NON-FINAL REJECTION, FIRST DETAILED ACTION
Status of Prosecution
The present application, 18/404,280 filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
The application was filed on January 4, 2024.
Claims 1-20 are pending and all are rejected. Claims 1, 9 and 17 are independent.
Status of the Claims
Claims 1-20 are rejected under 35 U.S.C. § 101 because the claimed invention is directed to an abstract idea without significantly more.
Claims 1-5, 8-13 and 16-20 are rejected under 35 USC. § 103 as being unpatentable over non-patent literature, Kingetsu et al. (“Kingetsu”), “Neural Network Module Decomposition and Recomposition with Superimposed Masks,” published in 2023 in view of non-patent literature, Fang et al. (“Fang”), “DepGraph: Towards Any Structural Pruning,” published in 2023 in further view of Jiang et al. (“Jiang”), United States Patent Application Publication 2022/0103823 published on March 31, 2022.
Claims 6-7 and 14-15 are rejected under 35 USC. § 103 as being unpatentable over, Kingetsu in view of Fang in further view of Jiang in further view of non-patent literature, Lin et al. (“Lin”), “Towards Optimal Structured CNN Pruning via Generative Adversarial Learning,” published in 2019.
Claim Rejections – § 101 Subject Matter Eligibility
Claims 1-20 are rejected under 35 U.S.C. § 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding representative claim 1, at step 1, the claim recites a processor, and therefore is a manufacture, which is a statutory category of invention. See MPEP § 2106.03.
At step 2A, prong one, the claim recites a processor comprising one or more circuits.
The following limitations are the abstract idea of a mathematical calculation. See MPEP § 2106.04(a)(2)(I)(C):
to generate one or more second masks to modify one or more second portions of a neural network based, at least in part, on one or more first masks of one or more first portions of the neural network from which the one or more second portions of the neural network depend.
Therefore, the claim recites at least one abstract idea per this part of the analysis.
At step 2A prong 2, the claim language is analyzed to determine whether it recites additional elements that integrate the judicial exception into a practical application. See MPEP § 2106.04(d).
The claim element of the processor comprising one more circuits to perform the abstract idea is an additional element that generally links the use of the judicial exception to a particular technological environment or field of use, specifically model training and prediction. See MPEP §§ 2106.04(d), 2106.05(h). Additionally, it is an additional element that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer. See MPEP § 2106.05(f)(1).
Accordingly, these additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea. The claim is therefore directed to an abstract idea.
Next, at step 2B of the analysis, the claim is considered if it recites additional elements that amount to significantly more than the judicial exception. See MPEP § 2106.05.
As discussed above with respect to integration of the abstract idea into a practical application, the additional element of the processor comprising one or more circuits amount to nothing more than linking the use of the judicial exception to a particular technological environment or field of use. See MPEP § 2106.05(h).
Therefore, claim 1 is ineligible.
As to dependent claims 2-8, the analysis of the parent claim is incorporated. In the step 2A, prong 2 analysis, the additional limitations are merely also the abstract idea of a mathematical calculation. See MPEP § 2106.04(a)(2)(I)(C):
The claims are also ineligible.
As to independent claim 9, the analysis of claim 1 is incorporated. Where it differs is in the step zero analysis, in which the system includes a processor is a manufacture and thus statutory.
As to dependent claims 10-16, they are similarly rejected as to claims 2-8.
As to independent claim 17, the analysis of the claim 1 is incorporated. Where it differs is in the step 0 analysis, in which the method is a process and thus statutory.
As to dependent claims 18-20, they are similarly rejected.
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 of this title, 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.
A.
Claims 1-5, 8-13 and 16-20 are rejected under 35 USC. § 103 as being unpatentable over non-patent literature, Kingetsu et al. (“Kingetsu”), “Neural Network Module Decomposition and Recomposition with Superimposed Masks,” published in 2023 in view of non-patent literature, Fang et al. (“Fang”), “DepGraph: Towards Any Structural Pruning,” published in 2023 in further view of Jiang et al. (“Jiang”), United States Patent Application Publication 2022/0103823 published on March 31, 2022.
