DETAILED ACTION
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claims 1-20 have been presented for examination based on the application filed on 10/25/2022.
Claim 16 is rejected under 35 U.S.C. 102(a)(1) as being anticipated by NPL by C. Yeung et al (“Global Inverse Design across Multiple Photonic Structure Classes Using Generative Deep Learning”,. Adv. Optical Mater. 2021, 9, 2100548. https://doi.org/10.1002/adom.202100548).
Claim(s) 1-4, 8-15, 17-20 are rejected under 35 U.S.C. 103 as being unpatentable over NPL by C. Yeung et al (“Global Inverse Design across Multiple Photonic Structure Classes Using Generative Deep Learning”,. Adv. Optical Mater. 2021, 9, 2100548. https://doi.org/10.1002/adom.202100548), in view of NPL by M. Lee et al (“Controllable Generative Adversarial Network," in IEEE Access, vol. 7, pp. 28158-28169, 2019, doi: 10.1109/ACCESS.2019.2899108) further in view of Jang; Wonik et al. (US PGPUB No. US 20210158173 A1).
Claim(s) 5-7 are rejected under 35 U.S.C. 103 as being unpatentable over Yeung, in view of Lee, in view of Jang, further in view of NPL by M. Lee et al (Lee-2 hereafter). "Regularization methods for generative adversarial networks: An overview of recent studies." arXiv preprint arXiv:2005.09165 (2020).
This action is made Non-Final.
Priority
Acknowledgment is made of applicant's claim for foreign priority under 35 U.S.C. 119(a)-(d). The certified copy has been electronically retrieved on 12/8/2022.
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 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.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claim 16 is rejected under 35 U.S.C. 102(a)(1) as being anticipated by NPL by C. Yeung et al (“Global Inverse Design across Multiple Photonic Structure Classes Using Generative Deep Learning”,. Adv. Optical Mater. 2021, 9, 2100548. https://doi.org/10.1002/adom.202100548).
Regarding Claim 16
Yeung teaches An inverse design method for designing a dielectric pattern of an image sensor using a controllable Generative Adversarial Network (cGAN) (Yeung : Abstract "... To overcome these challenges, a global deep learning-based inverse design framework is presented, where a conditional deep convolutional generative adversarial network is trained on colored images encoded with a range of material and structural parameters, including refractive index, plasma frequency, and geometric design...."; Fig.2a-b; See e.g. in Fig.1 various designs being generated), the method comprising: collecting simulation data related to transmittance for each wavelength among a plurality of wavelengths relevant to the dielectric pattern (Yeung: Pg.2 Col.1-2 "...The encoded images are used to train a customized conditional deep convolutional generative adversarial network (cDCGAN), which we evaluate by inputting a variety of target absorption spectra...."
PNG
media_image1.png
674
770
media_image1.png
Greyscale
); training a generation model based on the cGAN driven in a computing system using the simulation data to provide a trained generation model (Yeung: Fig.2 above showing the training and design process, Pg.3 Col.2-Pg.4 Col.2 detailing the training data set - "... cDCGANs have previously been used to generate domain-specific images in response to input conditions.[41–43] Implemented in the PyTorch framework, the cDCGAN consists of a generator and a discriminator. Initially, batches of absorption spectra (y) are fed into the generator, along with a latent vector (z), to generate “fake” images (G) that are similar to the “real” images (x) from the training set...."; Fig.2a ); generating a dielectric pattern image by inputting a maximum transmittance wavelength to the trained generation model (Yeung: See Fig,2a showing the input wavelength spectra to generate the pattern as shown in Fig.3 "... Figure 3 presents a series of tests performed with inputs that originate from the validation dataset (10% of the training dataset). Here, the blue lines represent randomly selected inputs (across both classes of structures), and the orange lines are the simulated spectra of the cDCGAN-generated designs....") ; and designing the image sensor including the dielectric pattern in accordance with the dielectric pattern image (Yeung: Pg.4 Col.2 "... After training the cDCGAN, we developed an image processing workflow to convert the generated images into full 3D metasurface designs (Figure 2b). ...") .
---- This page is left blank after this line ----
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 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.
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.
