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 .
Response to Amendment and Arguments
Applicant’s amendment filed on January 29, 2026 has been entered and made of record. Claims 1 and 3-24 are pending and are being examined in this application.
In light of Applicant’s amendments to the claims, the 112(f) claim interpretation, 112(b) rejection, and 101 rejections are withdrawn.
Applicant’s arguments with respect to the 102 and 103 rejections have been considered, but are moot in view of the new ground(s) of rejection provided below.
Allowable Subject Matter
Claims 12, 13, and 22 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Claim Rejections - 35 USC § 102
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.
Claims 1, 5-9, 14-17, 19-21, and 23 are rejected under 35 U.S.C. 102(a)(1) and (a)(2) as being anticipated by Mehnert (US Pub. 20210192285).
Referring to claim 1, Mehnert discloses A processor comprising: one or more circuits to [fig. 1; par. 33; system 100 executes training hardware 10 and inference hardware 20]:
generate a first set of results using a sparse neural network [fig. 1; pars. 34-36; inference hardware 20 includes limited hardware capacity than training hardware 10, so it is not possible to represent complete neural network 11 on interference hardware 20 at once; this means that second neural network 21 of inference hardware 20 includes only a portion of neural network 11 (i.e., is sparser)] based on input information [fig. 1; pars. 35 and 36; image data is fed to first neural network 21, which generates output data 22];
generate a second set of results using an altered version of the sparse neural network based on the input information [fig. 1; pars. 35 and 36; the same image data is fed to second neural network 11, which generates output data 12], the altered version comprising a dense version of the sparse neural network [fig. 1; pars. 34-36; note that neural network 11 is a complete (i.e., denser) version of neural network 21];
compare the first and second sets of results to determine one or more inconsistent results [fig. 1; pars. 35 and 36; an error ascertainment unit 30 ascertains a deviation (i.e., difference) or error between output data 22 and output data 21, which means output data 22 and output data 21 are inconsistent]; and
identify a flaw in the altered version of the sparse neural network based at least on the one or more inconsistent results [fig. 1; pars. 35 and 36; note the deviation or error].
Referring to claim 5, Mehnert discloses The processor of claim 1, wherein the one or more circuits are to perform one or more data perturbations on a first portion of the input information to generate a second portion of the input information, wherein the inconsistent results are generated from the second portion of the input information [pars. 21, 34-36, 38, 39, 43, and 52; the image fed into neural network 11 may be changed (i.e., perturbed) according to noise parameters; the input fed into neural network 11 and neural network 12 would then comprise the changed image / added noise (i.e., second portion) and the original image (i.e., first portion)].
Referring to claim 6, Mehnert discloses The processor of claim 1, wherein the one or more circuits are to compute a loss function based, at least in part, on the one or more inconsistent results [pars. 36-38; neural network 11 is iteratively trained until the error is essentially zero; this is performed using back propagation, which entails computing a loss function].
Referring to claim 7, Mehnert discloses The processor of claim 5, wherein the first and second sets are determined as consistent results based, at least in part, on using the first portion of the input information being identical to the second input information [pars. 36-38; neural network 11 is iteratively trained until the error is essentially zero (i.e., output 21 and output 22 are bit-identical)].
Referring to claim 8, Mehnert discloses The processor of claim 1, wherein the input information comprises images [par. 35; note the image].
Referring to claim 9, see at least the rejection for claim 1. Mehnert further discloses A computer-implemented method comprising: causing two or more neural networks to perform the claimed steps [fig. 1; pars. 35 and 36; note neural network 11 and neural network 21].
Referring to claim 14, Mehnert discloses The computer-implemented method of claim 9, wherein causing the sparse neural network and the altered version to generate the inconsistent results includes training one or more neural networks based, at least in part, on distillation loss and prediction loss [pars. 36-38; neural network 11 is iteratively trained until the error is essentially zero; the error in this case is based on ascertaining the deviation (i.e., differences) between output 22 and output 21 (i.e., knowledge distillation loss)]; the iterative training is performed using back propagation, which entails computing a (prediction) loss function].
Referring to claim 15, Mehnert discloses The computer-implemented method of The computer-implemented method of wherein the altered version of the sparse neural network has a measure of sparsity that is different from the sparse neural network [fig. 1; pars. 34-36; note that neural network 21 is an incomplete (i.e., sparser) version of neural network 11].