As to Claim 1, Kingetsu teaches: A processor comprising:
one or more circuits to generate one or more second masks to modify one or more second portions of a neural network and one or more first masks of one or more first portions of the neural network (Kingetsu: Fig. 2, Introduction, “In our approach, we modularize the DNN by pruning edges using supermasks [3]. A supermask is a binary score matrix that indicates whether each edge of the network should be pruned.” Each supermask (i.e. one of the first or second masks) for pruning (i.e. pruning) edges (i.e. portions of a neural network).).
PNG
media_image1.png
521
1389
media_image1.png
Greyscale
Kingetsu may not explicitly teach: one or more circuits to generate one or more second masks to modify one or more second portions of a neural network based, at least in part, on one or more first masks of one or more first portions of the neural network from which the one or more second portions of the neural network depend.
Fang teaches in general concepts related to modeling dependency between layers and comprehensively group coupled parameters for pruning (Fang: Abstract). Specifically, Fang teaches that dependencies between different layers may be captured, for different types of dependencies (Fang: Fig. 2, Sec. 1., “In this paper, we strive for a generic scheme towards any structural pruning, where structural pruning over arbitrary network architectures is executed in an automatic fashion. At the heart of our approach is to estimate the Dependency Graph (DepGraph), which explicitly models the interdependency between paired layers in neural networks”). A dependency graph is constructed to facilitate this (Fang: Sec. 3.2, grouping matrix is used).
PNG
media_image2.png
270
992
media_image2.png
Greyscale
Fang however does not explicitly teach the use of masks for relating the pruning decisions and the dependencies.
Jiang teaches in general concepts related to using one or more binary masks for to generate weight coeffcients of layers of a neural network to decode video data (Jiang: Abstract). Specifically, Jiang teaches that micro-structure pruning on different portions of the neural network is based on binary masks represented as weight matrices (Jiang: par. 0034).
It would have been obvious to a person having ordinary skill in the art at a time before the effective filing date of the invention to have modified the Kingetsu device by including using masks that incorporate dependency information for the modifications of the neural network as taught and disclosed by Fang and Jiang. Such a person would have been motivated to do so with a reasonable expectation of success to do so to allow for a generalized, any-dependency, use of masking approach to making changes to the neural network effectively (Fang: Sec. 2, “Unfortunately, existing techniques still depend on empirical rules or predefined architectural patterns, rendering them insufficiently versatile for all structural pruning applications. In this study, we present a general approach to resolve this challenge, demonstrating that addressing parameter dependency effectively generalizes structural pruning across a wide array of networks, resulting in satisfactory performance on several tasks.”).
As to Claim 2, Kingetsu, Fang and Jiang teach the limitations of claim 1.
Kingetsu, Fang and Jiang as combined further teaches: identify a dependency between the one or more first portions of the neural network and the one or more second portions of the neural network (Fang: Fig. 2, Sec. 1., “At the heart of our approach is to estimate the Dependency Graph (DepGraph), which explicitly models the interdependency between paired layers in neural networks);
identify a neuron dimension of the one or more first masks (Fang: Sec. 3.3, each of the K prunable dimension is indexed with parameter w, in w[k], per equation 4, which is a neuron dimension; Jiang: par. 0031, binary masks have multidimensional (5d) weight tensors, including channel, convolution kernels, etc.); and
identify a neuron dimension of the one or more second masks using the neuron dimension of the one or more first masks (Examiner asserts per the combination, the dimensions are cross-channel and related and the second mask dimensions would be thus inter-identified).
As to Claim 3, Kingetsu, Fang and Jiang teach the limitations of claim 1.
Kingetsu, Fang and Jiang as combined further teaches wherein the one or more circuits are to apply the one or more second masks to the one or more second portions of the neural network to modify the one or more second portions of the neural network (Examiner asserts that the combination for the group pruning would be applicable to the different corresponding masked-portions).
As to Claim 4, Kingetsu, Fang and Jiang teach the limitations of claim 1.
Fang further teaches: wherein the one or more first portions of the neural network comprise a concatenation operation or addition (Fang: Fig. 2(c), concatenation dependency).
As to Claim 5, Kingetsu, Fang and Jiang teach the limitations of claim 1.
Fang further teaches: wherein the one or more circuits are to indicate one or more dependencies of the one or more first portions of the neural network and the one or more second portions of the neural network using one or more graphs (Fang: Fig. 2, Sec. 1., “At the heart of our approach is to estimate the Dependency Graph (DepGraph), which explicitly models the interdependency between paired layers in neural networks; Sec. 3.2, grouping matrix is used).