Claim(s) 1-4, 8-15, 17-20 are rejected under 35 U.S.C. 103 as being unpatentable over NPL by C. Yeung et al (“Global Inverse Design across Multiple Photonic Structure Classes Using Generative Deep Learning”,. Adv. Optical Mater. 2021, 9, 2100548. https://doi.org/10.1002/adom.202100548), in view of NPL by M. Lee et al (“Controllable Generative Adversarial Network," in IEEE Access, vol. 7, pp. 28158-28169, 2019, doi: 10.1109/ACCESS.2019.2899108) further in view of Jang; Wonik et al. (US PGPUB No. US 20210158173 A1).
Regarding Claim 1
Yeung teaches An inverse design system configured to design a pattern of a structure (Yeung : Abstract "... To overcome these challenges, a global deep learning-based inverse design framework is presented, where a conditional deep convolutional generative adversarial network is trained on colored images encoded with a range of material and structural parameters, including refractive index, plasma frequency, and geometric design...."; Fig.2a-b) in an image sensor (Yeung: See Fig.2a) , the inverse design system comprising: (Yeung: Pg.4 Col.2- Pg.5 Col.1 "... We evaluated the performance of our trained cDCGAN and image processing method by inputting a set of absorption spectra (coupled with randomly sampled latent vectors) and analyzing the resulting designs....") , wherein the cGAN corresponding to a target characteristic (Yeung: Pg.4 Col.2- Pg.5 Col.1 "... We evaluated the performance of our trained cDCGAN and image processing method by inputting a set of absorption spectra (coupled with randomly sampled latent vectors) and analyzing the resulting designs...." – input spectra, latent vector fed into GAN as results compared as shown in Fig.3, See Pg.5 Col.2-Pg.6 Col.1 stating "...In addition, across a wide range of input spectra, we observe that the network synthesized designs that are noticeably different from the known structures (either in resonator shape or property /thickness). Despite this difference, the generated designs exhibit responses that strongly match the input targets....")- synthesized designs as image of structure; Pg.4 Col.2 "... After training the cDCGAN, we developed an image processing workflow to convert the generated images into full 3D metasurface designs (Figure 2b)...."); and an input/output (I/O) interface configured to receive a training data set used to train the cGAN (Yeung: Pg.3 Col.2-Pg.4 Col.2 detailing the training data set - "... cDCGANs have previously been used to generate domain-specific images in response to input conditions.[41–43] Implemented in the PyTorch framework, the cDCGAN consists of a generator and a discriminator. Initially,
batches of absorption spectra (y) are fed into the generator, along with a latent vector (z), to generate “fake” images (G) that are similar to the “real” images (x) from the training set...."; Fig.2a ) , communicate the training data set to the CPU, and output the image generated by the cGAN (Yeung: Pg.3 Col.2-Pg.4 Col.2) , wherein the cGAN includes a generator configured to generate a fake image of the structure (Yeung: Pg.3 Col.2-Pg.4 Col.2"... Initially,
batches of absorption spectra (y) are fed into the generator, along with a latent vector (z), to generate “fake” images (G) that are similar to the “real” images (x) from the training set..."; Pg.4 Col.2 "... After training the cDCGAN, we developed an image processing workflow to convert the generated images into full 3D metasurface designs (Figure 2b)...." ) , a discriminator configured to determine whether the fake image is fake or real (Yeung: Pg.4 Col.1 "... generator, along with a latent vector (z), to generate “fake” images (G) that are similar to the “real” images (x) from the training set.... Both G and x are then fed into the discriminator (D), which attempts to distinguish the generated images from the real…") , and a classifier configured to classify (Yeung Pg.4 Col.2 "...In this workflow,
the material property (ωP or n) and thickness values (t) are calculated
by taking the average pixel-values in their respective channels (based on structure classification), then reversing the normalization performed in the encoding step...").
PNG
media_image2.png
458
566
media_image2.png
Greyscale
Lee teaches a classifier configured to classify a class label associated with the fake image and corresponding to the target characteristic (Lee: Pg. 28161 , §III, Fig.2 – Notice the classifier (C) generates a label L which is associated with face image G,
).
Yeung and Lee do not specifically teach the system components of processor and memory, but would be obvious in view that designing using GAN cannot be performed manually for any practical application.
Jang teaches computer implemented GAN with a processor and memory (Jang: Fig.1 element 15 & 16 [0027]-[0028] implementing the GAN as generator/discriminator network) for purposes of real or fake image identification in semiconductor process monitoring application).
