claim 2 and Yamashita, Sect. 5.1 MNIST & Page 234) including one or more ALUs (see, e.g., Yamashita, Sect. 2, p. 229 “to reduce learning time, vast parallel computing with GPU is often used. … GPU parallelization is used when updating parameters. … GPU helps updating the parameters faster … implemented simple data augmentation in Python on CPU while parameter updating is computed on GPU”, Sect. 4.3 Architecture of Networks, p. 232: “the recognition task utilizes the convolutional, max pooling, maxout, fully connected and classification layers, but not the binarization layer” and Sect. 5.1 MNIST, p. 234: “We implemented the methods using Theano library and train the network on a NVIDIA GT640 2GB GPU” discloses arithmetic logic operations performed by ALUs of a GPU)), claim 32 is rejected for reasons set forth in the rejection of claim 2.
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 text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
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.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 3, 5, 9, 11, 15, 17, 21, 23, 27, 29, 33, 35 are rejected under 35 U.S.C. 103 as being unpatentable over Yamashita (“Improving Quality of Training Samples Through Exhaustless Generation and Effective Selection for Deep Convolutional Neural Networks”) in view of Goodfellow (“Explaining and Harnessing Adversarial Examples”).
Regarding claim 3, As discussed above, Yamashita discloses the one or more processors of claim 2.
Yamashita does not explicitly disclose wherein an activation for a selected classification is maximized for forward passes through the one or more neural networks to produce an inference and a cross-entropy loss.
However, in the same field, analogous art Goodfellow teaches wherein an activation for a selected classification is maximized for forward passes through the one or more neural networks to produce an inference and a cross-entropy loss (4 LINEAR PERTURBATION OF NON-LINEAR MODELS & Page 3 “Let θ be the parameters of a model, x the input to the model, y the targets associated with x (for machine learning tasks that have targets) and J(θ, x, y) be the cost used to train the neural network. We can linearize the cost function around the current value of θ, obtaining an optimal max-norm constrained pertubation [sic – perturbation] of η = ε sign (∇xJ(θ, x, y)). We refer to this as the “fast gradient sign method” of generating adversarial examples. Note that the required gradient can be computed efficiently using backpropagation” teaches J(θ, x, y) training loss corresponding to cross entropy; inference is implied as the process of computing output from the input. Also, input modified in the direction of the gradient of the training cost which boosts the network’s activation for that target class (corresponds to activation maximized).
Yamashita and Goodfellow are analogous art because they are both directed to training a neural network.
It would have been obvious for one of ordinary skill in the arts before the effective filing date of the claimed invention to incorporate the limitations above as taught by Goodfellow to the disclosed invention of Yamashita.
One of ordinary skill in the arts would have been motivated to make this modification because of the following, “We found that training with an adversarial objective function based on the fast gradient sign method was an effective regularizer”, as suggested by Goodfellow (Goodfellow, 6 ADVERSARIAL TRAINING OF DEEP NETWORKS & Page 5).
Claim 5:
As discussed above, Yamashita discloses the one or more processors of claim 2.
Yamashita does not explicitly disclose wherein the set of noise images is modified according to noise gradients determined for the number of passes through the one or more neural networks.
However, in the same field, analogous art Goodfellow teaches wherein the set of noise images is modified according to noise gradients determined for the number of passes through the one or more neural networks (6 ADVERSARIAL TRAINING OF DEEP NETWORKS & Page 5 “Let θ be the parameters of a model, x the input to the model, y the targets associated with x (for machine learning tasks that have targets) and J(θ, x, y) be the cost used to train the neural network. We can linearize the cost function around the current value of θ, obtaining an optimal max-norm constrained pertubation [sic – perturbation] of η = ε sign (∇xJ(θ, x, y)). We refer to this as the “fast gradient sign method” of generating adversarial examples. Note that the required gradient can be computed efficiently using backpropagation” teaches fast gradient sign method computes cost function respect to input and the (∇xJ(θ, x, y) gradient of the cost function is used to modify the input).
Yamashita and Goodfellow are analogous art because they are both directed to training a neural network.
It would have been obvious for one of ordinary skill in the arts before the effective filing date of the claimed invention to incorporate the limitations above as taught by Goodfellow to the disclosed invention of Yamashita.
One of ordinary skill in the arts would have been motivated to make this modification because of the following, “We found that training with an adversarial objective function based on the fast gradient sign method was an effective regularizer”, as suggested by Goodfellow (Goodfellow, 6 ADVERSARIAL TRAINING OF DEEP NETWORKS & Page 5).
