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 .
Status of the Claims
Claims 1, 8, 15, and 22 have been amended. Claims 4, 6, 11, 13, 18, 20, and 24-25 have been canceled. Claims 1-3, 5, 7-10, 12, 14-17, 19, 21-23, and 26-30 are currently pending and have been considered by the Examiner.
Drawings
Fig. 3 of the drawings is objected to. The equations in the bottom right of Fig. 3 and “already computed” products appear to be incorrect. For example, it appears the first row should read σ1*(x2+x6+x3+x7) with the already computed products being σ1 times (x2+x6). Applicant should review all equations and indicate support in the specification.
Fig. 4 of the drawings is objected to. The equations resulting from the convolution with 3 x 3 kernel appear to be incorrect. For example, it appears the second row should read α2*(x2+x3+x5+x6). In element 404, the text in the first 2 x 2 matrix is illegible. Applicant should review all equations and indicate support in the specification.
Corrected drawing sheets in compliance with 37 CFR 1.121(d) are required in reply to the Office action to avoid abandonment of the application. Any amended replacement drawing sheet should include all of the figures appearing on the immediate prior version of the sheet, even if only one figure is being amended. The figure or figure number of an amended drawing should not be labeled as “amended.” If a drawing figure is to be canceled, the appropriate figure must be removed from the replacement sheet, and where necessary, the remaining figures must be renumbered and appropriate changes made to the brief description of the several views of the drawings for consistency. Additional replacement sheets may be necessary to show the renumbering of the remaining figures. Each drawing sheet submitted after the filing date of an application must be labeled in the top margin as either “Replacement Sheet” or “New Sheet” pursuant to 37 CFR 1.121(d). If the changes are not accepted by the examiner, the applicant will be notified and informed of any required corrective action in the next Office action. The objection to the drawings will not be held in abeyance.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-3, 5, 7-10, 12, 14-17, 19, 21-23, and 26-30 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
The term “efficient” in claim 1, line 6 is a relative term which renders the claim indefinite. The term “efficient” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. The term “efficient” is a subjective term under MPEP 2173.05(b), subsection IV. because one of ordinary skill in the art would not be able to ascertain whether or not a “variable-length vector multiplication unit” is efficient based on the written disclosure. Examiner treats “an efficient variable-length vector multiplication unit” as “a variable-length vector multiplication unit”.
Claims 2-3, 5, and 7 are rejected for failing to cure the deficiencies of claim 1.
Claim 8 recites a system which implements the same features as the method of claim 1 and is therefore rejected for at least the same reasons.
Claims 9-10, 12, and 14 are rejected for failing to cure the deficiencies of claim 8.
Claims 15 recites a product which implements the same features as the method of claim 1 and is therefore rejected for at least the same reasons.
Claims 16-17, 19, and 21 are rejected for failing to cure the deficiencies of claim 15.
Claim 22 recites the term “efficient” in line 10. This is a relative term which renders the claim indefinite for the same reasons claim 1 is rendered indefinite.
Claims 23 and 26-30 are rejected for failing to cure the deficiencies of claim 22.
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 1-3, 5, 7-10, 12, 14-17, 19, 21-23, and 26-30 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1: Claims 1-3, 5, 7, 22-23, and 26-30 each recites a method. Claims 8-10, 12, and 14 each recites a processing system comprising processors. Claims 15-17, 19, and 21 each recites a non-transitory computer-readable medium (a product). A method, a system, and a product each falls under one of the four statutory categories of patent eligible subject matter.
Claim 1
Step 2A Prong 1: Generating a set of basis masks for a convolution layer of a machine learning model, wherein each basis mask comprises a binary mask is a mathematical calculation based on specification paragraphs [0043]-[0044] and [0051]-[0061].
Determining a set of scaling factors, wherein each scaling factor of the set of scaling factors corresponds to a basis mask in the set of basis masks is a judgement and evaluation mental process which can reasonably be performed in the human mind with the aid of pencil and paper. A scaling factor is merely a numerical value for scaling a basis mask, and a human mind can reasonably determine a numerical value to use in a mathematical calculation.
Generating… a composite kernel based on the set of basis masks and the set of scaling factors is a mathematical calculation based on specification paragraphs [0043]-[0044] and [0051]-[0061].