It would have been obvious to one (e.g. a designer) of ordinary skill in the art before the effective filing date of the claimed invention to apply the teachings of Lee (2019) to Yeung (2021). The motivation to combine would have been that data augmentation (DA) and Labels as computed by the classifier can lead to low classification loses and generator losses (Pg.28161 Col.2-28162 Col.1" ControlGAN employs DA for the training of the classifier, classification loss for the test set can signficantly be reduced, which corresponds to solve the overfitting problem existing in ACGAN. Therefore, the classifier in ControlGAN guides the generator better since the generator learns from the classification loss.... It is expected that label-focus samples can be generated with low EO values since the second loss in (8) takes the most of the generator loss...."). Further motivation to combine would be that Yeung and Lee are analogous arts to the claimed invention in the field of using generative adversarial networks (GAN) based detection of real and fake data and design (Yeung: Abstract and Lee: Abstract).
It would have been obvious to one (e.g. a designer) of ordinary skill in the art before the effective filing date of the claimed invention to apply the teachings of Jang to Yeung to show computer based implementation of the generative adversarial networks (GAN) showing the obvious aspect of computer based implementation for any practical application, such as in semiconductor layout design field ( Yeung: Abstract Lee: [0023][0029] ) . Further motivation to combine would have been that Yeung and Lee are analogous arts to the instant claim in the field of using generative adversarial networks (GAN) in the field of semiconductor layout design (Yeung: Abstract Lee: [0003] [0023][0029] ).
Regarding Claim 2
Yeung, Lee and Jang all teach The system of claim 1, wherein the generator is further configured to combine the class label with random noise (Yeung : Pg.6 Col.2-Pg.7 Col.1 "... For each spectrum (shown in their individual plots), a second query was performed after resampling the latent vector and slightly perturbing the starting spectrum. While not perturbing the spectrum still produced unique results on the second run (as shown in Figure S7, Supporting Information), adding small perturbations (<0.01 shifts in amplitude at various wavelengths) increased the overall uniqueness of the new designs...." - perturbations is understood as noise; Lee: Introduction "... A generator produces fake samples from random noises, while
a discriminator attempts to distinguish between these fake samples and real samples...."Jang: [0039]) to generate the fake image (Yeung: Fig.2 fake image as G) .
Regarding Claim 3
Yeung and Lee teaches The system of claim 1, wherein the discriminator if further configured to operate in response to a discriminator loss function (Yeung : Pg.4 Col.2) including a gradient penalty to stabilize weights during training of the cGAN (Lee: Pg.28162 "...The discriminator in Control-GAN is regularized by gradient penalty to assist the convergence. The gradient penalty parameter is set at 10, same as AC-WGAN-GP....").
Regarding Claim 4
Lee teaches the system of claim 1, wherein the classifier is further configured to generate a classification result for the fake image, and the classification result is feedback data applied to the generator during training of the generator (Lee: Fig.2 see Label (L) applied back to the generator) .
Regarding Claim 8
Yeung teaches the system of claim 1, wherein the structure is a dielectric pattern formed in a dielectric layer of the image sensor (Yeung: Fig.1 & Pgs, 2-4 showing the metal-insulator-metal (MIM) structures, where the insulator is shown as dielectric material which corresponds to target wavelength as in Fig.2 ) , and the target characteristic is a maximum transmittance wavelength Xmax of the dielectric layer (Yeung: See Fig.2 where the dielectric design is changed to meet the target wavelength (as in Fig.2a at least) and verification in Fig.2b & 3) .
