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 .
The amendments were received 10/17/2025. Claims 1-4, 7, and 9 where claims 1-4, 7, and 9 were previously presented and claims 5, 6, and 8 were cancelled.
Claim Objections
Claims 1 and 9 are objected to because of the following informalities: Claims 1 and 9 were amended to recite a new phrase “a CP decomposition” where the acronym “CP” is not first defined before usage. Appropriate correction is required.
35 USC § 101
The applicant amended the claims to illustrate the usage of a neural network including monitoring it so that the system can perform a quantization process on the neural network. In view of the amendments, including the usage of the quantized network, the respective claims are directed towards a practical application and thus the respective 35 USC 101 rejections for being directed towards an abstract idea have been withdrawn.
Claim Rejections - 35 USC § 101
Claim 9 is rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter. The claim(s) does/do not fall within at least one of the four categories of patent eligible subject matter because the claim is directed towards software per se. The respective components of database constructor, parameter generator, and network constructor are illustrated in Figure 1 where paragraphs 15, 16, and 37 discuss that the components in Figure 1 are the functional configuration where the functions are implemented “via software”. Therefore, the respective claim limitations are directed towards software and the claim is rejected for being directed towards software per se.
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.
Claims 1 and 9 are rejected under 35 U.S.C. 103 as being unpatentable over Gibson et al [US 2017/0323197 A1] (from IDS) in view of Baskaran et al [US 10,936,569].
With respect to claim 1, Gibson teaches an implementation method of implementing a neural network on a computer, the implementation method comprising:
performing, by using a processor, a plurality of tests by inputting a plurality of test datasets to the neural network (see paragraphs [0056] and [0057]; initial training of a neural network is performed, i.e. preparing the neural network);
obtaining, by using the processor, through the plurality of tests, statistical information of tensors handled by the neural network (paragraph [0059]; the system can determine statistical information including distribution/range of data values for each layer);
generating, by using a processor, a quantization parameter set by quantizing values of the tensors based on the statistical information and the neural network (see paragraph [0056]; the system utilizes the statistical information of each layer to determine a quantized parameter set including quantization type/format);
generating and outputting, by using the processor, a quantized network by quantizing the neural network using of the quantized parameter set (see paragraphs [0055]-[0056]; a quantization step is performed to convert the weights and/or input data to the fixed point number format according to the determined);
and implementing the quantized network in the computer (see paragraphs [0056] and [0088]; the system can implement the quantized network and utilize it for future inputs),
wherein the generating includes determining a quantization type for each of a plurality of layers that make up the neural network (see paragraphs [0055]-[0056]; the system can determine a quantization type/format for each of the layers),
the quantization type is determined based on a redundancy of the tensor included in each of the plurality of layers (see Gibson, paragraphs [0051] and [0059]; the system determines the redundancy of the tensor for each layer).
Gibson does not appear to explicitly teach:
the redundancy is defined as (J-K)/J or K/J, wherein the redundancy is determined based on a result of tensor decomposition of the tensor, the tensor being a J-dimensional tensor that is a multidimensional array with J dimensions, and the tensor decomposition being a CP decomposition that decomposes the tensor into a K-dimensional core tensor and J factor matrices, where J is an integer greater than or equal to 2 and K is an integer less than J and greater than or equal to 1.
Baskaran teaches wherein the redundancy is determined based on a result of tensor decomposition of the tensor, the tensor being a J-dimensional tensor that is a multidimensional array with J dimensions, and the tensor decomposition being a CP decomposition that decomposes the tensor into a K-dimensional core tensor and J factor matrices, where J is an integer greater than or equal to 2 and K is an integer less than J and greater than or equal to 1 (see col 1, line 28 through col 2, line 6; the system can decompose the multi-dimensional tensor into factor matrices and a core tensor).
It would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to modify the quantization method of Gibson by using widely known and used tensor decomposition techniques as taught by Baskaran in order to decompose tensors into a simpler form to allow for easier processing of multi-dimensional data where the system is able to utilize widely-used and known techniques to perform the decomposition (see Baskaran, col 1, lines 47-50).
Gibson in view of Baskaran teach the redundancy is defined as (J-K)/J or K/J (see Baskaran, col 1, line 28 through col 2, line 6; see Gibson, paragraphs [0051] and [0059]; the system can determine the size of the full tensor and the core tensor).
With regard to claim 9, this claim is substantially similar to claim 1 and is rejected for similar reasons as discussed above.
Claim 2 is rejected under 35 U.S.C. 103 as being unpatentable over Gibson et al [US 2017/0323197 A1] in view of Baskaran et al [US 10,936,569] in further view of Lee et al [US 2018/0341857 A1].
With regard to claim 2, Gibson in view of Baskaran teach all the claim limitations of claim 1 as discussed above.
Gibson in view of Baskaran teach wherein the determining includes selecting the quantization type from among a plurality of numerical transformation types each performing different numerical transformations on the tensor; (see paragraphs [0055]-[0056] and [0075]; the system determines the quantization type/format and as well as other information to perform the numerical transformations to do the quantization conversion).
Gibson in view of Baskaran do not appear to explicitly teach the plurality of numerical transformation types include logarithmic transformation.
Lee teaches the plurality of numerical transformation types include logarithmic transformation (see paragraphs [0013] and [0065] and [0067]; the system has means to perform logarithmic transformation).
