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 .
Information Disclosure Statement
The information disclosure statements (IDS) submitted on October 26, 2023, and January 10, 2025 were considered by the examiner. The submission is in compliance with the provisions of 37 CFR 1.97.
Claim Interpretation
The following is a quotation of 35 U.S.C. 112(f):
(f) Element in Claim for a Combination. – An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph:
An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked.
As explained in MPEP § 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph:
(A) the claim limitation uses the term “means” or “step” or a term used as a substitute for “means” that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function;
(B) the term “means” or “step” or the generic placeholder is modified by functional language, typically, but not always linked by the transition word “for” (e.g., “means for”) or another linking word or phrase, such as “configured to” or “so that”; and
(C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function.
Use of the word “means” (or “step”) in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function.
Absence of the word “means” (or “step”) in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function.
Claim limitations in this application that use the word “means” (or “step”) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word “means” (or “step”) are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action.
This application includes one or more claim limitations that use the word “means,” which are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. Such claim limitations are: “A processing system, comprising: means for accessing a feature tensor generated based on a model input to a machine learning model; means for generating a sampling matrix based on the model input; means for generating an activation output, using an activation layer of the machine learning model, based on the feature tensor and the sampling matrix; and means for providing the activation output as output from the activation layer of the machine learning model,” in claim 28.
Because these claim limitations are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, they are being interpreted to cover the corresponding structure described in the specification as performing the claimed function, and equivalents thereof.
If applicant does not intend to have these limitation(s) interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitation(s) to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitation(s) recite(s) sufficient structure to perform the claimed function so as to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph.
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 1-30 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea (mental process or math concept) without significantly more.
Claim 1:
Regarding claim 1, in step 1 of the 101-analysis set forth in MPEP 2106, the claim recites “A processing system comprising: one or more memories comprising processor-executable instructions; and one or more processors configured to execute the processor-executable instructions and cause the processing system to: access a first feature tensor generated based on a model input to a machine learning model; generate a sampling matrix based on the model input; generate a first activation output, using a first activation layer of the machine learning model, based on the first feature tensor and the sampling matrix; and provide the first activation output as output from the first activation layer of the machine learning model”, and a system is one of the four statutory categories of invention.
In step 2A prong 1 of the 101-analysis set forth in the MPEP 2106, the examiner has determined that the following limitations recite a process that, under the broadest reasonable interpretation, covers a mathematical concept but for recitation of generic computer components:
generate a sampling matrix based on the model input; (This is considered a mathematical relationship, mathematical formula or equation, or mathematical calculation, see in paragraphs [0027-0028, 0030-31] from the specification, see MPEP 2106.04(a)(2), subsection I),
PNG
media_image1.png
686
1019
media_image1.png
Greyscale
generate a first activation output, using a first activation layer of the machine learning model, based on the first feature tensor and the sampling matrix; (This is considered a mathematical relationship, mathematical formula or equation, or mathematical calculation, see in paragraph [ 0027-0028,0101 ] from the specification, see MPEP 2106.04(a)(2), subsection I),
PNG
media_image2.png
716
991
media_image2.png
Greyscale
PNG
media_image3.png
246
962
media_image3.png
Greyscale
If claim limitations, under their broadest reasonable interpretation, covers performance of the limitations as a mathematical concept but for the recitation of generic computer components, then it falls within the mathematical concept grouping of abstract ideas. Accordingly, the claim “recites” an abstract idea.
In step 2A prong 2 of the 101-analysis set forth in MPEP 2106, the examiner has determined that the following additional elements do not integrate this judicial exception into a practical application:
A processing system comprising: one or more memories comprising processor-executable instructions (Mere instructions to apply an exception using generic computer – see MPEP 2106.05(f)),
… and one or more processors configured to execute the processor-executable instructions and cause the processing system (Mere instructions to apply an exception using generic computer – see MPEP 2106.05(f)),;
to: access a first feature tensor generated based on a model input to a machine learning model; (In step 2A, prong 2, this recites mere data gathering, which is considered an insignificant extra-solution activity – see MPEP 2106.05(g)),
provide the first activation output as output from the first activation layer of the machine learning model (In step 2A, prong 2, this recites mere data outputting, which is considered an insignificant extra-solution activity – see MPEP 2106.05(g)),
Since the claim as a whole, looking at the additional elements individually and in combination, does not contain any other additional elements that are indicative of integration into a practical application, the claim is “directed” to an abstract idea.
In step 2B of the 101-analysis set forth in the 2019 PEG, the examiner has determined that the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
As discussed above, additional elements iii and iv recite mere instructions to apply the judicial exception using generic computer components, which are not indicative of significantly more. The additional elements v and vi, recite mere data gathering or data outputting, and are considered insignificant extra-solution activities. In step 2B, these insignificant extra-solution activities are well understood routine and conventional activities, which include receiving or transmitting data over a network from court case Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information); TLI Communications LLC v. AV Auto. LLC, 823 F.3d 607, 610, 118 USPQ2d 1744, 1745 (Fed. Cir. 2016), – see MPEP 2106.05(d) (II)(i)), as well as from data gathering and outputting, court case Mayo, 566 U.S. at 79, 101 USPQ2d at 1968; and court case OIP Techs., Inc. v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1092-93 (Fed. Cir. 2015) (presenting offers and gathering statistics amounted to mere data gathering).
Considering the additional elements individually and in combination, and the claim as a whole, the additional elements do not provide significantly more than the abstract idea. Therefore, the claim is not patent eligible.
Claim 2:
Regarding claim 2, it is dependent upon claim 1,and thereby incorporates the limitations of, and corresponding analysis applied to claim 1. Further, claim 2 recites the following additional element:
The processing system of claim 1, wherein, to generate the sampling matrix, the one or more processors are configured to execute the processor-executable instructions and cause the processing system to use a set of parameters having values learned during training of the machine learning model, (In step 2A, prong 2, this is considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)). (In step 2B, this is also considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)).
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 3:
Regarding claim 3, it is dependent upon claim 2, and thereby incorporates the limitations of, and corresponding analysis applied to claim 2. Further, claim 3 recites the following additional element:
The processing system of claim 2, wherein the set of parameters corresponds to at least one of (i) an equivariant multilayer perceptron (MLP) or (ii) an equivariant convolutional layer. (In step 2A, prong 2, this is considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)). (In step 2B, this is also considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)).
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 4:
Regarding claim 4, it is dependent upon claim 1,and thereby incorporates the limitations of, and corresponding analysis applied to claim 1.
Further, claim 4 recites the following abstract idea:
The processing system of claim 1, wherein the first feature tensor comprises Fourier coefficients generated by a first layer of the machine learning model, (This is considered a mathematical relationship, mathematical formula or equation, or mathematical calculation, see in paragraphs [0019] state “Pointwise non-linearity (e.g., using an activation function such as a rectified linear unit (ReLU)) can then be applied to the transformed features. A discretized Fourier transform can then be applied to transform the output of the nonlinearity operation back to Fourier coefficients,” and in [0030] stating “In some aspects, the discretized Fourier transform component 135 processes the second intermediate tensor to generate the activation output 140 for the nonlinear block 110. In some aspects, the activation output 140 is defined using Equation 2 below, where {circumflex over (f)}(x) is the activation output 140 and A(x)+ is the pseudoinverse of the matrix A(x). In some aspects, A(x)+=(A(x)TA(x))−1A(x)T.” and later for details in [0034] from the specification, describing Fourier transform involves math, see MPEP 2106.04(a)(2), subsection I),
If claim limitations, under their broadest reasonable interpretation, covers performance of the limitations as a mathematical concept but for the recitation of generic computer components, then it falls within the mathematical concept grouping of abstract ideas. Accordingly, the claim “recites” an abstract idea.
Further, claim 4 recites the following additional element:
and wherein the one or more processors are configured to further execute the processor-executable instructions to cause the processing system to use the sampling matrix to perform an inverse Fourier transform operation on the first feature tensor, (In step 2A, prong 2, this is considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)). (In step 2B, this is also considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)).
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 5:
Regarding claim 5, it is dependent upon claim 1,and thereby incorporates the limitations of, and corresponding analysis applied to claim 1. Further, claim 5 recites the following additional element:
The processing system of claim 1, wherein the sampling matrix comprises a respective column for each respective irreducible representation (irrep) of a set of irreps of a transformation group to which the first activation layer of the machine learning model is equivariant. (In step 2A, prong 2, this is considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)). (In step 2B, this is also considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)).