Claims 9 and 11:
Claims 9 and 11 recite systems performing operations corresponding to the operations performed by using the processor of claims 3 and 5.
As Yamashita performs operations/steps using a processor (Yamashita, 5.1 MNIST & Page 234) in a system (see, e.g., Yamashita, Sect. 4, p. 231, “We illustrated the entire training system in Fig.1.The system consists of two processes, data augmentation and ConvNets training”), claims 9 and 11 are rejected for reasons set forth above in the rejections of claims 3 and 5, respectively.
Claims 15 and 17:
Claims 15 and 17 recite methods with steps corresponding to the operations performed by using the processor of claims 3 and 5. As Yamashita performs operations/steps using a processor (Yamashita, 5.1 MNIST & Page 234) in a system (see, e.g., Yamashita, Sect. 4, p. 231, “We illustrated the entire training system in Fig.1.The system consists of two processes, data augmentation and ConvNets training”), claims 15 and 17 are rejected for reasons set forth above in the rejections of claims 3 and 5, respectively.
Claims 21 and 23:
Claims 21 and 23 recite non-transitory machine-readable media having stored thereon a set of instructions, which if performed by using a processor, carry out operations corresponding to the operations performed by using the processor of claims of claims 3 and 5. As Yamashita performs operations/steps using a processor (Yamashita, 5.1 MNIST & Page 234) in a system (see, e.g., Yamashita, Sect. 4, p. 231, “We illustrated the entire training system in Fig.1.The system consists of two processes, data augmentation and ConvNets training”), claims 21 and 23 are rejected for reasons set forth above in the rejections of claims 3 and 5, respectively.
Claims 27 and 29:
Claims 27 and 29 recite a training system performing operations corresponding to the operations performed by using the one or more processors of claims 3 and 5.
As Yamashita performs operations/steps using a processor (Yamashita, 5.1 MNIST & Page 234) in a system (see, e.g., Yamashita, Sect. 4, p. 231, “We illustrated the entire training system in Fig.1.The system consists of two processes, data augmentation and ConvNets training”) t, claims 27 and 29 are rejected for reasons set forth above in the rejections of claims 3 and 5, respectively.
Claims 33 and 35:
Claims 33 and 35 recite one or more processors, comprising: one or more arithmetic logic units (ALUs) to cause one or more neural networks to classify image data, wherein the one or more neural networks were trained by at least (Yamashita, 5.1 MNIST teaches model training using the dataset which using a GPU comprising an ALU) to perform operations corresponding to the operations performed by using the processor of claims 3 and 5.
As Yamashita discloses performing the operations of claims 3 and 5 using a processor (see, e.g., Yamashita, 5.1 MNIST & Page 234) in a system (see, e.g., Yamashita, Sect. 4, p. 231, “We illustrated the entire training system in Fig.1.The system consists of two processes, data augmentation and ConvNets training”), claims 33 and 35 are rejected for reasons set forth above in the rejections of claims 3 and 5, respectively.
Claims 4, 6, 10, 12, 16, 18, 22, 24, 28, 30, 34 and 36 are rejected under 35 U.S.C. 103 as being unpatentable over Yamashita (“Improving Quality of Training Samples Through Exhaustless Generation and Effective Selection for Deep Convolutional Neural Networks”) in view of Goodfellow (“Explaining and Harnessing Adversarial Examples”) and further in view of Zhai (“Generative adversarial networks as variational training of energy based models”).
Claim 4.
As discussed above, Yamashita in view of Goodfellow teaches the one or more processors of claim 3,
Yamashita in view of Goodfellow does not explicitly teach wherein backpropagation through the one or more neural networks utilizes feature distribution regularization to maintain mean and variance values.
However, in the same field, analogous art Zhai teaches wherein backpropagation through the one or more neural networks utilizes feature distribution regularization to maintain mean and variance values (Algorithm 2 & Page 6 teaches training both generator and energy model using stochastic gradient descent 5 BOUNDED MULTI-MODAL ENERGY & Page 5, 1st paragraph “the generated samples to be diverse, as H ˜(pg) reaches its minimum if G(z) collapses to one single point. Moreover, in the outer loop while minimizing the NLL w.r.t. E, we find it helpful to also maximize
PNG
media_image1.png
24
253
media_image1.png
Greyscale
as well, which acts as a regularizer to E to encourage the average activation of each expert σj (·) to close to 0.5” teaches maintaining average close to .5 is mean and to prevent G(z) collapses is to maintaining variance).
