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 .
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-16 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., an abstract idea) without significantly more. Apparatus claims 9-16 will be addressed first, followed by method claims 1-8.
Regarding claim 9, under the Alice Framework Step 1 analysis, the claim falls within the four statutory categories of patentable subject matter: an apparatus.
Under the Alice Framework Step 2A Prong 1 analysis, claim 9 recites Mathematical Concepts. The claim recites Mathematical Calculations, which is specifically identified as an exemplar in the Mathematical Concepts grouping of abstract ideas:
“converting the data into embedding vectors corresponding to respective indices through an embedding layer;
generating a masked vector through element-wise multiplication of a mask vector and the embedding vector;
calculating a loss using output based on the masked vector;”
See specification ([0045-0046], [0053], [0087-0088]) describing converting. See specification ([0054], [0056], [0058], [0072], [0081]) describing generating. See specification ([0064-0066], [0091]) describing calculating. For these reasons, the claim recites Mathematical Concepts.
Under the Alice Framework Step 2A Prong 2 analysis, the claim recites the combination of the following additional elements: a memory in which at least one program is recorded, a processor for executing the program, wherein the program includes instructions for performing, receiving data configured with indices of discrete entities, and training a model based on the loss. A memory in which at least one program is recorded, a processor for executing the program, and wherein the program includes instructions for performing are recited at a high level of generality, and are examples of generic computing elements, and/or merely generally linked to a particular technological environment (see MPEP 2106.05(h)(vi): Limiting the abstract idea of collecting information, analyzing it, and displaying certain results of the collection analysis to data related to the electric power grid, because limiting application of the abstract idea to power-grid monitoring is simply an attempt to limit the use of the abstract idea to a particular technological environment). The training limitation is an example of mere instructions to apply the exception, merely resulting in the words “apply it” (see MPEP 2106.05(f): Mere Instructions). The receiving limitation is an example of insignificant extra-solution activity, mere data gathering (see MPEP 2106.05(g): Insignificant Extra-Solution Activity). Taken alone or in combination, they fail to integrate the judicial exception into a practical application.
Under the Alice Framework Step 2B Analysis, the additional elements recited above, taken alone or in combination, do not amount to significantly more than the judicial exception. As discussed in the Step 2A Prong 2 Analysis, the claim recites a memory in which at least one program is recorded, a processor for executing the program, and wherein the program includes instructions for performing at a high level of generality, which merely result in “apply it” on a computer. Additionally, the training limitation merely results in “apply it” on a computer. The receiving limitation described above as an insignificant extra-solution activity is also well-understood, routine, or conventional (see MPEP 2106.05(d)(II)(iv): Receiving or transmitting data over a network). Since the claim does not include additional elements that, alone or in combination, amount to significantly more than the judicial exception, claim 9 is ineligible.
Claims 10-14, 16 merely further limits the mathematical concepts. Claims 10-14, 16 do not recite any new additional elements.
Under the Alice Framework Step 2A Prong 1 analysis, claim 15 recites Mathematical Concepts. The claim recites Mathematical Calculations, which is specifically identified as an exemplar in the Mathematical Concepts grouping of abstract ideas:
“calculating the loss comprises calculating a final loss based on a first loss corresponding to a difference between output and a correct answer and on a second loss corresponding to a difference between a target sparsity and a sparsity of a masking vector generated based on a weight vector configured with respective weight values for the discrete entities;”
See specification ([0065-0066], [0091-0093]) describing calculating. For these reasons, the claim recites Mathematical Concepts.
Under the Alice Framework Step 2A Prong 2 analysis, the claim recites the combination of the following additional elements: a neural network model. A neural network model is recited at a high level of generality, and is an example of generic computing elements, and/or merely generally linked to a particular technological environment (see MPEP 2106.05(h)(vi): Limiting the abstract idea of collecting information, analyzing it, and displaying certain results of the collection analysis to data related to the electric power grid, because limiting application of the abstract idea to power-grid monitoring is simply an attempt to limit the use of the abstract idea to a particular technological environment). Taken alone or in combination, they fail to integrate the judicial exception into a practical application.
Under the Alice Framework Step 2B Analysis, the additional elements recited above, taken alone or in combination, do not amount to significantly more than the judicial exception. As discussed in the Step 2A Prong 2 Analysis, the claim recites a neural network model at a high level of generality, which merely result in “apply it” on a computer. Since the claim does not include additional elements that, alone or in combination, amount to significantly more than the judicial exception, claim 15 is ineligible.