As to Claim 8, Kingetsu, Fang and Jiang teach the limitations of claim 1.
Fang, Jiang and Kingetsu as combined further teaches: wherein the one or more second portions of the neural network are modified to prune one or more neurons (Examiner asserts that the combination for the group pruning would be applicable to the different corresponding masked-portions’s neurons).
As to Claim 9, it is rejected for similar reasons as claim 1.
As to Claim 10, it is rejected for similar reasons as claim 2.
As to Claim 11, it is rejected for similar reasons as claim 3.
As to Claim 12, it is rejected for similar reasons as claim 4.
As to Claim 13, it is rejected for similar reasons as claim 5.
As to Claim 16, it is rejected for similar reasons as claim 8.
As to Claim 17, it is rejected for similar reasons as claim 1.
As to Claim 18, it is rejected for similar reasons as claim 2.
As to Claim 19, it is rejected for similar reasons as claim 3.
As to Claim 20, it is rejected for similar reasons as claim 5.
B.
Claims 6-7 and 14-15 are rejected under 35 USC. § 103 as being unpatentable over non-patent literature, Kingetsu et al. (“Kingetsu”), “Neural Network Module Decomposition and Recomposition with Superimposed Masks,” published in 2023 in view of non-patent literature, Fang et al. (“Fang”), “DepGraph: Towards Any Structural Pruning,” published in 2023 in further view of Jiang et al. (“Jiang”), United States Patent Application Publication 2022/0103823 published on March 31, 2022 in further view of non-patent literature, Lin et al. (“Lin”), “Towards Optimal Structured CNN Pruning via Generative Adversarial Learning,” published in 2019.
As to Claim 6, Kingetsu, Fang and Jiang teach the limitations of claim 1.
Fang, Jiang and Kingetsu may not explicitly teach: wherein the one or more second masks are generated by a concatenation of the one or more first masks.
Lin teaches in general concepts related to structured pruning using jointly pruning filters as well as other structures in an end-to-end manner (Lin: Abstract). Specifically, Lin teaches that a soft mask may be used, where a concatenation of all weights together to remove redundancy of different networks used in joint learning (Lin: Sec. 3.4, eq. 14).
It would have been obvious to a person having ordinary skill in the art at a time before the effective filing date of the invention to have modified the Kingetsu-Fang-Jiang device by including using combined masks trhough concatenation as taught and disclosed by Lin. Such a person would have been motivated to do so with a reasonable expectation of success to do so to allow for enhancing ifnromation flow by removal of redundancies via the combined soft mask (Lin: Sec. 3.4).
As to Claim 7, Kingetsu, Fang and Jiang teach the limitations of claim 1.
Fang, Jiang and Kingetsu as combined further teaches: wherein the one or more second masks are generated by a split of the one or more first masks.
Lin teaches in general concepts related to structured pruning using jointly pruning filters as well as other structures in an end-to-end manner (Lin: Abstract). Specifically, Lin teaches that a soft mask may be used, where a concatenation of all weights together to remove redundancy of different networks used in joint learning (Lin: Sec. 3.4, eq. 14). While this is not a teaching of a split mask, Examiner asserts a reverse concatenation may be necessary for certain operations other than pruning and the splitting would be the corresponding operation.
It would have been obvious to a person having ordinary skill in the art at a time before the effective filing date of the invention to have modified the Kingetsu-Fang-Jiang device by including using split masks through concatenation as taught and disclosed by Lin. Such a person would have been motivated to do so with a reasonable expectation of success to do so to allow for enhancing ifnromation flow by removal of redundancies via the altered soft mask (Lin: Sec. 3.4).
As to Claim 14, it is rejected for similar reasons as claim 6.
As to Claim 15, it is rejected for similar reasons as claim 7.
Conclusion
Additional prior art: Li et al., US PG Pub 2023/0084203 (Mar. 16, 2023) (teaching model pruning using a graph neural network).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JAMES T TSAI whose telephone number is (571)270-3916. The examiner can normally be reached M-F 8-5 Eastern.
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, Viker Lamardo can be reached at 571-270-5871. 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.
/JAMES T TSAI/ Primary Examiner, Art Unit 2147
Prosecution Insights