Yamashita, Goodfellow and Zhai are analogous art because they are each directed to training a neural network.
It would have been obvious for one of ordinary skill in the arts before the effective filing date of the claimed invention to incorporate the limitations above as taught by Zhai to the disclosed invention of Yamashita in view of Goodfellow.
One of ordinary skill in the arts would have been motivated to make this modification because of the following, “we propose VGAN, which bridges GANs and EBMs and combines the benefits from both worlds” and “we show that the mini-max game of GANs is approximately equivalent to minimizing a variational lower bound of the negative log likelihood (NLL) of an EBM” (Zhai, 1 INTRODUCTION & Page 1-2).
Claim 6:
As discussed above, Yamashita in view of Goodfellow teaches the one or more processors of claim 5,
Yamashita in view of Goodfellow doesn’t explicitly teach wherein the set of noise images is further modified using at least one image prior.
However, Zhai teaches wherein the set of noise images is further modified using at least one image prior (2 GENERATIVE ADVERSARIAL NETWORKS & Page 2, Paragraph 2 “As the two-player, mini-max game reaches the Nash equilibrium, G defines an implicit distribution pg(x) that recovers the data distribution, i.e., pg(x) = pdata(x)” teaches Pdata(x) is prior image and Equation 1 teaches generated based on Pdata(x)).
Yamashita, Goodfellow and Zhai are analogous art because they are both directed to training a neural Network.
It would have been obvious for one of ordinary skill in the arts before the effective filing date of the claimed invention to incorporate the limitation(s) above as taught by Zhai to the disclosed invention of Yamashita in view of Goodfellow.
One of ordinary skill in the arts would have been motivated to make this modification because of the following, “we propose VGAN, which bridges GANs and EBMs and combines the benefits from both worlds” and “we show that the mini-max game of GANs is approximately equivalent to minimizing a variational lower bound of the negative log likelihood (NLL) of an EBM” (Zhai, 1 INTRODUCTION & Page 1-2).
Claims 10 and 12:
Claims 10 and 12 recite a system to perform operations corresponding to the operations performed by using the processor claims 4 and 6.
As Yamashita in view of Goodfellow performs the operations of claims 4 and 6 using one or more processors (see, e.g., Yamashita, Sect. 5.1 MNIST & Page 234) in a system (see, e.g., Yamashita, Sect. 4, p. 231, “We illustrated the entire training system in Fig.1.The system consists of two processes, data augmentation and ConvNets training”), claims 10 and 12 are rejected for reasons set forth above in the rejections of claims 4 and 6, respectively.
Claims 16 and 18:
Claims 16 and 18 recite methods with steps corresponding to the operations performed by the processor of claims 4 and 6.
As Yamashita in view of Goodfellow performs the operations of claims 4 and 6 using one or more processors (see, e.g., Yamashita, Sect. 5.1 MNIST & Page 234) and a method (see, e.g., Yamashita, Sect. 4 PROPOSED METHOD, p. 231, “we propose here a method for an asynchronous exhaustless sample generation”), claims 16 and 18 are rejected for reasons set forth in the rejections of claims 4 and 6, respectively.
Claims 22 and 24:
Claims 22 and 24 recite machine-readable media having stored thereon a set of instructions to perform operations by corresponding to the operations performed by using the processor of claims 4 and 6.
As Yamashita in view of Goodfellow performs the operations of claims 4 and 6 using one or more processors (see, e.g., Yamashita, Sect. 5.1 MNIST & Page 234) and a machine-readable medium (see, e.g., Sect. 4, p. 231 “MNIST contains 50000 images of each class for training and 10000 images for testing”, Sects. 5.1-5.2, p. 234, “We implemented the methods using Theano library and train the network on a NVIDIA GT640 2GB GPU” and “We stored a set amount of augmented samples in the package, here 1000 images.” discloses a NVIDIA GT640 2GB GPU is part of a computer system that can store and process data and the MNIST dataset is a structured set of digit images and labels stored in a digital format in a machine-readable medium, which the computer/machine can read and use for training), claims 22 and 24 are rejected for reasons set forth above in the rejections of claims 4 and 6, respectively.
Claims 28 and 30:
Claims 28 and 30 recite a training system performing operations corresponding to the operations performed by using the processor of Claims 4 and 6.