It would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to modify the quantization method of Gibson in view of Baskaran by utilizing statistical analysis to determine to use a log quantization method as taught by Lee in order to be able to address situations where more quantization levels are needed for near-zero weights when the weights have a high frequency of near-zero weights so as to not diminish the influence of those small weights thus allowing significant reduction in size of the model and the computational amount while minimizing any accuracy loss.
Claim 3 is rejected under 35 U.S.C. 103 as being unpatentable over Gibson et al [US 2017/0323197 A1] in view of Baskaran et al [US 10,936,569] in further view of Sather et al [US 12,136,039].
With regard to claim 3, Gibson in view of Baskaran teach all the claim limitations of claim 1 as discussed above.
Gibson in view of Baskaran teach wherein the determining includes selecting the quantization type from among a plurality of fineness types each having different degrees of fineness of quantization, and the plurality of fineness types include an N-bit fixed-point type, where N is an integer greater than or equal to 2 (see paragraphs [0055]-[0056]; the system can utilize a wide variety of different fixed points for each of the layers).
Gibson in view of Baskaran do not appear to explicitly teach a ternary type.
Sather teaches a ternary type (see col 4, lines 5-7; the system can utilize a ternary method of quantization).
It would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to modify the quantization method of Gibson in view of Baskaran by known quantization methods including ternary as taught by Sather in order to provide options to the design/implementation of the various layers be being able to not only have fixed point representation but can also utilize ternary format to provide removal of influence of non-important weights while focusing on the weights that most contribute to the loss/accuracy of the system.
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Gibson et al [US 2017/0323197 A1] in view of Baskaran et al [US 10,936,569] in further view of Lee et al [US 2018/0341857 A1] and in further view of Sather et al [US 12,136,039].
With regard to claim 4, Gibson in view of Baskaran in further view of Lee teach all the claim limitations of claims 1 and 2 as discussed above.
Gibson in view of Baskaran in further view of Lee teach wherein the determining includes selecting the quantization type from among a plurality of fineness types each having different degrees of fineness of quantization, and the plurality of fineness types include an N-bit fixed-point type, where N is an integer greater than or equal to 2 (see paragraphs [0055]-[0056]; the system can utilize a wide variety of different fixed points for each of the layers).
Gibson in view of Baskaran in further view of Lee do not appear to explicitly teach a ternary type.
Sather teaches a ternary type (see col 4, lines 5-7; the system can utilize a ternary method of quantization).
It would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to modify the quantization method of Gibson in view of Baskaran in further view of Lee by known quantization methods including ternary as taught by Sather in order to provide options to the design/implementation of the various layers be being able to not only have fixed point representation but can also utilize ternary format to provide removal of influence of non-important weights while focusing on the weights that most contribute to the loss/accuracy of the system.
Claim 7 is rejected under 35 U.S.C. 103 as being unpatentable over Gibson et al [US 2017/0323197 A1] in view of Baskaran et al [US 10,936,569] in further view of Li et al [US 2019/0034784 A1].
With regard to claim 7, Gibson in view of Baskaran teach all the claim limitations of claim 1 discussed above.
Gibson in view of Baskaran do not appear to explicitly teach wherein the quantization type is determined as a type with lower fineness as the redundancy increases.
Li teaches lower fineness as the redundancy increases (see paragraph [0063]; that the accuracy or fineness will decrease as the compression increases which is similar to redundancies since the redundancies can be removed/reduced via tensor decomposition).
It would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to modify the quantization method of Gibson in view of Baskaran by a determination for a target compression ratio or amount of redundancies to remove/prune from the network as taught by Li in order to reduce the space/storage consumption as well as the computational consumption of the system while still having means to minimize the error and maintain a desired accuracy of the system.
Gibson in view of Baskaran in further view of Li teach wherein the quantization type is determined as a type with lower fineness as the redundancy increases (see Li, paragraph [0063]; see Gibson, paragraphs [0055]-[0056]; the system can determine a quantization type/format for each of the layers while maintaining a preferred compression ratio where the higher the redundancies, the more compressed the model can be which in turn increases the error/lower the precision/accuracy of the system).
Response to Arguments
Applicant’s arguments (see the second to last paragraph on page 5 through the third paragraph on page 7) with respect to the 35 USC 101 rejections have been fully considered and are persuasive. The 35 USC 101 rejections of the claims have been withdrawn. The applicant provides amendments and arguments, including on page 6, that discuss the improvement or technical effect of the claim limitations. In view of the amendments and respective arguments relating to the analysis and updating of the neural network, the respective 35 USC 101 rejections have been withdrawn.
Applicant’s arguments (see section 35 USC 102/103 on page 7 through last paragraph on page 9) with respect to the rejection(s) of claim(s) under 35 USC 102/103 have been fully considered and are persuasive. Therefore, the rejection has been withdrawn. However, upon further consideration, a new ground(s) of rejection is made in view of Baskaran. The applicant amended the claims to include new limitations that required further search and consideration. As seen from the 35 USC 103 rejections, a new reference was found that, when combined, would teach or suggest the claim limitations as recited including the CP decomposition.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MARC S SOMERS whose telephone number is (571)270-3567. The examiner can normally be reached M-F 11-8 EST.
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, Ann Lo can be reached at 5712729767. 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.
/MARC S SOMERS/Primary Examiner, Art Unit 2159 11/20/2025
Prosecution Insights