Training the machine learning model… through a training process that imposes a structured property associated with the composite kernel on the convolutional layer of the machine learning model is a mathematical calculation. Specification paragraphs [0105]-[0106], [0109] discloses training the machine learning model by optimizing a total loss function. This amounts to performing an optimization algorithm, which is a mathematical calculation and a mathematical concept.
The training process comprises performing a convolution operation based on the composite kernel is a mathematical calculation. Specification paragraphs [0043]-[0044] and [0051]-[0061] disclose equations for performing the convolution operation based on the composite kernel.
The training process reduces a number of parameters of the machine learning model as a result of imposing the structured property associated with the composite kernel on the convolutional layer of the machine learning model is a mathematical calculation. Specification paragraph [0061] discloses mathematical concepts of reducing a number of parameters.
The imposing of the structured property associated with the composite kernel on the convolutional layer of the machine learning model is accomplished by a sum-pooling operation… comprising a convolution with a Toeplitz matrix that is based on kernel made of all ones is a mathematical calculation based on specification paragraphs [0099]-[0101]. The claim recites an abstract idea.
Step 2A Prong 2 and Step 2B: An efficient variable-length vector multiplication unit (VMU) of a hardware accelerator amounts to a generic computer hardware component for applying the abstract ideas on a generic computer under MPEP 2106.05(f).
Using the hardware accelerator amounts to a generic computer hardware component for applying the abstract ideas on a generic computer under MPEP 2106.05(f).
Perform by an extract sum unit (ESU) of the hardware accelerator amounts to a generic computer hardware component for applying the abstract ideas on a generic computer under MPEP 2106.05(f).
The additional elements as disclosed above, alone or in combination, do not integrate the abstract ideas into a practical application as they are generic computer functions that are implemented to perform the abstract ideas disclosed above. The claim is directed to an abstract idea. The additional elements as disclosed above, in combination with the abstract ideas, are not sufficient to amount to significantly more than the abstract ideas as they are generic computer functions that are implemented to perform the abstract ideas disclosed above. The claim is not patent eligible.
Claim 2 incorporates the rejection of claim 1.
Step 2A Prong 1: The abstract ideas of claim 1 are incorporated. Performing the convolution operation based on the composite kernel comprises: … for each respective basis mask in the set of basis masks associated with the composite kernel: extracting a subset of the input data for processing based on the respective basis mask, computing a basis sum for the respective basis mask based on the subset of the input data for the respective basis mask, and computing a partial convolution layer output by applying a scaling factor corresponding to the respective basis mask to the basis sum are mathematical calculations based on specification paragraphs [0043]-[0044] and [0051]-[0061].
Generating a convolution layer output by summing each partial convolution layer output associated with each basis mask in the set of basis masks is a mathematical calculation.
Step 2A Prong 2: Receiving input data amounts to mere data-gathering, an insignificant extra-solution activity under MPEP 2106.05(g).
Step 2B: Receiving input data is recognized by the courts as a well-understood, routine, conventional activity under MPEP 2106.05(d)(II). The claim is not patent eligible.
Claim 3 incorporates the rejection of claim 1.
Step 2A Prong 1: The abstract ideas of claim 1 are incorporated. The composite kernel comprises a structured kernel is a mathematical calculation. The convolution operation comprises a structured convolution is a mathematical calculation.
Step 2A Prong 2 and Step 2B: The claim does not recite any additional elements which would integrate the abstract ideas into a practical application or which, in combination with the abstract ideas, would be sufficient to amount to significantly more than the abstract ideas. The claim is not patent eligible.
Claim 5 incorporates the rejection of claim 1.
Step 2A Prong 1: The abstract ideas of claim 1 are incorporated. Training the machine learning model with a structural regularization term is a mathematical calculation based on paragraphs [0105]-[0106].
Step 2A Prong 2 and Step 2B: The claim does not recite any additional elements which would integrate the abstract ideas into a practical application or which, in combination with the abstract ideas, would be sufficient to amount to significantly more than the abstract ideas. The claim is not patent eligible.
Claim 7 incorporates the rejection of claim 1.
Step 2A Prong 1: The abstract ideas of claim 1 are incorporated. Applying a structural decomposition to the convolution layer to generate a decomposed convolution layer is a mathematical calculation based on paragraphs [0043]-[0044] and [0051]-[0061].
Training the machine learning model using the decomposed convolution layer and a task loss function based on paragraphs [0105]-[0106].