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 6:
Regarding claim 6, it is dependent upon claim 1,and thereby incorporates the limitations of, and corresponding analysis applied to claim 1.
Further, claim 6 recites the following abstract idea:
…to: generate a second feature tensor (This recites a mental process, a person can mentally evaluate and generate a second feature tensor which is an algebraic object that describes a multilinear relationship between sets of algebraic objects related to a vector space. Tensors can map between different objects such as vectors, scalars, and even other tensors, see MPEP 2106.04(a)(2)(III)),
generate a downsampled sampling matrix (This recites a mental process, a person can mentally evaluate and generate a downsampled (or data with a smaller number of data points) sampling matrix, see MPEP 2106.04(a)(2)(III)),
If claim limitations, under their broadest reasonable interpretation, covers performance of the limitations as a mental process but for the recitation of generic computer components, then it falls within the mental process grouping of abstract ideas. Accordingly, the claim “recites” an abstract idea.
Further, claim 6 recites the following additional element:
The processing system of claim 1, wherein the one or more processors are configured to further execute the processor-executable instructions to cause the processing system… (In step 2A, prong 2, this is considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)). (In step 2B, this is also considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)).
…based on processing the first activation output using a second layer of the machine learning model; (In step 2A, prong 2, this is considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)). (In step 2B, this is also considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)).
based on processing the sampling matrix using a downsampling operation of the machine learning model; (In step 2A, prong 2, this is considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)). (In step 2B, this is also considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)).
and generate a second activation output, using a second activation layer of the machine learning model, based on the second feature tensor and the downsampled sampling matrix. (In step 2A, prong 2, this is considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)). (In step 2B, this is also considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)).
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 7:
Regarding claim 7, it is dependent upon claim 6, and thereby incorporates the limitations of, and corresponding analysis applied to claim 6. Further, claim 7 recites the following additional element:
The processing system of claim 6, wherein the downsampling operation comprises at least one of: (i) a linear interpolation operation or (ii) a convolution operation using one or more learned weights. (In step 2A, prong 2, this is considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)). (In step 2B, this is also considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)).
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 8:
Regarding claim 8, it is dependent upon claim 1, and thereby incorporates the limitations of, and corresponding analysis applied to claim 1. Further, claim 8 recites the following abstract idea:
to generate a respective sampling matrix for each of a plurality of points in the model input and wherein the model input comprises point cloud data, (This recites a mathematical relationship, mathematical formula or equation, or mathematical calculation, see in paragraphs [0027-0028, 0030-31 ] from the specification, see MPEP 2106.04(a)(2), subsection I),
PNG
media_image1.png
686
1019
media_image1.png
Greyscale
Further, claim 8 recites the following additional element:
The processing system of claim 1, wherein the one or more processors are configured to further execute the processor-executable instructions to cause the processing system…, (In step 2A, prong 2, this is considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)). (In step 2B, this is also considered mere instructions to implement an abstract idea using generic computer – see MPEP 2106.05(f)).
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 9:
Regarding claim 9, it is dependent upon claim 1, and thereby incorporates the limitations of, and corresponding analysis applied to claim 1. Further, claim 9 recites the following abstract idea:
i. The processing system of claim 1, wherein the activation output is generated according to
PNG
media_image4.png
72
446
media_image4.png
Greyscale
and wherein: f~(x) is the activation output for the first feature tensor generated based on the model input, x is the model input, n indicates a number of rows in the sampling matrix, A(x) is the sampling matrix generated based on the first feature tensor, and f^(x) is the first feature tensor. (This is considered a mathematical relationship, mathematical formula or equation, or mathematical calculation, see in paragraphs [ 0030-0031 ] stating “In some aspects, the discretized Fourier transform component 135 processes the second intermediate tensor to generate the activation output 140 for the nonlinear block 110. In some aspects, the activation output 140 is defined using Equation 2 below, where {circumflex over (f)}(x) is the activation output 140 and A(x)+ is the pseudoinverse of the matrix A(x). In some aspects, A(x)+=(A(x)TA(x))−1A(x)T… [0031] In some aspects, as computing the inverse of (A(x)TA(x))−1 (to compute A(x)+) may be computationally expensive during a forward pass through the model (as well as being difficult to backpropagate through on a backward pass,” from the specification, see MPEP 2106.04(a)(2), subsection I),
If claim limitations, under their broadest reasonable interpretation, covers performance of the limitations as a mathematical concept but for the recitation of generic computer components, then it falls within the mathematical concept grouping of abstract ideas. Accordingly, the claim “recites” an abstract idea.
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 10:
Regarding claim 10, in step 1 of the 101-analysis set forth in MPEP 2106, the claim recites “A processor-implemented method, comprising: accessing a first feature tensor generated …”, and a method is one of the four statutory categories of invention.
Since claim 10 recites similar limitations as corresponding claim 1 listed above, the claim is rejected for similar reasons under 35 U.S.C. 101.
Claim 19:
Regarding claim 19, in step 1 of the 101-analysis set forth in MPEP 2106, the claim recites
“One or more non-transitory computer-readable media comprising processor-executable instructions that, when executed by one or more processors of a processing system, cause the processing system…”, and a non-transitory computer readable medium is a system and is one of the four statutory categories of invention.
In step 2A prong 1 of the 101-analysis set forth in the MPEP 2106, the examiner has determined that the following limitations recite a process that, under the broadest reasonable interpretation, covers a mathematical concept but for recitation of generic computer components:
generate a sampling matrix based on the model input; (This is considered a mathematical relationship, mathematical formula or equation, or mathematical calculation, see in paragraphs [0027-0028, 0030-31] from the specification, see MPEP 2106.04(a)(2), subsection I),
PNG
media_image1.png
686
1019
media_image1.png
Greyscale
generate a first activation output, using a first activation layer of the machine learning model, based on the first feature tensor and the sampling matrix; (This is considered a mathematical relationship, mathematical formula or equation, or mathematical calculation, see in paragraph [ 0027-0028,0101 ] from the specification, see MPEP 2106.04(a)(2), subsection I),
PNG
media_image2.png
716
991
media_image2.png
Greyscale
PNG
media_image3.png
246
962
media_image3.png
Greyscale
If claim limitations, under their broadest reasonable interpretation, covers performance of the limitations as a mathematical concept but for the recitation of generic computer components, then it falls within the mathematical concept grouping of abstract ideas. Accordingly, the claim “recites” an abstract idea.
In step 2A prong 2 of the 101-analysis set forth in MPEP 2106, the examiner has determined that the following additional elements do not integrate this judicial exception into a practical application:
One or more non-transitory computer-readable media comprising processor-executable instructions that, when executed by one or more processors of a processing system, cause the processing system … (In step 2A, prong 2, this is considered a generic computer component being used as a tool. – see MPEP 2106.05(f)),
to: access a first feature tensor generated based on a model input to a machine learning model; (In step 2A, prong 2, this recites mere data gathering, which is considered an insignificant extra-solution activity – see MPEP 2106.05(g)),
provide the first activation output as output from the first activation layer of the machine learning model (In step 2A, prong 2, this recites mere data outputting, which is considered an insignificant extra-solution activity – see MPEP 2106.05(g)),
Since the claim as a whole, looking at the additional elements individually and in combination, does not contain any other additional elements that are indicative of integration into a practical application, the claim is “directed” to an abstract idea.
In step 2B of the 101-analysis set forth in the 2019 PEG, the examiner has determined that the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
As discussed above, additional element iii recites mere instructions to apply the judicial exception using generic computer components, which is not indicative of significantly more. The additional elements iv and v, recite mere data gathering or data outputting, and are considered insignificant extra-solution activities. In step 2B, these insignificant extra-solution activities are well understood routine and conventional activities, which include receiving or transmitting data over a network from court case Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information); TLI Communications LLC v. AV Auto. LLC, 823 F.3d 607, 610, 118 USPQ2d 1744, 1745 (Fed. Cir. 2016), – see MPEP 2106.05(d) (II)(i)), as well as from data gathering and outputting, court case Mayo, 566 U.S. at 79, 101 USPQ2d at 1968; and court case OIP Techs., Inc. v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1092-93 (Fed. Cir. 2015) (presenting offers and gathering statistics amounted to mere data gathering).