Claims 1-8 are directed towards a method that would be performed by the apparatus of claims 9-16. The claims 9-16 analysis equally applies, and claims 1-8 are similarly rejected.
Regarding claim 1, the preamble is given patentable weight. Claim 1 contains the
limitation “the model” in the body, which is referring to the limitations as recited in the
preamble, respectively “a neural network model”. A skilled person in the art reading the claims would consider the claim in view of the body and preamble, and identify them limited to the technological environment of measuring a weight of a discrete entity in a neural network model. The body of the claim depends on the preamble for completeness, and gives life, meaning, and vitality to this claim. Therefore, the preamble of claim 1 should be afforded patentable weight.
Additionally, due to the preamble of claim 1 receiving patentable weight, claim 1 recites “a method for measuring a weight for a discrete entity”, and “performed in a neural network model configured with multiple layers” as additional elements. The performing in a neural network model is recited at a high level of generality, and are examples of generic computing elements, and/or merely generally linked to a particular technological environment (see MPEP 2106.05(h)(vi): Limiting the abstract idea of collecting information, analyzing it, and displaying certain results of the collection analysis to data related to the electric power grid, because limiting application of the abstract idea to power-grid monitoring is simply an attempt to limit the use of the abstract idea to a particular technological environment). The method for measuring limitation is an example of mere instructions to apply the exception, merely resulting in the words “apply it” (see MPEP 2106.05(f): Mere Instructions). Taken alone or in combination, they fail to integrate the judicial exception into a practical application.
Under the Alice Framework Step 2B Analysis, the additional elements recited above, taken alone or in combination, do not amount to significantly more than the judicial exception. As discussed in the Step 2A Prong 2 Analysis, the claim recites performing in a neural network model at a high level of generality, which merely result in “apply it” on a computer. Additionally, the method for measuring limitation merely results in “apply it” on a computer. Since the claim does not include additional elements that, alone or in combination, amount to significantly more than the judicial exception, claim 1 is ineligible.
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)(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.
Apparatus claims 9-16 will be addressed first, followed by method claims 1-8.
Claims 1 and 9 are rejected under 35 U.S.C. 102(a)(2) as being anticipated by US 20220164655 A1 Gomez et al. (hereinafter “Gomez”) .
Regarding claim 9, Gomez teaches an apparatus for measuring a weight of a discrete entity, comprising:
memory (Fig. 9, 903, [0057]) in which at least one program is recorded ([0059] program modules); and
a processor for executing the program (Fig. 9, 901, [0059]), wherein the program includes instructions ([0059] program logic) for performing
receiving data configured with indices of discrete entities (Fig. 8, 810, [0055]; Fig. 3, 301, [0030]; description of ‘discrete entities’ provided with respect to Figs. 1A-B and 2, [0020], [0022] ‘sequence of elements’);
converting the data into embedding vectors corresponding to respective indices through an embedding layer (Fig. 8, 811, [0055]; Fig. 3, 302 & 303, [0030]; description of ‘embedding vectors’ & ‘respective indices’ provided with respect to Figs. 1A-B and 2, [0022] ‘input embeddings’ & ‘positional information’);
generating a masked vector through element-wise multiplication (Fig. 8, 813, [0055]; Fig. 4, outputs from 401-403 in 304 [0031], outputs: sliced query matrix 404, sliced key matrix 405, and sliced value matrix 406 [0032]; Fig. 6A-C, results of Mat Mul, [0037-0040]) of a mask vector (Fig. 4, 401 and 402 mask M1, 403 mask M2, [0051]; Fig. 6A-B, 611, 612, [0036-0038]; Fig. 6C, 631, [0040]) and the embedding vector (Fig. 8, 812, [0055]; Fig. 6A-C, 601, [0033]);
calculating a loss ([0050] error; note: ‘error’ is used in place of ‘loss’ as it similarly represents a parameter of the network based on the training data with respect to an optimization function, the model’s weights are updated to reduce error) using output based on the masked vector (Fig. 7, Training Mask Set 1 and 2, [0051]); and
training a model (Fig. 8, 814, [0055]; Figs. 3 & 4, 304, [0030-0031]) based on the loss ([0050-0051]).
Claim 1 is directed towards a method that would be performed by the apparatus of claim 9. The claim 9 analysis equally applies, and claim 1 is similarly rejected.