Referring to claim 16, Mehnert discloses The computer-implemented method of claim 11, further including: generating the second portion of the input information based, at least in part, on performing one or more data perturbations on the first portion of the input information [pars. 21, 34-36, 38, 39, 43, and 52; the image fed into neural network 11 may be changed (i.e., perturbed) according to noise parameters; the input fed into neural network 11 and neural network 12 would then comprise the changed image / added noise (i.e., second portion) and the original image (i.e., first portion)].
Referring to claim 17, see at least the rejection for claim 1. Mehnert further discloses A computer system comprising: one or more processors and memory storing executable instructions that, if performed by the one or more processors: perform the claimed steps [fig. 1; par. 33; system 100 executes training hardware 10 and inference hardware 20].
Referring to claim 19, Mehnert discloses The computer system of claim 17, wherein the one or more processors are to cause the sparse neural network and the altered version to generate the inconsistent results by modifying a first portion of the input information to generate a second portion of the input information to satisfy one or more similarity conditions [pars. 21, 34-39, 43, and 52; the image fed into neural network 11 may be changed (i.e., perturbed) according to noise parameters; the input fed into neural network 11 and neural network 12 would then comprise the changed image / added noise (i.e., second portion) and the original image (i.e., first portion); the noise parameters are updated via iterative training until the error is essentially zero (i.e., output 21 and output 22 are bit-identical)].
Referring to claim 20, Mehnert discloses The computer system of claim 17, the one or more processors are to cause the sparse neural network and the altered version to generate the inconsistent results by training one or more other neural networks based, at least in part, on a prediction loss of the inconsistent results [pars. 36-38; neural network 11 is iteratively trained until the error is essentially zero; the iterative training is performed using back propagation, which entails computing a (prediction) loss function].
Referring to claim 21, Mehnert discloses The computer system of claim 17, wherein the one or more processors are to cause the sparse neural network and the altered version to generate the inconsistent results by training one or more other neural networks based, at least in part, on a confidence measure of the inconsistent results [pars. 30 and 36-38; neural network 11 is iteratively trained until the error is essentially zero; the iterative training is performed using back propagation, which entails computing a (prediction) loss function; a certain reliability (i.e., threshold confidence) may be ensured].
Referring to claim 23, Mehnert discloses The computer system of claim 19, wherein the one or more processors are to generate the second portion of the input information based, at least in part, on one or more changes to the first portion of the input information [pars. 21, 34-36, 38, 39, 43, and 52; the image fed into neural network 11 may be changed (i.e., perturbed) according to noise parameters; the input fed into neural network 11 and neural network 12 would then comprise the changed image / added noise (i.e., second portion) and the original image (i.e., first portion)].
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 3 and 11 are rejected under 35 U.S.C. 103 as being unpatentable over Mehnert in view of Casagrande (US Pub. 20220383851).
Referring to claim 3, Mehnert does not appear to explicitly disclose The processor of claim 1, wherein the one or more circuits are to use another neural network to generate a next iteration of input information based, at least in part, on the inconsistent results.
However, Casagrande discloses The processor of claim 1, wherein the one or more circuits are to use another neural network to generate a next iteration of input information based, at least in part, on the inconsistent results [pars. 46-50; a perturbation neural network generates perturbed spectral representations of input to a vocoder neural network, which are used to train the vocoder neural network].
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the iterative training of a neural network taught by Mehnert so that the changed image is generated by a perturbation neural network as taught by Casagrande, with a reasonable expectation of success. The motivation for doing so would have been to train the neural network using a more diverse training set [Casagranda, par. 50].
Referring to claim 11, Mehnert does not appear to explicitly disclose The computer-implemented method of The computer-implemented method of wherein causing the sparse neural network and the altered version to generate the inconsistent results includes: training another neural network to generate a second portion of the input information based, at least in part, on one or more modifications to a first portion of the input information
However, Casagrande discloses The computer-implemented method of The computer-implemented method of wherein causing the sparse neural network and the altered version to generate the inconsistent results includes: training another neural network to generate a second portion of the input information based, at least in part, on one or more modifications to a first portion of the input information [pars. 46-50; a perturbation neural network generates perturbed spectral representations of input to a vocoder neural network, which are used to train the vocoder neural network].