Considering the additional elements individually and in combination, and the claim as a whole, the additional elements do not provide significantly more than the abstract idea. Therefore, the claim is not patent eligible.
Claims 11-18, and 20-27:
All of claim 10’s dependent claims 11-18 follow the deficiencies of their parent claim. Claims 11-18, recite similar limitations as corresponding claims 2-9, and are rejected for similar reasons under 35 U.S.C. 101.
All of claim 19’s dependent claims 20-27 follow the deficiencies of their parent claim. Claims 20-27, recite similar limitations as corresponding claims 2-9, and are rejected for similar reasons under 35 U.S.C. 101.
Claim 28:
Regarding claim 28, in step 1 of the 101-analysis set forth in MPEP 2106, the claim recites
“A processing system, comprising … “, and a system is one of the four statutory categories of invention.
In step 2A prong 1 of the 101-analysis set forth in the MPEP 2106, the examiner has determined that the following limitations recite a process that, under the broadest reasonable interpretation, covers a mathematical concept but for recitation of generic computer components:
… generating a sampling matrix based on the model input; (This is considered a mathematical relationship, mathematical formula or equation, or mathematical calculation, see in paragraphs [0027-0028, 0030-31] from the specification, see MPEP 2106.04(a)(2), subsection I),
PNG
media_image1.png
686
1019
media_image1.png
Greyscale
… generating an activation output, using an activation layer of the machine learning model, based on the feature tensor and the sampling matrix; (This is considered a mathematical relationship, mathematical formula or equation, or mathematical calculation, see in paragraph [ 0027-0028,0101 ] from the specification, see MPEP 2106.04(a)(2), subsection I),
PNG
media_image2.png
716
991
media_image2.png
Greyscale
PNG
media_image3.png
246
962
media_image3.png
Greyscale
If claim limitations, under their broadest reasonable interpretation, covers performance of the limitations as a mathematical concept but for the recitation of generic computer components, then it falls within the mathematical concept grouping of abstract ideas. Accordingly, the claim “recites” an abstract idea.
In step 2A prong 2 of the 101-analysis set forth in MPEP 2106, the examiner has determined that the following additional elements do not integrate this judicial exception into a practical application:
A processing system, comprising: … (Mere instructions to apply an exception using generic computer – see MPEP 2106.05(f)),
… accessing a feature tensor generated based on a model input to a machine learning model; (In step 2A, prong 2, this recites mere data gathering, which is considered an insignificant extra-solution activity – see MPEP 2106.05(g)),
and … providing the activation output as output from the activation layer of the machine learning model, (In step 2A, prong 2, this recites mere data outputting, which is considered an insignificant extra-solution activity – see MPEP 2106.05(g)),
means for accessing a feature tensor… (Mere instructions to apply an exception using generic computer – see MPEP 2106.05(f)),
means for generating a sampling matrix…(Mere instructions to apply an exception using generic computer – see MPEP 2106.05(f)),
means for generating an activation output …(Mere instructions to apply an exception using generic computer – see MPEP 2106.05(f)),
and means for providing the activation output… (Mere instructions to apply an exception using generic computer – see MPEP 2106.05(f)),
Since the claim as a whole, looking at the additional elements individually and in combination, does not contain any other additional elements that are indicative of integration into a practical application, the claim is “directed” to an abstract idea.
In step 2B of the 101-analysis set forth in the 2019 PEG, the examiner has determined that the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
As discussed above, additional elements iii, vi, vii, viii and ix recite mere instructions to apply the judicial exception using generic computer components, which are not indicative of significantly more. The additional elements iv and v, recite mere data gathering or data outputting, and are considered insignificant extra-solution activities. In step 2B, these insignificant extra-solution activities are well understood routine and conventional activities, which include receiving or transmitting data over a network from court case Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information); TLI Communications LLC v. AV Auto. LLC, 823 F.3d 607, 610, 118 USPQ2d 1744, 1745 (Fed. Cir. 2016), – see MPEP 2106.05(d) (II)(i)), as well as from data gathering and outputting, court case Mayo, 566 U.S. at 79, 101 USPQ2d at 1968; and court case OIP Techs., Inc. v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1092-93 (Fed. Cir. 2015) (presenting offers and gathering statistics amounted to mere data gathering).
Considering the additional elements individually and in combination, and the claim as a whole, the additional elements do not provide significantly more than the abstract idea. Therefore, the claim is not patent eligible.
Claims 29, 30:
All of claim 28’s dependent claims follow the deficiencies of their parent claim. Since claims 29, and 30 recite similar limitations as corresponding claims 2, and 9 respectively listed above, and are rejected for similar reasons under 35 U.S.C. 101.
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.
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.
Claims 1, 2, 6, 7, 10, 11, 15, 16, 19, 20, 24, 25, 28, and 29 are rejected under 35 U.S.C. 103 over Koyuncu, A. et al., (Pub. WO2023066473A1), published on April 27, 2023, (hereafter, Koyuncu), in view of Lee, H. et al., (Pub. No. WO2014205231A1), published on December 24, 2014, (hereafter, Lee), further in view of Langhammer, M. et al., (US PG Pub. No. US20230325665A1 ), published on October 12, 2023, (hereafter, Langhammer).
Claim 1:
Regarding claim 1, Koyuncu teaches “A processing system comprising: one or more memories comprising processor-executable instructions; and one or more processors configured to execute the processor-executable instructions and cause the processing system…”
See Koyuncu in page 10, lines 1-7 describe "In an exemplary implementation, a computer program stored on a non-transitory medium and including code instructions, which, when executed on one or more processors, cause the one or more processors to execute steps of the method according to any of the methods described above. According to an embodiment, an apparatus is provided for entropy encoding of a latent tensor, comprising: processing circuitry configured to: separate the latent tensor into a plurality of segments in the spatial dimensions". Also, see Koyuncu in page 45, lines 7-8 describe “In other examples, data is retrieved from a local memory, streamed over a network, or the like.” Here, Koyuncu describes a system that contains processors and a memory.
Further, Koyuncu teaches “access a first feature tensor generated based on a model input to a machine learning model;”
See Koyuncu in page 25, lines 23 - 33 describe "In the transformer model, an input tensor x is first fed to a neural network layer in order to extract features of the input tensor. Thereby a so-called embedding tensor e 5010 is obtained, which includes the latent space elements that are used as an input to a transformer. The input tensor x and embedding tensor e have the size of S x d input , and S x de , where S is the number of sequential elements and d is the dimensionality of the each sequential element. Positional encodings 5020 may be added to the embedding tensors. The positional encodings 5020 enable a transformer to take into account a sequential order of the input sequence. Such a positional encoding provides a representation of the position of an element within an arrangement of the elements of the input tensor." Here, Koyuncu mentions using a model input x of a transformer model, to extract features (i.e. first feature tensor) of that input tensor.
However, Koyuncu did not explicitly teach “generate a sampling matrix based on the model input; generate a first activation output, using a first activation layer of the machine learning model, based on the first feature tensor and the sampling matrix; and provide the first activation output as output from the first activation layer of the machine learning model.”
In an analogous field, Lee teaches “generate a sampling matrix based on the model input;”
See Lee in [0075] describe “The projection matrix {prs} is a sparse, non-negative matrix of dimension R2 x S. Note that the projection matrix is specific to each image since it depends on the structure of the superpixel graph. Due to the deterministic connection, the pooling layer is actually a virtual layer that only exists to map between the superpixel nodes and the hidden nodes. The GLOC model can also be viewed as having a set of grid-structured nodes that performs average pooling over the adjacent superpixel nodes.” The examiner construes sampling matrix to be any matrix that maps unstructured points (like superpixels) to a structured form (like a grid) for image data. Here, Lee mentions the projection matrix processes each image, and helps perform processing of data.
Further, see Lee in [00139] describe “Max-pooling refers to operations where a local neighborhood (e.g., 2 x 2 grid) of feature detection outputs is shrunk to a pooling node by computing the maximum of the local neighbors. Max-pooling makes the feature representation more invariant to local translations in the input data, and has been shown to be useful in computer vision.” Here, Lee elaborates that this method is applied for input image data.