Regarding Claim 9
Yeung teaches A method of training an inverse design system using a controllable Generative Adversarial Network (cGAN) (Yeung: Abstract) , (Yeung: See Fig.2 where the dielectric design is changed to meet the target wavelength (as in Fig.2a at least) and verification in Fig.2b & 3) , and generates a structure pattern (Yeung : Pg. 2 starting near end of Col.1 "... The encoded images are used to train a customized conditional deep convolutional generative adversarial network (cDCGAN), which we evaluate by inputting a variety of target absorption spectra. In response to the input spectra, the network generates corresponding metasurface designs that are validated through full-wave electromagnetic (EM) simulations...."; Also see Fig.1 with plurality of designs) for a nano-optical device (Yeung: Abstract "... Understanding how nano- or micro-scale structures and material properties can be optimally configured to attain specific functionalities remains a fundamental challenge. Photonic meta surfaces, for instance, can be spectrally tuned through material choice and structural geometry to achieve unique optical responses....") , the method comprising: training a discriminator and a classifier of the cGAN using simulation data associated with the nano-optical device (Yeung: Pg. 1 "... Current state-of-the-art machine learning methods involve training neural networks to learn the underlying relationships between photonic structures and corresponding
optical phenomena... In the photonics context, 1D tandem networks were used to design core-shell nanoparticles,[16] multilayer thin films,[17] and supercell-class metasurfaces."; Pg. 4 Col.1 "... the generator is trained to produce convincing images that deceive the discriminator, while the discriminator is trained not to be
deceived – a competition which leads to the joint and stepwise improvement of both networks via their loss functions....") ; generating a fake image of the structure pattern using a generator of the cGAN by combining the target characteristic and random noise (Yeung : Pg.6 Col.2-Pg.7 Col.1 "... For each spectrum (shown in their individual plots), a second query was performed after resampling the latent vector and slightly perturbing the starting spectrum. While not perturbing the spectrum still produced unique results on the second run (as shown in Figure S7, Supporting Information), adding small perturbations (<0.01 shifts in amplitude at various wavelengths) increased the overall uniqueness of the new designs...." - perturbations is understood as noise) ;
Yeung does not specifically teach calculating a discrimination error used to determine whether the fake image is real or fake by providing the fake image to the discriminator; determining a characteristic classification error for the fake image by providing the fake image to the classifier; and competitively training the generator, the discriminator, and the classifier with reference to the discrimination error and the characteristic classification error.
Lee teaches generating a fake image of the structure pattern using a generator of the cGAN by combining the target characteristic and random noise ( Lee: Introduction "... A generator produces fake samples from random noises, while a discriminator attempts to distinguish between these fake samples and real samples...."); calculating a discrimination error used to determine whether the fake image is real or fake by providing the fake image to the discriminator (Lee: Pg.28160 Col.1 "... Consequently, the generator in ACGAN is trained by both errors, the discrimination and the classification error, obtained from the discriminator with an auxiliary classification layer:...") & Eqns. (5) & (6)) ; determining a characteristic classification error for the fake image by providing the fake image to the classifier (Lee: See Fig.2 fake image (G(Z, L,theta.sub.g)) is provided to classifier C
PNG
media_image2.png
458
566
media_image2.png
Greyscale
) ; and
competitively training the generator, the discriminator, and the classifier with reference to the discrimination error and the characteristic classification error (Lee: Pg.28161-28162 § III Col.1 Eqn (7)-(9) and (10)-(12) in context of Fig.2 showing training of generator, discriminator and classifier).
Jang teaches computer implemented GAN with a processor and memory (Jang: Fig.1 element 15 & 16 [0027]-[0028] implementing the GAN as generator/discriminator network) for purposes of real or fake image identification in semiconductor process monitoring application).
It would have been obvious to one (e.g. a designer) of ordinary skill in the art before the effective filing date of the claimed invention to apply the teachings of Lee (2019) to Yeung (2021). The motivation to combine would have been that data augmentation (DA) and Labels as computed by the classifier can lead to low classification loses and generator losses (Pg.28161 Col.2-28162 Col.1" ControlGAN employs DA for the training of the classifier, classification loss for the test set can signficantly be reduced, which corresponds to solve the overfitting problem existing in ACGAN. Therefore, the classifier in ControlGAN guides the generator better since the generator learns from the classification loss.... It is expected that label-focus samples can be generated with low EO values since the second loss in (8) takes the most of the generator loss...."). Further motivation to combine would be that Yeung and Lee are analogous arts to the claimed invention in the field of using generative adversarial networks (GAN) based detection of real and fake data and design (Yeung: Abstract and Lee: Abstract).