Regarding claim 1, the preamble is given patentable weight. Claim 1 contains the
limitation “the model” in the body, which is referring to the limitations as recited in the
preamble, respectively “a neural network model”. A skilled person in the art reading the claims would consider the claim in view of the body and preamble, and identify them limited to the technological environment of measuring a weight of a discrete entity in a neural network model. The body of the claim depends on the preamble for completeness, and gives life, meaning, and vitality to this claim. Therefore, the preamble of claim 1 should be afforded patentable weight.
Additionally, Gomez teaches a method ([0002]) for measuring a weight (Fig. 7, weight updates, [0050-0051]) for a discrete entity ([0020], [0022]), performed in a neural network model configured with multiple layers ([0005]; Fig. 3, [0030]).
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.
Apparatus claims 9-16 will be addressed first, followed by method claims 1-8.
Claims 2-3, 10-11 are rejected under 35 U.S.C. 103 as being unpatentable over Gomez in view of US 6721247 B2 Watanabe (hereinafter “Watanabe”).
Regarding claim 10, in addition to the teachings addressed in the claim 9 analysis, the rejection of claim 9 is incorporated and Gomez teaches the apparatus wherein:
generating the masked vector (see claim 9 mapping) includes
performing a floor operation on a weight value for the discrete entity (Fig. 4, Wq, Wk, Wv, [0031]; Figs. 6A-C, 603, [0034]);
assigning 1 to an index (Figs. 6A-C, 614, 622, 632, [0038-0040]) corresponding to an integer equal to or less than a value resulting from the floor operation; and
assigning 0 to an index (Figs. 6A-C, 614, 622, 632, [0038-0040]) corresponding to an integer greater than the value resulting from the floor operation.
Gomez is silent with disclosing performing a floor operation; assigning 1 to an index corresponding to an integer equal to or less than a value resulting from the floor operation; and assigning 0 to an index corresponding to an integer greater than the value resulting from the floor operation.
Watanabe teaches performing a floor operation (Col. 12, line 7, floor(x)); assigning 1 to an index corresponding to an integer equal to or less than a value resulting from the floor operation (Col. 12, lines 10-11, (x) = 1); and assigning 0 to an index corresponding to an integer greater than the value resulting from the floor operation (Col. 12, lines 8-10, (x) = 0).
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify Gomez’s apparatus for measuring a weight of a discrete entity with Watanabe’s floor operations because they are in the claimed invention’s same field of endeavor of digital arithmetic processing (Col. 10, lines 36-39, 53-58). It would have been obvious to one of ordinary skill in the art to perform the floor operation, as it allows the system to perform calculations on simplified values post-floor operation, reduced to 1’s and 0’s (Col. 12, lines 7-13). Making this modification would be beneficial for Gomez’s apparatus, as Gomez’s post processing circuitry could have reduced computational complexity, thus making it more efficient at processing more data more effectively and having greater use in other applications.
Regarding claim 11, in addition to the teachings addressed in the claim 10 analysis, the rejection of claim 10 is incorporated and Gomez teaches the apparatus wherein:
the mask vector (Fig. 4, 401 and 402 mask M1, 403 mask M2, [0051]; Fig. 6A-B, 611, 612, [0036-0038]; Fig. 6C, 631, [0040]) represents values to be used as 1 (Figs. 6A-C, unchanged indices of 611 and 612 in Fig. 6A-B, 631 in Fig. 6C, [0035-0036]) and represents values not to be used as 0 (Figs. 6A-C, sliced indices of 611 and 612 in Fig. 6A-B, 631 in Fig. 6C, [0035-0036]), among elements of the embedding vector (Fig. 6A-C, 601, [0033]).
Claims 2-3 are directed towards a method that would be performed by the apparatus of claims 10-11. The claims 10-11 analysis equally applies, and claims 2-3 are similarly rejected.
Claims 4-6, 12-14 are rejected under 35 U.S.C. 103 as being unpatentable over Gomez in view of Watanabe in view of US 20200107072 A1 Lomada et al. (hereinafter “Lomada”).
Regarding claim 12, in addition to the teachings addressed in the claim 10 analysis, the rejection of claim 10 is incorporated and Gomez teaches the apparatus wherein:
generating the masked vector (see claim 9 mapping) further includes
adding a value corresponding to a gate function for learning the weight value to each of elements of the masked vector (see claim 10 mapping).
Gomez is silent with disclosing adding a value corresponding to a gate function for learning the weight value.