Later, see Lee in [0091] First, the superpixels and features were generated for each image and then ran the GLOC model to get label guesses for each superpixel, and finally mapped back to pixels for evaluation”. Lee shows that each image generates superpixels and features, and later let a projection matrix processes each image input from [0075].
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the base reference of Koyuncu with the teachings of Lee by using Koyuncu’s teachings of accessing a feature tensor created from model input, and incorporate with Lee’s teaching of generating a sampling matrix from model input.
One of ordinary skill in the art would be motivated to do so because by integrating Lee’s framework into the methods of Koyuncu would bring “ The object centered cropped images brought improvement in classification accuracies, such as 74.9% to 76.8% with RBM, and 77.8% to 78.9% with CRBM using 30 training images per class, respectively...This suggests that the classification performance can be improved by localizing the object better than simply cropping the center region,” (see Lee in [0066]).
However, Koyuncu in view of Lee did not teach “generate a first activation output, using a first activation layer of the machine learning model, based on the first feature tensor and the sampling matrix; and provide the first activation output as output from the first activation layer of the machine learning model.”
In an analogous art, Langhammer teaches “generate a first activation output, using a first activation layer of the machine learning model, based on the first feature tensor and the sampling matrix;”
See Langhammer in [0052] describe “as a part of the convolution, MAC operations can be performed on a 3×3×3 subtensor 215 (which is highlighted with dot patterns in FIG. 2 ) in the input tensor 210 and each filter 220. The result of the MAC operations on the subtensor 215 and one filter 220 is an output activation. In some embodiments (e.g., embodiments where the convolution is an integral convolution), an output activation may include 8 bits, e.g., one byte.” Here, Langhammer shows that the result of a subtensor (i.e. feature tensor ) and filter (i.e. sampling matrix) is an output activation.
Also, See Langhammer in [0041] describe “The convolutional layer 110 may receive several images as input and calculate the convolution of each of them with each of the kernels. This process can be repeated several times. For instance, the OFM 160 is passed to the subsequent convolutional layer 110 (i.e., the convolutional layer 110 following the convolutional layer 110 generating the OFM 160 in the sequence). The subsequent convolutional layers 110 perform a convolution on the OFM 160 with new kernels and generates a new feature map.” Langhammer describes information regarding features is transferred onto an initial convolutional layer and receive several images or data as input. Further, see Langhammer in [0057] describe “For instance, the local memory 410 may store the input tensor, convolutional kernels, or output tensor of a convolution in a convolutional layer of a DNN, e.g., the convolutional layer 30. The output tensor can be transmitted from a local memory of a compute block 330 to the local memory 410 through the DMA engine 320.” Later, see Langhammer in [0051] describe “ An activation in the output tensor 230 is a data point in the output tensor 230. The output tensor 230 has a spatial size Hout × Wout × Cout, where Hout is the height of the 3D matrix (i.e., the length along the Y axis, which indicates the number of output activations in a column in the 2D matrix of each output channel), Wout is the width of the 3D matrix (i.e., the length along the X axis, which indicates the number of output activations in a row in the 2D matrix of each output channel), and Cout is the depth of the 3D matrix.” Here, Langhammer shows activation in output tensor, where its information is conveyed within a layer (this is viewed as an activation layer).
Further, Langhammer teaches “and provide the first activation output as output from the first activation layer of the machine learning model.”
See Langhammer in [0020] describe “Example tensors include a vector, which is a one-dimensional tensor, and a matrix, which is a two-dimensional tensor. There can also be three-dimensional tensors and even higher dimensional tensors. A DNN layer may have an input tensor (also referred to as “input feature map (IFM)”) including one or more input activations (also referred to as “input elements” or “activations”), a weight tensor including one or more weights, and an output tensor (also referred to as “output feature map (OFM)”) including one or more output activations (also referred to as “output elements” or “activations”).” Here, Langhammer shows providing an output activation from a layer of the deep neural network machine learning model .
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the references of Koyuncu and Lee, and incorporate with the teachings of Langhammer by using the teachings of Koyuncu and Lee of accessing a feature tensor created from model input and creating a sampling matrix, and incorporate with Langhammer’s teaching of generating an activation output.
One of ordinary skill in the art would be motivated to do so because by integrating Langhammer’s framework into the methods of Koyuncu and Lee would bring “by reducing the number of multiplication operations from eight to two, the MAC operation in the PE 500 is accelerated. As a DNN accelerator usually performs a large number of MAC operations in the execution of a DNN, the sparsity acceleration can significantly improve the efficiency and performance of the DNN accelerator,” (see Langhammer in [0093]).
Claim 2:
Regarding claim 2, Koyuncu in view of Lee, further in view of Langhammer, teach the limitations of claim 1.
Further, Koyuncu teaches “2. The processing system of claim 1, wherein, to generate the sampling matrix, the one or more processors are configured to execute the processor-executable instructions and cause the processing system to use a set of parameters having values learned during training of the machine learning model.”
See Koyuncu in page 27, lines 7-12 describe "In other words, an attention layer obtains a plurality of representations of an input sequence, for example the Keys, Queries and Values. To obtain a representation out of said plurality representations, the input sequence is processed by a respective set of weights. This set of weights may be obtained in a training phase. These set of weights may be learned jointly with the remaining parts of a neural network including such an attention layer. During inference, the output is computed as the weighted sum of the processed input sequence." The examiner construes sampling matrix to be any information, that is represented in a format that relates to processing of image data. Here, KOYUNCU shows that to obtain a representation (i.e. sampling matrix) the system takes a set of weights (i.e. parameters) obtained during training of a neural network model.
Claim 6:
Regarding claim 6, Koyuncu in view of Lee, further in view of Langhammer, teach the limitations of claim 1.
Further, Koyuncu teaches “generate a downsampled sampling matrix based on processing the sampling matrix using a downsampling operation of the machine learning model;”
See Koyuncu in page 20, lines 18-22 describe "the compression in the encoder 310 may be achieved, e.g. by applying a neural network, or in general any processing network with one or more layers. In such network, the compression may be performed by cascaded processing including downsampling, which reduces size and/or number of channels of the input. Thus, the encoder may be referred to, e.g. as a neural network (NN) based encoder," and also see Koyuncu in page 28, lines 19-25 mention "Image data to be compressed may be represented as a three-dimensional tensor 311 with the size of H x W x C where H and W are the height and width of the image and C is the number of color channels. The input image may be processed by an autoencoding convolutional neural network 310 as explained above with reference to Fig. 3a. Such an autoencoder 310 downsamples the input image by applying multiple convolutions and non-linear transformations," Here, Koyuncu mentions creating a downsampled input image data that is “compressed and may be represented as a three-dimensional tensor 311 with the size of H x W x C” using the image operation performed by an autoencoding convolutional neural network described in figure 3a.
PNG
media_image5.png
830
1284
media_image5.png
Greyscale
Further, Langhammer teaches “generate a second feature tensor based on processing the first activation output using a second layer of the machine learning model;”
See Langhammer in [0041] describe "The OFM 160 is then passed to the next layer in the sequence. In some embodiments, the OFM 160 is passed through an activation function. An example activation function is Reu. Elu is a calculation that returns the value provided as input directly, or the value zero if the input is zero or less. The convolutional layer 110 may receive several images as input and calculate the convolution of each of them with each of the kernels. This process can be repeated several times. For instance, the OFM 160 is passed to the subsequent convolutional layer 110 (i.e., the convolutional layer 110 following the convolutional layer 110 generating the OFM 160 in the sequence). The subsequent convolutional layers 110 perform a convolution on the OFM 160 with new kernels and generates a new feature map. The new feature map may also be normalized and resized. The new feature map can be kernelled again by a further subsequent convolutional layer 110, and so on.” Here, when Langhammer mentions that “OFM 160 is passed through an activation function… Elu is a calculation that returns the value provided as input”, this shows this initial or first activation output calculation that returns the value is later used by subsequent layers (which includes second layer). Langhammer shows that the information from the output feature map or OFM 160 is passed to the next convolutional layer in the network sequence. This next layer treats the OFM 160 as its Input Feature Map (IFM). Then, the subsequent layer slides new kernels across this OFM 160 to extract deeper, more complex spatial features, generating a new, second feature tensor or map.
Further, Langhammer teaches “generate a second activation output, using a second activation layer of the machine learning model, based on the second feature tensor and the downsampled sampling matrix.”