It would have been obvious to one (e.g. a designer) of ordinary skill in the art before the effective filing date of the claimed invention to apply the teachings of Jang to Yeung to show computer based implementation of the generative adversarial networks (GAN) showing the obvious aspect of computer based implementation for any practical application, such as in semiconductor layout design field ( Yeung: Abstract Lee: [0023][0029] ) . Further motivation to combine would have been that Yeung and Lee are analogous arts to the instant claim in the field of using generative adversarial networks (GAN) in the field of semiconductor layout design (Yeung: Abstract Lee: [0003] [0023][0029] )
Regarding Claim 10
Yeung teaches method of claim 9, wherein the structure pattern is a dielectric pattern formed in a dielectric layer of the nano-optical device (Yeung: Fig.1 & Pgs, 2-4 showing the metal-insulator-metal (MIM) structures, where the insulator is shown as dielectric material which corresponds to target wavelength as in Fig.2 of the nano-optical device as discussed in Abstract; Pg.2 Col.2 nanophotonics structures as nano-optical device structures).
Regarding Claim 11
Yeung teaches the method of claim 10, wherein the target characteristic is a maximum transmittance wavelength of light transmitted through the dielectric pattern (Yeung: See Fig.2 where the dielectric design is changed to meet the target wavelength (as in Fig.2a at least) and verification in Fig.2b & 3).
Regarding Claim 12
Lee teaches the method of claim 9, wherein a gradient of the discrimination error is limited by applying a discriminator loss function having a gradient penalty term when training the discriminator (Lee: Pg.28162 Col.1 "... The discriminator in Control- GAN is regularized by gradient penalty to assist the con-
vergence. The gradient penalty parameter is set at 10, same as AC-WGAN-GP....").
Regarding Claim 13
Lee teaches the method of claim 9, wherein the competitively training of the generator, the discriminator, and classifier includes using an Adaptive Moment Estimation optimization algorithm to optimize neural network weightings (Lee: Pg.28162 Col.1 § B. OPTIMIZATION TECHNIQUES USED FOR
CONVERGENCE OF CONTROLGAN stating "... Adam optimization [This is mapped as Adaptive Moment Estimation optimization algorithm] is used for optimizing the parameters in ControlGAN. The parameters of the adam optimizer beta1
and beta2 are set at 0 and 0.9, respectively, which are identical to previous studies [27]....") .
Regarding Claim 14
Lee teaches the method of claim 13, wherein the competitively training of the generator, the discriminator, and classifier includes using a Two Time-scale Update Rule to stabilize the competitively training of the generator and the discriminator (Lee: Pg.28162 Col.1 § B. OPTIMIZATION TECHNIQUES USED FOR CONVERGENCE OF CONTROLGAN stating "... To reduce computing time, Tow-Timescale Update Rule (TTUR) [28] is used...." ) .
Regarding Claim 151
Lee teaches the method of claim 9, wherein the competitively training of the generator, the discriminator, and classifier includes using one of conditional batch normalization and hierarchical normalization during the competitively training of the generator and the discriminator (Lee: Pg. 28162 Col.1 "... Conditional batch normalization is used only for generators...") .
Regarding Claim 17
Teachings of claim 16 are shown by Yeung in parent claim 16.
Yeung does not specifically teach claim 17 limitations.
Lee teaches method of claim 16, wherein the generation model comprises: a generator configured to generate a fake image by combining a class label and random noise (Lee: See Fig.2 Generator G with inputs class label (L) and noise (Z)) ; a discriminator configured to calculate a discrimination error for determining whether the fake image is real or fake (Lee: Fig.2 Discriminator D with discriminator error as discussed in Pg.28159 Col.2 §II "... Simultaneously, the generator learns to deceive the discriminator by discriminative errors of the generated samples....") ; and a classifier configured to calculate a characteristic classification error for the fake image generated by the generator (Lee: Pg.28160 Col.1 "... Consequently, the generator in ACGAN is trained by both errors, the discrimination and the classification error, obtained from the discriminator with an auxiliary classification layer:...") & Eqns. (5) & (6))) .
Regarding Claim 18
Lee teaches method of claim 17, wherein the generator, the discriminator, and the classifier are competitively trained in accordance with the discrimination error and the characteristic classification error during the training of the generation model (Lee: §III showing in Eqns(7)-(9) use of training parameter gamma to show competitive training using the discrimination error (Eqn (7) and classification error (Eqn(9)).