Watanabe is also silent with disclosing adding a value corresponding to a gate function for learning the weight value.
Gomez in view of Watanabe are silent with disclosing adding a value corresponding to a gate function for learning the weight value.
Lomada teaches adding a value ([0113-0114]
b
f
,
b
i
,
b
o
) corresponding to a gate function (Fig. 4B,
f
t
,
i
t
,
o
t
, [0112-0113]) for learning ([0120] tuning biases based on error loss feedback vector) the weight value.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify Gomez in view of Watanabe’s modified apparatus for measuring a weight of a discrete entity with Lomada’s gate functionality because they are in the claimed invention’s same field of endeavor of neural networks ([0008]). It would have been obvious to one of ordinary skill in the art to implement the gate functions, as it can assist with training of the neural network layers ([0036], [0114]) and as features for machine learning algorithms ([0027], [0037]). Making this modification would be beneficial, as Gomez in view of Watanabe’s modified apparatus could have further support for training and tuning of their neural network model, thus yielding more accurate and correct results by having improved and efficient training.
Regarding claim 13, in addition to the teachings addressed in the claim 12 analysis, the rejection of claim 12 is incorporated and Gomez teaches the weight value (see claim 10 mapping).
Gomez is silent with disclosing the gate function has a value close to 0 such that learning of the weight value is possible, and is a function that is differentiable in a preset section.
Watanabe is silent with disclosing the gate function has a value close to 0 such that learning of the weight value is possible, and is a function that is differentiable in a preset section.
Gomez in view of Watanabe are silent with disclosing the gate function has a value close to 0 such that learning of the weight value is possible, and is a function that is differentiable in a preset section.
Lomada teaches the gate function has a value close to 0 ([0112] gates comprises sigmoid neural network layers, sigmoid neural network layer outputs) such that learning of the weight value is possible ([0112] output indicates how much should pass), and is a function that is differentiable in a preset section (Fig. 4B,
f
t
,
i
t
, and
o
t
are computed from their individually respective
σ
blocks, [0112-0113] differentiable in the sense of being distinguishable).
The motivation to combine provided with respect to claim 12 equally applies.
Regarding claim 14, in addition to the teachings addressed in the claim 13 analysis, the rejection of claim 13 is incorporated.
Gomez is silent with disclosing the gate function is a function of Equation (1) below, and L is a positive integer equal to or greater than 1000.
B
x
=
L
x
-
L
x
L
…(1)
Watanabe is silent with disclosing the gate function is a function of Equation (1) below, and L is a positive integer equal to or greater than 1000.
B
x
=
L
x
-
L
x
L
…(1)
Lomada teaches the gate function ([0112]) is a function of Equation (1) below, and L is a positive integer equal to or greater than 1000.
B
x
=
L
x
-
L
x
L
…(1) ([0113]
f
t
,
i
t
,
o
t
equations (1), (2), (5), respectively).
It is noted that “Lx” and “B(x)” are not defined in equation (1) in claim 13 or any previous claims. Given this lack of definition and it further limiting claim 13, Lomada effectively teaches the gate function and its associated equation. Lomada teaches the gate function and the equation by virtue of teaching “yielding a value close to 0” (Lomada, [0112]), as previously recited in claim 13. It would have been obvious to try the
f
t
,
i
t
,
o
t
equations (1), (2), (5), respectively, given the finite number of equations in practicality to perform the functionality of yielding a value close to 0. Utilizing equations already integrated in the neural network system would achieve predictable results as using them requires operation of the system already containing such formulations and already meets the functionality requirements of yielding a value close to 0.
Claims 4-6 are directed towards a method that would be performed by the apparatus of claims 12-14. The claims 12-14 analysis equally applies, and claims 4-6 are similarly rejected.
Claims 7-8, 15-16 are rejected under 35 U.S.C. 103 as being unpatentable over Gomez in view of Watanabe in view of Lomada in view of US 11887583 B1 Strimel et al. (hereinafter “Strimel”).
Regarding claim 15, in addition to the teachings addressed in the claim 13 analysis, the rejection of claim 13 is incorporated and Gomez teaches the apparatus wherein:
calculating the loss ([0050-0051]); a neural network model ([0005], Fig. 3, [0030]); a masking vector generated based on a weight vector (Fig. 6A, 614, [0035]; Fig. 6B, 622, [0039]; Fig. 6C, 632, [0040]) configured with respective weight values (Figs. 6A-C, 603, [0034], [0041]) for the discrete entities (description provided with respect to Figs. 1A-B and 2, [0020], [0022] ‘sequence of elements’).