See Langhammer in [0148] describe “The multipliers 1230 may perform multiple rounds of multiplication operations. A multiplier 1230 may use the same weight operand but different activation operands in different rounds. For instance, the multiplier 1230 performs a sequence of multiplication operations on a first activation operand stored in a first input register file in a first round, versus a second activation operand stored in a second input register file in a second round. In the second round, a different multiplier 1230 may use the first activation operand and a different weight operand to perform another sequence of multiplication operations. That way, the first activation operand is reused in the second round. The first activation operand may be further reused in additional rounds, e.g., by additional multipliers 1230.” Here, Langhammer describes a multiplier, which is a physical chip processor that performs math calculations for the layer, and the activation operand (i.e. activation output) contains data that flows into the multiplier. Langhammer also uses a second activation operand, representing second activation layer, and generates second output in multiple rounds. Overall, Langhammer explains the efficiency strategy used by hardware multipliers to perform the multiple, iterative calculations required to generate those activation outputs.
Also, see Langhammer in [0041] describe “The subsequent convolutional layers 110 perform a convolution on the OFM 160 with new kernels and generates a new feature map. The new feature map may also be normalized and resized. The new feature map can be kernelled again by a further subsequent convolutional layer 110, and so on.” Here, Langhammer shows each feature map kernelled (i.e. second sampling matrix) by subsequent convolutional layer.
Further, Langhammer mentions in [0043] that “The pooling layers 120 down-sample feature maps generated by the convolutional layers, e.g., by summarizing the presence of features in the patches of the feature maps.” Langhammer shows using the down-sampled summaries of presence features in feature maps (i.e. sampling matrix), and later in [0046] mentions “In some embodiments, the fully connected layers 130 classify the input image 105 and return an operand of size N, where N is the number of classes in the image classification problem. In the embodiments of FIG. 1 , N equals 3, as there are 3 objects 115, 125, and 135 in the input image. Each element of the operand indicates the probability for the input image 105 to belong to a class. To calculate the probabilities, the fully connected layers 130 multiply each input element by weight, make the sum, and then apply an activation function (e.g., logistic if N=2, softmax if N>2). This is equivalent to multiplying the input operand by the matrix containing the weights.” Langhammer here mentions using the layers to return an operand of size N, and later using a matrix to calculate along with this operand. Langhammer overall shows the strategy used by hardware multipliers to perform the multiple, iterative calculations required to generate those activation outputs.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the references of Koyuncu and Lee, and incorporate with the teachings of Langhammer by using the teachings of Koyuncu and Lee of accessing a feature tensor created from model input and creating a sampling matrix, and incorporate with Langhammer’s teaching of generating a second activation output.
One of ordinary skill in the art would be motivated to do so because by integrating Langhammer’s framework into the methods of Koyuncu and Lee would bring “by reducing the number of multiplication operations from eight to two, the MAC operation in the PE 500 is accelerated. As a DNN accelerator usually performs a large number of MAC operations in the execution of a DNN, the sparsity acceleration can significantly improve the efficiency and performance of the DNN accelerator,” (see Langhammer in [0093]).
Claim 7:
Regarding claim 7, Koyuncu in view of Lee, further in view of Langhammer, teach the limitations of claim 1. Further, Koyuncu teaches “7. The processing system of claim 6, wherein the downsampling operation comprises at least one of: (i) a linear interpolation operation or (ii) a convolution operation using one or more learned weights.”
See Koyuncu in page 17, lines 13-18 describe "The factorized entropy model 342 works as a codebook whose parameters are available on the decoder side. An entropy decoder 343 recovers the quantized hyper-latent tensor from the bitstream 341 by using the factorized entropy model 342. The recovered quantized hyper- latent tensor is up-sampled in the hyper-decoder 350 by applying multiple convolution operations and non-linear transformations." and in "The name “convolutional neural network” (CNN) indicates that the network employs a mathematical operation called convolution. Convolution is a specialized kind of linear operation. Convolutional networks are neural networks that use convolution in place of a general matrix multiplication in at least one of their layers."
Also, see Koyuncu mention in page 27, lines 8-12 describe learned weights "To obtain a representation out of said plurality representations, the input sequence is processed by a respective set of weights. This set of weights may be obtained in a training phase. These set of weights may be learned jointly with the remaining parts of a neural network including such an attention layer. During inference, the output is computed as the weighted sum of the processed input sequence." Here, KOYUNCU shows that the weights are learned for the calculations of the model.
Further, see Koyuncu in page 22, lines 5-8, describe "The name "convolutional neural network" (CNN) indicates that the network employs a mathematical operation called convolution. Convolution is a specialized kind of linear operation. Convolutional networks are neural networks that use convolution in place of a general matrix multiplication in at least one of their layers." Koyuncu describes using a convolution operation in page 22 that includes learned weights of model from page 27, and elaborates convolution is a specialized type of linear operation.
Claim 10:
Regarding claim 10, Koyuncu further teaches “A processor-implemented method…”
See Koyuncu in page 10, lines 1-7 describe in "In an exemplary implementation, a computer program stored on a non-transitory medium and including code instructions, which, when executed on one or more processors, cause the one or more processors to execute steps of the method according to any of the methods described above. According to an embodiment, an apparatus is provided for entropy encoding of a latent tensor, comprising: processing circuitry configured to: separate the latent tensor into a plurality of segments in the spatial dimensions". Here, Koyuncu describes using a computer program stored on a non-transitory medium run by processing circuitry and processors to implement a method and system.
Regarding claim 10, the claim recites similar limitations as corresponding independent claim 1 and is rejected for similar reasons as claim 1 using similar teachings and rationale.
Claim 19:
Regarding claim 19, Koyuncu further teaches “One or more non-transitory computer-readable media comprising processor-executable instructions that, when executed by one or more processors of a processing system…”
See Koyuncu in page 10, lines 1-7 describe in "In an exemplary implementation, a computer program stored on a non-transitory medium and including code instructions, which, when executed on one or more processors, cause the one or more processors to execute steps of the method according to any of the methods described above. According to an embodiment, an apparatus is provided for entropy encoding of a latent tensor, comprising: processing circuitry configured to: separate the latent tensor into a plurality of segments in the spatial dimensions". Here, Koyuncu describes using a computer program stored on a non-transitory medium run by processing circuitry and processors to implement a method and system.
Regarding claim 19, the claim recites similar limitations as corresponding independent claim 1 and is rejected for similar reasons as claim 1 using similar teachings and rationale.
Claim 28:
Regarding claim 28, where Koyuncu teaches “means for generating a sampling matrix based on the model input,”
See Koyuncu mention in page 44, lines 17-23 “The encoder 20 may be implemented via processing circuitry 46 to embody the various modules as discussed with respect to encoder of Fig. 3b and/or any other encoder system or subsystem described herein. The decoder 30 may be implemented via processing circuitry 46 to embody the various modules as discussed with respect to decoder of Fig. 3c and/or any other decoder system or subsystem described herein. The processing circuitry may be configured to perform the various operations as discussed later.” Later, see Koyuncu in page 3, lines 9-10 describe “Processing the input of the neural network to extract the features of the plurality of segments may enable a focus of the attention layer on independent deep features of the input.” Koyuncu shows that the system (such as the encoder with processing circuitry) includes processors or machinery that evaluates a model to extract features of segments, which relates to gathering data from input. See Koyuncu in page 30, lines 20-25 for more details on describing a matrix “The mask may be described with a S x S matrix, where its lower triangle (including the diagonal) contains 1 and the upper triangle part (excluding the diagonal) consists of minus infinity (softmax(-∞) = 0). The masked attention may be formulated as:
PNG
media_image6.png
238
773
media_image6.png
Greyscale
”. From the specification in [0054] stating “At block 410, the machine learning system generates a sampling matrix based on the input tensor,” the corresponding structure relates to any system part of a computer.