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the iterative training of a neural network taught by Mehnert so that the changed image is generated by a perturbation neural network as taught by Casagrande, with a reasonable expectation of success. The motivation for doing so would have been to train the neural network using a more diverse training set [Casagranda, par. 50].
Claims 4 and 24 are rejected under 35 U.S.C. 103 as being unpatentable over Mehnert in view of Abeloe (US Pat. 10324467).
Referring to claim 4, Mehnert does not appear to explicitly disclose The processor of claim 1, wherein the one or more circuits cause another neural network to detect differences between the first and the second sets of results.
However, Abeloe discloses The processor of claim 1, wherein the one or more circuits are cause another neural network to detect differences between the first and the second sets of results [col. 7, line 53 – col. 8, line 8; a second neural network receives outputs from a first neural network and a third neural network, the first neural network and the third neural network being substantially identical; the second neural network compares the outputs to detect a difference between the output form the first neural network and the second neural network].
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the iterative training of a neural network taught by Mehnert so that another neural network is used to detect differences between outputs as taught by Abeloe, with a reasonable expectation of success. The motivation for doing so would have been to automatically self-correct one or more neural networks after detecting a triggering event [Abeloe, col. 1, lines 17-22].
Referring to claim 24, Mehnert does not appear to explicitly disclose The computer system of claim 19, wherein the one or more processors are to implement a neural network to detect differences between the first portion of the input information and the second portion of the input information.
However, Abeloe discloses The computer system of claim 19, wherein the one or more processors are to implement a neural network to detect differences between the first portion of the input information and the second portion of the input information [col. 7, line 53 – col. 8, line 8; a second neural network receives outputs from a first neural network and a third neural network, the first neural network and the third neural network being substantially identical; the second neural network compares the outputs to detect a difference between the output from the first neural network and the second neural network].
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the iterative training of a neural network taught by Mehnert so that another neural network is used to detect differences between outputs as taught by Abeloe, with a reasonable expectation of success. The motivation for doing so would have been to automatically self-correct one or more neural networks after detecting a triggering event [Abeloe, col. 1, lines 17-22].
Claims 10 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Mehnert in view of Gao et al. (US Pub. 20220261649).
Referring to claim 10, Mehnert does not appear to explicitly disclose The computer-implemented method of claim 9, wherein the sparse neural network is generated based, at least in part, on pruning the dense version of the sparse neural network.
However, Gao discloses The computer-implemented method claim 9, wherein the sparse neural network is generated based, at least in part, on pruning the dense version of the sparse neural network [pars. 46-50; a perturbation neural network generates perturbed spectral representations of input to a vocoder neural network, which are used to train the vocoder neural network].
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the iterative training of a neural network taught by Mehnert so that the changed image is generated by a perturbation neural network as taught by Gao, with a reasonable expectation of success. The motivation for doing so would have been to train the neural network using a more diverse training set [Casagranda, par. 50].
Referring to claim 18, Mehnert does not appear to explicitly disclose The computer system of claim 17, wherein the sparse neural network is a compressed version of the dense version of the sparse neural network.
However, Gao discloses The computer system of claim 17, wherein the sparse neural network is a compressed version of the dense version of the sparse neural network [par. 73; a neural network is compressed via pruning based on a target sparsity].
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the inference hardware neural network taught by Mehnert so that it is a compressed version of the complete neural network as taught by Gao, with a reasonable expectation of success. The motivation for doing so would have been to compress the neural network according to a target sparsity [Gao, par. 73].
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.
The following prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Kim (US Pub. 20210182670) discloses comparing results from two neural networks.
Khalegi et al. (US Pub. 20210326756) discloses testing the accuracy of a sparse model on a validation dataset, which is a part of the original training dataset.
Contact Information
Any inquiry concerning this communication or earlier communications from the examiner should be directed to GRACE PARK whose telephone number is (571)270-7727. The examiner can normally be reached M-F 8AM-5PM.
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, TAMARA KYLE can be reached at (571)272-4241. 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.
/Grace Park/Primary Examiner, Art Unit 2144
Prosecution Insights