Regarding Claim 19
Lee teaches the method of claim 18, wherein during the training of the generation model, the discriminator applies a gradient penalty to a discriminator loss function to stabilize weightings (Lee: Pg.28162 Col.1 "... The discriminator in Control- GAN is regularized by gradient penalty to assist the con-
vergence. The gradient penalty parameter is set at 10, same as AC-WGAN-GP....").
Regarding Claim 20
Lee teaches the method of claim 17, wherein during the training of the generation model, an adaptive moment estimation optimization algorithm is used to optimize neural network weightings of at least one of the generator, the discriminator and the classifier (Lee: Pg.28162 Col.1 § B. OPTIMIZATION TECHNIQUES USED FORCONVERGENCE OF CONTROLGAN stating "... Adam optimization [This is mapped as Adaptive Moment Estimation optimization algorithm] is used for optimizing the parameters in ControlGAN. The parameters of the adam optimizer beta1 and beta2 are set at 0 and 0.9, respectively, which are identical to previous studies [27]....") .
---- This page is left blank after this line ----
Claim(s) 5-7 are rejected under 35 U.S.C. 103 as being unpatentable over Yeung, in view of Lee, in view of Jang, further in view of NPL by M. Lee et al (Lee-2 hereafter). "Regularization methods for generative adversarial networks: An overview of recent studies." arXiv preprint arXiv:2005.09165 (2020).
Regarding Claim 5
Teachings of Yeung, Lee and Jang are shown in the parent claim 1.
Lee teaches The system of claim 1, wherein at least one of the generator and the discriminator is trained using at least one of conditional batch normalization (Lee: Pg.28162 Col.2 "... Conditional batch normalization is used only for generators [30]....") .
Lee-2 teaches wherein at least one of the generator and the discriminator is trained using … layer normalization (Lee-2: §3.5)
It would have been obvious to one (e.g. a designer) of ordinary skill in the art before the effective filing date of the claimed invention to apply the teachings of Lee-2 to Lee as they are works of the same inventor Minhyeok Lee in the same field of application of Generative Adversarial Network (GAN) (See Abstracts in Lee and Lee-2).
Regarding Claim 6
Teachings of Yeung, Lee and Jang are shown in the parent claim 1.
Lee states spectral normalization is not used specifically (Lee: Pg.28162 Col.2).
Lee-2 teaches the system of claim 1, wherein at least one of the generator and the discriminator is trained using spectral normalization (Lee-2: Pg. 8 "... Spectral Normalization (SN) [47] is one of the most conventional weight normalization methods, which introduces the spectral norm of weight matrices for the GAN training. The...") . Motivation to combine would be as presented in parent claim.
Regarding Claim 7
Lee teaches the system of claim 6, wherein the discriminator is trained using a hinge loss function (Lee: Pg.12 "... Recently, a more improved loss function for the discriminator, called hinge loss, was proposed and employed in various GANs..." ). Motivation to combine would be as presented in parent claim.
Conclusion
All claims are rejected.
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Examiner’s Note: Examiner has cited particular columns and line numbers in the references applied to the claims above for the convenience of the applicant. Although the specified citations are representative of the teachings of the art and are applied to specific limitations within the individual claim, other passages and figures may apply as well. It is respectfully requested from the applicant in preparing responses, to fully consider the references in their entirety as potentially teaching all or part of the claimed invention, as well as the context of the passage as taught by the prior art or disclosed by the Examiner.
In the case of amending the claimed invention, Applicant is respectfully requested to indicate the portion(s) of the specification which dictate(s) the structure relied on for proper interpretation and also to verify and ascertain the metes and bounds of the claimed invention.
Communication
Any inquiry concerning this communication or earlier communications from the examiner should be directed to AKASH SAXENA whose telephone number is (571)272-8351. The examiner can normally be reached Mon-Fri, 7AM-3:30PM.
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, RYAN PITARO can be reached on (571) 272-4071. 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.
AKASH SAXENA
Primary Examiner
Art Unit 2188
/AKASH SAXENA/Primary Examiner, Art Unit 2188 Friday, November 28, 2025
1 Also see US 20220335689 A1 [0104] In this implementation, the generator and the base network of the discriminator and classifier is embodied by separate convolutional neural networks with strided convolutions. The generator employs batch normalization with LeakyReLU activation functions. The base network of the discriminator and classifier employs LeakyReLU and Dropout activation functions as well as LeakyReLU activation function with batch normalization.