Further, Koyuncu teaches “means for generating an activation output, …”
See Koyuncu in page 3, lines 27-28 through page 4, lines 1-3, describe “In an exemplary implementation, the neural network includes a second neural subnetwork, the second neural subnetwork processing an output of the attention layer. The neural subnetwork may process the features outputted by the attention layer to provide probabilities for the symbols used in the encoding and thus enabling an efficient encoding and/or decoding.” Further, see Koyuncu in page 13, lines 10-12 describe “For instance, it is understood that a disclosure in connection with a described method may also hold true for a corresponding device or system configured to perform the method and vice versa.” The system is the corresponding structure that helps generate an activation output. See specification in [0055] describe “At block 415, the machine learning system computes an activation output (e.g., the activation output 140 of FIG. 1 ) for the nonlinear block based on the input tensor and the sampling matrix,” which shows any system that runs machine learning can be a corresponding structure.
Further, Koyuncu teaches “and means for providing the activation output as output from the activation layer of the machine learning model.”
See Koyuncu in page 23, lines 15-21, describe “As a result, the network learns filters that activate when it detects some specific type of feature at some spatial position in the input. Stacking the activation maps for all filters along the depth dimension forms the full output volume of the convolution layer. Every entry in the output volume can thus also be interpreted as an output of a neuron that looks at a small region in the input and shares parameters with neurons in the same activation map. A feature map, or activation map, is the output activations for a given filter.” Further, see Koyuncu in page 13, lines 10-12 describe “For instance, it is understood that a disclosure in connection with a described method may also hold true for a corresponding device or system configured to perform the method and vice versa.” The system is the corresponding structure that helps provide an activation output as output. See specification in [0055] describe “At block 415, the machine learning system computes an activation output (e.g., the activation output 140 of FIG. 1 ) for the nonlinear block based on the input tensor and the sampling matrix,” which shows any system that runs machine learning can be a corresponding structure.
However, Koyuncu did not teach “ means for generating a sampling matrix…” or “means for accessing a feature tensor generated based on a model input to a machine learning model”
In an analogous art, Lee teaches “means for generating a sampling matrix…”
See Lee in [0075] describe “The projection matrix {prs} is a sparse, non-negative matrix of dimension R2 x S. Note that the projection matrix is specific to each image since it depends on the structure of the superpixel graph. Due to the deterministic connection, the pooling layer is actually a virtual layer that only exists to map between the superpixel nodes and the hidden nodes. The GLOC model can also be viewed as having a set of grid-structured nodes that performs average pooling over the adjacent superpixel nodes.” The examiner construes sampling matrix to be any matrix that maps unstructured points (like superpixels) to a structured form (like a grid) for image data. Here, Lee mentions the projection matrix processes each image, and helps perform processing of data to create a projection matrix (i.e. sampling matrix).
Further, Lee teaches “means for accessing a feature tensor generated based on a model input to a machine learning model”
See Lee in [00188] describe “In recent years, unsupervised feature learning algorithms have emerged as promising tools for learning representations from data. In particular it is an important problem to learn invariant representations that are robust to variability in high-dimensional data (e.g., images, speech, etc.) since they will enable machine learning systems to achieve good generalization performance while using a small number of labeled training examples.” Also, see Lee in [0006] mention for details “ In another aspect, an automated technique is provided for identifying features in an image. This technique employs a feature recognition model that combines an energy function for a restricted Boltzmann machine with an energy function for conditional random fields. The method includes: receiving data for an image captured by an imaging device; segmenting pixels of the image data into two or more regions using the feature recognition model,” Here, Lee mentions that a machine learning system helps to access a feature tensor created from model input. The specification in [0053] states “At block 405, the machine learning system accesses an input tensor. In some aspects, the input tensor is a feature tensor used as input to a nonlinear operation or layer (e.g., the nonlinear block 110 of FIGS. 1 and/or 3 ) in a CNN,” showing that any machine learning system is a corresponding structure that accesses a feature tensor from model input.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the base reference of Koyuncu with the teachings of Lee by using Koyuncu’s teachings of accessing a feature tensor created from model input, and incorporate with Lee’s teaching of generating a sampling matrix from model input.
One of ordinary skill in the art would be motivated to do so because by integrating Lee’s framework into the methods of Koyuncu would bring “ The object centered cropped images brought improvement in classification accuracies, such as 74.9% to 76.8% with RBM, and 77.8% to 78.9% with CRBM using 30 training images per class, respectively...This suggests that the classification performance can be improved by localizing the object better than simply cropping the center region,” (see Lee in [0066]).
Regarding claim 28, the claim recites similar additional limitations as corresponding independent claim 1 and is rejected for similar reasons as claim 1 using similar teachings and rationale.
Claims 11, 15, 16, 20, 24, 25, 29:
Regarding claims 11, 20, and 29; claims 15 and 24; and claims 16 and 25; they recite similar limitations as corresponding claims 2; 6; and 7, respectively, and are rejected for similar reasons using similar teachings and rationale.
Claims 3, 5, 12, 14, 21, and 23 are rejected under 35 U.S.C. 103 over Koyuncu in view of Lee, further in view of Langhammer, and further in view of Kondor, I. et al., (Pub. No. WO2019246397A1), published on December 26, 2019, (hereafter, Kondor_I).
Claim 3:
Regarding claim 3, Koyuncu in view of Lee, further in view of Langhammer, teach the limitations of claim 1. However, Koyuncu in view of Lee, further in view of Langhammer, did not teach “3. The processing system of claim 2, wherein the set of parameters corresponds to at least one of (i) an equivariant multilayer perceptron (MLP) or (ii) an equivariant convolutional layer.”
In an analogous field, Kondor_I teach “3. The processing system of claim 2, wherein the set of parameters corresponds to at least one of (i) an equivariant multilayer perceptron (MLP) or (ii) an equivariant convolutional layer.”
See Kondor_I in [0093] describe "this gives a sparse tensor representation of dimension N * T * 2b * 2b for every molecule. ..[0093] The CG spherical CNN architecture described herein has the same parameters and hyperparameters as described above, except that v = 15 for all layers, increasing the number of parameters to 1.1 M... The feature vector for each atom is projected onto 150 dimensions using a MLP. These embeddings are summed over atoms, and then the regression target is trained using another MLP having 50 hidden units. Both of these MLPs are jointly trained. The final results are presented below, which show that the present CG-net method outperforms the Spherical CNN of Cohen et al. The only method that appears to provide better performance is a MLP trained on randomly permuted Coulomb matrices " Here, Kondor_I show using MLP for projections of feature vectors.
See Kondor_I in [0097] for details “[0097] Figure 5 is a flow chart of an example method 500, according to example embodiments. Specifically, example method 500 may be used for computationally processing data with a multi-layer convolutional neural network (CNN) implemented in the computing device and having an input layer, an output layer, and one or more intermediate layers.” And [0098] for details “At step 502, the computing device may receive digital image data corresponding to input data that are represented in a form of evaluations of one or more continuous functions on a sphere.”
Also, see Kondor_I in [0110] describe "[0110] An SO(3)-equivariant neural network architecture for spherical data that operates completely in Fourier space has been presented herein. In accordance with example embodiments, this approach circumvents a major drawback of earlier models that need to switch back and forth between Fourier space and “real” space. This achieved by a novel and unconventionally approach that uses the Clebsch-Gordan decomposition as the only source of nonlinearity. While the specific focus is on spheres and SO(3)-equivariance, the approach is more widely applicable, suggesting a general formalism for designing folly Fourier neural networks that are equivariant to the action of any compact continuous group." Kondor_I also describe that the model applies to equivariant neural network structure.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the references of Koyuncu, Lee, and Langhammer, and incorporate with the teachings of Kondor_I by using the teachings of Koyuncu, Lee, and Langhammer of accessing a feature tensor created from model input and creating a sampling matrix, and incorporate with Kondor_I’s teaching of having equivariant MLP projections.
One of ordinary skill in the art would be motivated to do so because by integrating Kondor_I’s framework into the methods of Koyuncu, Lee, and Langhammer, would bring “Clebsch-Gordan CNNs result in real and practical improvements in terms of reduced computational costs (e.g., resources consumed) and increased speed and efficiency, as well as in enhanced overall performance, in comparison with previous approaches,” (see Kondor_I in [0016]).
Claim 5:
Regarding claim 5, Koyuncu in view of Lee, further in view of Langhammer, teach the limitations in claim 1. However, Koyuncu in view of Lee, further in view of Langhammer, did not teach “5. The processing system of claim 1, wherein the sampling matrix comprises a respective column for each respective irreducible representation (irrep) of a set of irreps of a transformation group to which the first activation layer of the machine learning model is equivariant.”