As Yamashita in view of Goodfellow performs the operations of claims 4 and 6 using one or more processors (see, e.g., Yamashita, Sect. 5.1 MNIST & Page 234) in a training system (see, e.g., Yamashita, Sect. 4, p. 231, “We illustrated the entire training system in Fig.1.The system consists of two processes, data augmentation and ConvNets training”), claims 28 and 30 are rejected for reasons set forth above in the rejections of claims 4 and 6, respectively.
Claims 34 and 36:
Claims 34 and 36 recite a processor comprising one or more arithmetic logic units (ALUs) performing the operations of claims 4 and 6. As Yamashita in view of Goodfellow performs the operations of claims 4 and 6 using one or more processors (see, e.g., Yamashita, Sect. 5.1 MNIST & Page 234) including one or more ALUs (see, e.g., Yamashita, Sect. 2, p. 229 “to reduce learning time, vast parallel computing with GPU is often used. … GPU parallelization is used when updating parameters. … GPU helps updating the parameters faster … implemented simple data augmentation in Python on CPU while parameter updating is computed on GPU” and Sect. 5.1, p. 234, “We implemented the methods using Theano library and train the network on a NVIDIA GT640 2GB GPU.” discloses arithmetic logic operations performed by ALUs of a GPU), claims 34 and 36 are rejected for reasons set forth above in the rejections of claims 4 and 6, respectively.
Response to Arguments
Applicant's arguments filed on 01/26/2026 with respect to 35 U.S.C. § 102 rejections of claims 1-3, 5, 7-9, 11, 13-15, 17, 19-21, 23, 25-27, 29, 31-33, and 35 have been fully considered but they are moot in view of newly-cited prior art Yamashita (“Improving Quality of Training Samples Through Exhaustless Generation and Effective Selection for Deep Convolutional Neural Networks”).
Regarding amended claim 1, applicant asserts “Goodfellow discloses using backpropagation to generate adversarial examples, which are described as inputs formed by "applying small but intentionally worse-case perturbations to examples from the dataset, such that the perturbed input results in the model outputting an incorrect answer with high confidence." Goodfellow at 1. For example, Goodfellow demonstrates that "by adding an imperceptibly small vector whose elements are equal to the sign of the elements of the gradient of the cost function with respect to the input, we can change GoogLeNet's classification of the image." Id. at 3. That is, Goodfellow discloses adjustments to input data made to intentionally misclassify the adjusted input. In contrast, the claim recites adjustments to an input made to represent features learned by the neural network and correspond to the original training data. Therefore, because Goodfellow does not disclose, at least, "wherein the adjustment causes the adjusted input to represent features learned by the one or more neural networks," it does not anticipate claim 1” (Remarks Pg. 9).
Examiner’s Response:
These arguments have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in this argument. A newly-cited prior art, non-patent literature Yamashita, (“Improving Quality of Training Samples Through Exhaustless Generation and Effective Selection for Deep Convolutional Neural Networks”, 2015) has been applied to teach the limitations referred to in this argument.
Applicant's arguments filed on 01/26/2026 with respect to 35 U.S.C. § 103 rejections of claims 4, 6, 10, 12, 16, 18, 22, 24, 28, 30, 34, and 36 have been fully considered but they are moot in view of newly-cited prior art Yamashita (“Improving Quality of Training Samples Through Exhaustless Generation and Effective Selection for Deep Convolutional Neural Networks”).
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).
The prior art made of record, listed on form PTO-892, and not relied upon, is considered pertinent to applicant's disclosure.
For example, Liu (U.S. Patent Application Pub. No. 2021/0097691 A1, hereinafter “Liu” discloses that “input to a first layer of a generator can be a random noise sampled from unit Gaussian, or segmentation map downsampled to an 8×8 resolution” where “a semantic layout can then be generated 512 using labeled regions of an image space. In at least one embodiment, a semantic layout can be provided 514 as input to an image synthesis network.”, “Cross-validation and adding Gaussian noise to a training dataset are techniques that can be useful for avoiding overfitting to any one dataset.” and “backpropagation can be utilized to calculate a gradient used for determining weights for a neural network.” (see, paragraphs 67, 74, 110 and 113).
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 RANDY K BALDWIN whose telephone number is (571)270-5222. The examiner can normally be reached on Mon - Fri 9:00-6:00.
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, Kamran Afshar can be reached at (571) 272-7796. 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.
/RANDALL K. BALDWIN/Primary Examiner, Art Unit 2125
Prosecution Insights