Gomez is silent with disclosing calculating the loss comprises calculating a final loss based on a first loss corresponding to a difference between output and a correct answer of a neural network model and on a second loss corresponding to a difference between a target sparsity and a sparsity of a masking vector generated based on a weight vector configured with respective weight values for the discrete entities.
Watanabe is silent with disclosing calculating the loss comprises calculating a final loss based on a first loss corresponding to a difference between output and a correct answer of a neural network model and on a second loss corresponding to a difference between a target sparsity and a sparsity of a masking vector generated based on a weight vector configured with respective weight values for the discrete entities.
Gomez in view of Watanabe are silent with disclosing calculating the loss comprises calculating a final loss based on a first loss corresponding to a difference between output and a correct answer of a neural network model and on a second loss corresponding to a difference between a target sparsity and a sparsity of a masking vector generated based on a weight vector configured with respective weight values for the discrete entities.
Lomada teaches calculating the loss comprises calculating a final loss ([0121] combined error loss feedback vector) based on a first loss ([0121] for each user trait change an amount of error loss due to inaccurate prediction is determined) corresponding to a difference between output ([0121] predicted user trait sequence) and a correct answer ([0121] ground truth sequence) of a neural network model and on a second loss corresponding to a difference between a target sparsity and a sparsity of a masking vector generated based on a weight vector configured with respective weight values for the discrete entities.
The motivation to combine provided with respect to claim 12 equally applies.
Although Lomada teaches a final loss and a first loss, it appears they are silent with disclosing a second loss corresponding to a difference between a target sparsity and a sparsity of a masking vector.
Strimel teaches a second loss (Col. 13, lines 20-30, loss; Col. 3, lines 34-61, updating model) corresponding to a difference between a target sparsity (Col. 13, lines 20-30, target sparsity; Col. 3, lines 34-61) and a sparsity (Col. 3, lines 34-61, a sparse mask) of a masking vector.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify Gomez in view of Watanabe in view of Lomada’s modified apparatus for measuring a weight of a discrete entity with Strimel’s second loss because they are in the claimed invention’s same field of endeavor of neural networks (Col. 2, lines 34-40). It would have been obvious to one of ordinary skill in the art to implement the second loss corresponding with sparsity, as it can assist with training of the neural network producing improved results (Col. 13, lines 27-30; Col. 3, lines 34-39). Making this modification would be beneficial, as Gomez in view of Watanabe in view of Lomada’s modified apparatus could have further support by incorporating an additional parameter for training and tuning of their neural network model, thus yielding more accurate and correct results by having improved and efficient training.
Regarding claim 16, in addition to the teachings addressed in the claim 15 analysis, the rejection of claim 15 is incorporated and Gomez teaches the apparatus wherein:
generating the masked vector (see claim 9 mapping) further includes
compensating for a rapid change in output ([0030], [0032] slicing matrices and ignoring the rest to preserve the regularization effect while reducing computational complexity and memory requirements, compensating in the sense of making up for), which is caused due to application of masking (Fig. 6A, 614, [0035]; Fig. 6B, 622, [0039]; Fig. 6C, 632, [0040]) to the embedding vector (Fig. 8, 812, [0055]; Fig. 6A-C, 601, [0033]).
Claims 7-8 are directed towards a method that would be performed by the apparatus of claims 15-16. The claims 15-16 analysis equally applies, and claims 7-8 are similarly rejected.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Jong-Ryul Lee, Yong-Ju Lee, and Yong-Hyuk Moon. 2021. Block-wise Word Embedding Compression Revisited: Better Weighting and Structuring. In Findings of the Association for Computational Linguistics: EMNLP 2021, pages 4379–4388, Punta Cana, Dominican Republic. Association for Computational Linguistics. (hereinafter “Lee”) is a NPL authored by the named inventors of the instant application. Lee provides a further explanation to the technologies utilized and an extensive discussion of the results gathered.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MARKUS A VILLANUEVA whose telephone number is (703)756-1603. The examiner can normally be reached M - F 8:30 am - 5:30 pm.
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, Andrew Caldwell can be reached at (571) 272-3702. 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.
/MARKUS ANTHONY VILLANUEVA/Examiner, Art Unit 2182
/EMILY E LAROCQUE/Primary Examiner, Art Unit 2182