Further, in an analogous art, Kondor_I teach “The processing system of claim 1, wherein the sampling matrix comprises a respective column for each respective irreducible representation (irrep) of a set of irreps of a transformation group to which the first activation layer of the machine learning model is equivariant.”
See Kondor_I in [0082] describe "The inventors have recognized that designing rotation equivariant (i.e., SO(3) equivariant) volumetric CNNs is a desirable goal. The Spherical CNN described above achieves SO(3) equivariance for a single spherical shell by extending functions on the sphere in terms of the irreducible representations of SO(3) and consistently applying only two types of operations: 1) Tensor products of vectors corresponding to different irreducible representations, followed by a Clebsch-Gordan decomposition;
2) Linear mixing of vectors corresponding to the same irreducible representation."
Also, see Kondor_I in [0053, 0055] describe “
PNG
media_image7.png
81
551
media_image7.png
Greyscale
Equation (12) describes the behavior of spherical harmonic vectors under rotations, while equation (15) describes the behavior of Fourier matrices. However, the latter is equivalent to saying that each column of the matrices separately transforms according to equation (12).” When Kondor_I mention each column of the matrices separately transforms based on the equation, Kondor_I means a respective column for each of the irreducible representations mentioned in [0082]. Since equation 12 describes the behavior of spherical harmonic vectors under rotations, and “SO(3) equivariance for a single spherical shell” relates by extending functions on the sphere in terms of the irreducible representations in form of “mixing of vectors that correspond to the same irreducible representation” from [0082], this shows that each column of the matrices relates to each of the irreducible representations.
Further, see Kondor_I in [0009] describe “First, while retaining the connection to noncommutative Fourier analysis, the approach disclosed herein relaxes the requirement that the activation of each layer of the network needs to be a (vector valued) function on SO(3), requiring only that it be expressible as a collection of some number of SO(3)-covariant vectors, referred to herein as “fragments,” corresponding to different irreducible representations of the group. In this sense, the proposed architecture is strictly more general than other recent spherical CNN architectures,” When Kondor_I mention different irreducible representations for the activation of each layer of the network, this relates to irreducible representation (irrep) of a set of irreps of a transformation group to which the first activation layer of the machine learning model is equivariant.
See Kondor_I in [0057] for more details describe “ Finally, by the theorem of complete reducibility of representations of compact groups, any fs that transforms under rotations linearly is reducible into a sequence of irreducible fragments as in equation (14). This means that equation (14) is really the most general possible form for an SO(3) equivariant neural network. As noted above, technically. the terms “equivariant” and “covariant” map to the same concept. The difference between them is one of emphasis. The term “equivariant” may be used when the same group is acting on two objects in a way that is qualitatively similar, as in the case of the rotation group acting on functions on the sphere and on cross-correlation functions on SO(3). The term “covariant” may be used if the actions are qualitatively different, as in the case of rotations of functions on the sphere and the corresponding transformations of equation (15) of the irreducible fragments in a neural network.” Here, Kondor_I describe the term equivariant and is applied to irreducible fragments in a neural network model that processes image data.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the references of Koyuncu, Lee, and Langhammer, and incorporate with the teachings of Kondor_I by using the teachings of Koyuncu, Lee, and Langhammer of accessing a feature tensor created from model input and creating a sampling matrix, and incorporate with Kondor_I’s teaching of having irreducible representations.
One of ordinary skill in the art would be motivated to do so because by integrating Kondor_I’s framework into the methods of Koyuncu, Lee, and Langhammer, would bring “Clebsch-Gordan CNNs result in real and practical improvements in terms of reduced computational costs (e.g., resources consumed) and increased speed and efficiency, as well as in enhanced overall performance, in comparison with previous approaches,” (see Kondor_I in [0016]).
Claims 12 and 21:
Regarding claims 12 and 21, the claims recite similar limitations as corresponding claim 3 and are rejected for similar reasons as claim 3 using similar teachings and rationale.
Claims 14 and 23:
Regarding claims 14 and 23, the claims recite similar limitations as corresponding claim 5, and are rejected for similar reasons as claim 5 using similar teachings and rationale.
Claims 4, 13, 22 are rejected under 35 U.S.C. 103 over Koyuncu in view of Lee, further in view of Langhammer, and further in view of Kondor R., et al., in “Clebsch–gordan nets: a fully fourier space spherical convolutional neural network,” published on November 10, 2018, available in journal Advances in Neural Information Processing Systems at https://proceedings.neurips.cc/paper/2018/file/a3fc981af450752046be179185ebc8b5-Paper.pdf, (hereafter, Kondor_R).
Claim 4:
Regarding claim 4, Koyuncu in view of Lee, further in view of Langhammer, teach the limitations of claim 1. However, Koyuncu in view of Lee, further in view of Langhammer, did not teach “4. The processing system of claim 1, wherein the first feature tensor comprises Fourier coefficients generated by a first layer of the machine learning model and wherein the one or more processors are configured to further execute the processor-executable instructions to cause the processing system to use the sampling matrix to perform an inverse Fourier transform operation on the first feature tensor.”
In an analogous art, Kondor_R teaches “4. The processing system of claim 1, wherein the first feature tensor comprises Fourier coefficients generated by a first layer of the machine learning model and wherein the one or more processors are configured to further execute the processor-executable instructions to cause the processing system to use the sampling matrix to perform an inverse Fourier transform operation on the first feature tensor.”
See Kondor_R on page 4 describe “the inverse Fourier transform is given by
PNG
media_image8.png
123
1058
media_image8.png
Greyscale
The spherical harmonics form an orthonormal basis for L2(S2), so (8) can be seen as a kind of Fourier series on the sphere, in particular, the elements of the f0,f1,f2,... coefficient vectors can be computed relatively easily by
PNG
media_image9.png
100
1322
media_image9.png
Greyscale
and similarly for h. Similarly to usual Fourier series, in practical scenarios spherical harmonic expansions are computed up to some limiting “frequency” L, which depends on the desired resolution. Noncommutative harmonic analysis [26, 27] tells us that functions on the rotation group also admit a type of generalized Fourier transform.”
PNG
media_image9.png
100
1322
media_image9.png
Greyscale
Here, Kondor_R teaches using a collection of matrices Gl to calculate Fourier transform in equation 9, as well as in inverse Fourier transform from page 4.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the references of Koyuncu, Lee, and Langhammer, and incorporate with the teachings of Kondor_R by using the teachings of Koyuncu, Lee, and Langhammer of accessing a feature tensor created from model input and creating a sampling matrix, and incorporate with Kondor_R’s teaching of having equivariant MLP projections.
One of ordinary skill in the art would be motivated to do so because by integrating Kondor_R’s framework into the methods of Koyuncu, Lee, and Langhammer, would achieve “Despite this, it is able to consistently come second or third in the competition, showing that it affords an efficient method to learn from spherical signals,” (See Kondor_R in page 8, section 4. Experiments, subsection 3D Shape Recognition).
Claims 13 and 22:
Regarding claims 13 and 22, the claims recite similar limitations as corresponding claim 4 and are rejected for similar reasons as claim 4 using similar teachings and rationale.
Claims 8, 17, 26 are rejected under 35 U.S.C. 103 over Koyuncu in view of Lee, further in view of Langhammer, and further in view of Cheng, R. et al., (US PG Pub. No. US20220381914A1), published on December 1, 2022, (hereafter, Cheng).
Claim 8:
Regarding claim 8, Koyuncu in view of Lee, further in view of Langhammer, teach the limitations of claim 1. However, Koyuncu in view of Lee, further in view of Langhammer, did not teach “8. The processing system of claim 1, wherein the one or more processors are configured to further execute the processor-executable instructions to cause the processing system to generate a respective sampling matrix for each of a plurality of points in the model input and wherein the model input comprises point cloud data.”
In an analogous art, Cheng teaches “8. The processing system of claim 1, wherein the one or more processors are configured to further execute the processor-executable instructions to cause the processing system to generate a respective sampling matrix for each of a plurality of points in the model input and wherein the model input comprises point cloud data.”
See Cheng in [0028] describe “In an example of preceding fourth aspect of the system, the trained neural network model may be trained to perform semantic segmentation, the sparse input point cloud may be a sparse 3D point cloud generated by a LIDAR unit, and the sparse labeled output point cloud may include semantic labels for each data point corresponding to the sparse input point cloud…” Here, Cheng shows that each data point corresponds to each of the input point cloud data that are part of the model input.
Further, Cheng in [0030] shows “In an example of any of the preceding examples of the preceding fourth aspect of the system, the instructions may configure the processor device to apply the encoder sparse inter-channel attention module or the decoder sparse inter-channel attention module by: obtaining, as input to the encoder sparse inter-channel attention module or the decoder sparse inter-channel attention module, a sparse tensor; processing the sparse tensor using a sparse global pooling squeeze layer to obtain a set of global attention weights; processing the set of global attention weights using a sparse linear excitation layer to obtain a set of channel-wise attention weights; and applying the set of channel-wise attention weights to the sparse tensor to output a scaled feature representation that is a processed sparse tensor having inter-channel attention applied.” Cheng here shows that the system creates for each point that correspond to the point cloud input, semantic labels. Further, Cheng in [0030] mentions, that the system further processes such an input to output a scaled feature representation with a processed sparse tensor having inter-channel attention applied (i.e. sampling matrix). The examiner construes a sampling matrix to also mean image data represented in matrix form that is used to map, shear, or resize an existing image. Here, Cheng mentions the inter-channel attention applied to process a sparse tensor, which includes scalars, vectors, and matrices (grids of numbers) into any number of dimensions, and outputs a processed scaled feature representation which relates to a sampling matrix since this is a filtered version of the original tensor, and all is creating this sampling matrix for each of the data points for the model input. See Cheng in [0068] describe “The input sparse tensor 112 may be mathematically represented as comprising a coordinate matrix of size N×4 and a feature matrix of size N×M.” This shows that the tensor can also be represented in a matrix form.
Later, see Cheng describe in [0118-0119] “The method 800 may be used to compute a block of the neural network model, in which the block includes the sparse intra-channel attention module 106. The neural network model may have been trained (e.g., using supervised training) to perform a perception task using unstructured higher dimensional data (e.g., 3D point cloud) as input… At 802, an input sparse tensor is obtained as input to the block of the neural network model. In general, the input sparse tensor may represent a feature map in which each data point includes coordinate information (representing the location of the data point in higher dimensional space, such as 3D space) and feature information (e.g., a feature vector). In some examples, the input sparse tensor may be the output from a prior layer or block of the neural network model. In some examples, the input sparse tensor may be the output from a preprocessing module (e.g., to preprocess unstructured, higher dimensional data such as a point cloud into the sparse tensor data structure).” Here, Cheng shows an input to the neural network model creates an input sparse tensor, which includes data from a point cloud into the sparse tensor data structure.
See Cheng in [0025] describe “In an example of any of the preceding examples of the third aspect of the system, other sparse convolution blocks in the series of sparse convolution blocks may have respective convolution kernels that leave the data points unchanged in order, and the duplicate sparse convolution block may have a convolution kernel with dimensions equal to the convolution kernel of the particular sparse convolution block in the series of sparse convolution blocks.” Here, Cheng shows the model generates respective kernels or data for each data point.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the references of Koyuncu, Lee, and Langhammer, and incorporate with the teachings of Cheng by using the teachings of Koyuncu, Lee, and Langhammer of accessing a feature tensor created from model input and creating a sampling matrix, and incorporate with Cheng’s teaching of a model input have point cloud data.
One of ordinary skill in the art would be motivated to do so because by integrating Cheng’s framework into the methods of Koyuncu, Lee, and Langhammer, would achieve “The disclosed sparse residual tower module may enable the neural network model to benefit from being able to process the sparse higher dimensional data more efficiently and to generate richer features, which may be useful for performing various perception tasks, for example to perform a semantic segmentation task,” (see Cheng in [0010]).
Claims 17 and 26:
Regarding claims 17 and 26, the claims recite similar limitations as corresponding claim 8 and are rejected for similar reasons as claim 8 using similar teachings and rationale.
Claims 9, 18, 27, and 30 are rejected under 35 U.S.C. 103 over Koyuncu in view of Lee, further in view of Langhammer, and further in view of Hui, J. in an article “Graph Convolutional Networks (GCN) & Pooling”, available on https://jonathan-hui.medium.com/graph-convolutional-networks-gcn-pooling-839184205692, published on February 23, 2021, (hereafter, Hui).
Claim 9:
Regarding claim 9, Koyuncu in view of Lee, further in view of Langhammer, teach the limitations of claim 1.
However, Koyuncu in view of Lee, further in view of Langhammer, did not teach “The processing system of claim 1, wherein the activation output is generated according to
PNG
media_image4.png
72
446
media_image4.png
Greyscale
and wherein: f~(x) is the activation output for the first feature tensor generated based on the model input, x is the model input, n indicates a number of rows in the sampling matrix, A(x) is the sampling matrix generated based on the first feature tensor, and f^(x) is the first feature tensor.”
In an analogous field, Hui teaches “The processing system of claim 1, wherein the activation output is generated according to
PNG
media_image4.png
72
446
media_image4.png
Greyscale
and wherein: f~(x) is the activation output for the first feature tensor generated based on the model input, x is the model input, n indicates a number of rows in the sampling matrix, A(x) is the sampling matrix generated based on the first feature tensor, and f^(x) is the first feature tensor.”
See Hui in pages 21-22, describe the equation
PNG
media_image10.png
322
875
media_image10.png
Greyscale
Where the H(k), j represents an output function, and has the form of sigma σ, which relates to an activation function, with the output function. The function has a Fourier transform on x (Ux^), along with its inverse (UTx), which is similar to A(x) and its inverse A(x)T, along with H:,i(k-1), which represents a feature map for input channel i from the previous layer (i.e. part of an initial feature tensor).
PNG
media_image11.png
705
1002
media_image11.png
Greyscale
Further, see Hui in pages 9-11 describe “ GCN wants  to be normalized to maintain the scale of the output feature vectors. One possibility is to multiple  with D̂⁻¹ where D̂ is the diagonal node degree matrix of  in measuring the degree of each node. At a high level, instead of summing up itself with its neighbor, multiplying the sum with the inverse D̂⁻¹ sort of averages them. Specifically, D̂ is a diagonal matrix with each diagonal element D̂ᵢᵢ counts the number of edges for the corresponding node i. And the output for each hidden layer becomes σ(D̂⁻¹ÂHⁱWⁱ), instead of σ(ÂHⁱWⁱ).” Here, Hui describes using a matrix within a similar equation for an output function σ(D̂⁻¹ÂHⁱWⁱ).
See Hui in page 30, in Sampling section describe “FastGCN refines the sampling algorithm further. Instead of sampling neighbors for each node, FastGCN uses importance sampling to reduce variance. It samples nodes (u) from a distribution q that reflects how well it is connected to other nodes. Then, it applies importance sampling to estimate the loss gradient of node v from u.” Here, the 1/tl sample vertices relate to 1/n or 1 divided by the sample size from the equation:
PNG
media_image4.png
72
446
media_image4.png
Greyscale
PNG
media_image12.png
356
875
media_image12.png
Greyscale
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the references of Koyuncu, Lee, and Langhammer, and incorporate with the teachings of Hui by using the teachings of Koyuncu, Lee, and Langhammer of accessing a feature tensor created from model input and creating a sampling matrix, and incorporate with Hui’s teaching of having similar equation format as
PNG
media_image4.png
72
446
media_image4.png
Greyscale
.
One of ordinary skill in the art would be motivated to do so because by integrating Hui’s framework into the methods of Koyuncu, Lee, and Langhammer, would bring “ a large percentage of machine learning (ML) problems will be much natural and effective to be modeled by a graph,” (see Hui in page 4).
Claims 18 and 27:
Regarding claims 18 and 27, the claims recite similar limitations as corresponding claim 9 and are rejected for similar reasons as claim 9 using similar teachings and rationale.
Claim 30:
Regarding claim 30, the claim recites similar limitations as corresponding independent claim 9 and is rejected for similar reasons as claim 9 using similar teachings and rationale.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to WENWEI ZENG whose telephone number is (571)272-7111. The examiner can normally be reached Monday-Friday, 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, Usmaan Saeed can be reached at (571) 272-4046. 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.
/WenWei Zeng/Examiner, Art Unit 2146
/USMAAN SAEED/Supervisory Patent Examiner, Art Unit 2146
Prosecution Insights