Step 2A Prong 2 and Step 2B: The claim does not recite any additional elements which would integrate the abstract ideas into a practical application or which, in combination with the abstract ideas, would be sufficient to amount to significantly more than the abstract ideas. The claim is not patent eligible.
Claim 8 recites a system which implements the same features as the method of claim 1 and is therefore rejected for at least the same reasons.
In Step 2A Prong 2 and Step 2B, the limitations “A processing system for machine learning, comprising: at least one memory comprising computer-executable instructions; and one or more processors configured to execute the computer-executable instructions and cause the processing system” amount to generic computer hardware components for applying the abstract ideas on a generic computer under MPEP 2106.05(f). The claim is not patent eligible.
Claims 9-10, 12, and 14 each recites a system which implements the same features as the method of claims 2-3, 5, and 7, respectively, and are therefore rejected for at least the same reasons.
Claim 15 recites a product which implements the same features as the method of claim 1 and is therefore rejected for at least the same reasons.
In Step 2A Prong 2 and Step 2B, the limitations “A non-transitory computer-readable medium comprising instructions that, when executed by one or more processors of a processing system, cause the processing system to perform a method of machine learning” amount to generic computer hardware components for applying the abstract ideas on a generic computer under MPEP 2106.05(f). The claim is not patent eligible.
Claims 16-17, 19, and 21 each recites a product which implements the same features as the method of claims 2-3, 5, and 7, respectively, and are therefore rejected for at least the same reasons.
Claim 22
Step 2A Prong 1: Generating a set of basis masks for a convolution layer of a machine learning model, wherein each basis mask comprises a binary mask is a mathematical calculation based on specification paragraphs [0043]-[0044] and [0051]-[0061].
Determining a set of scaling factors, wherein each scaling factor of the set of scaling factors corresponds to a basis mask in the set of basis masks is a judgement and evaluation mental process which can reasonably be performed in the human mind with the aid of pencil and paper. A scaling factor is merely a numerical value for scaling a basis mask, and a human mind can reasonably determine a numerical value to use in a mathematical calculation.
Generating a sum-pooled output based on input data to the convolution layer of the machine learning model is a mathematical calculation based on specification paragraph [0099].
Training… the machine learning model through a training process that imposes a structured property associated with a composite kernel, generated… based on the set of basis masks and the set of scaling factors, on the convolutional layer of the machine learning model is a mathematical calculation based on paragraphs [0043]-[0044], [0051]-[0061], [0105]-[0106], [0109].
The training process comprises generating a convolution layer output based on the sum-pooled output and the set of scaling factors is a mathematical calculation based on paragraphs [0043]-[0044] and [0051]-[0061].
The training process reduces a number of parameters of the machine learning model as a result of imposing the structured property associated with the composite kernel on the convolutional layer of the machine learning model is a mathematical calculation based on paragraph [0061].
The imposing of the structured property associated with the composite kernel on the convolutional layer of the machine learning model is accomplished by a sum-pooling operation… comprising a convolution with a Toeplitz matrix that is based on a kernel made of all ones to produce the sum-pooled output is a mathematical calculation based on paragraph [0099]-[0101]. The claim recites an abstract idea.
Step 2A Prong 2 and Step 2B: Using a hardware accelerator amounts to a generic computer hardware component for applying the abstract ideas on a generic computer under MPEP 2106.05(f).
An efficient variable-length vector multiplication unit (VMU) of the hardware accelerator amounts to a generic computer hardware component for applying the abstract ideas on a generic computer under MPEP 2106.05(f).
Perform by an extract sum unit (ESU) of the hardware accelerator amounts to a generic computer hardware component for applying the abstract ideas on a generic computer under MPEP 2106.05(f).
The additional elements as disclosed above, alone or in combination, do not integrate the abstract ideas into a practical application as they are generic computer functions that are implemented to perform the abstract ideas disclosed above. The claim is directed to an abstract idea. The additional elements as disclosed above, in combination with the abstract ideas, are not sufficient to amount to significantly more than the abstract ideas as they are generic computer functions that are implemented to perform the abstract ideas disclosed above. The claim is not patent eligible.
Claim 23 incorporates the rejection of claim 22.
Step 2A Prong 1: The abstract ideas of claim 22 are incorporated. Generating the sum-pooled output based on the input data to the convolution layer comprises: for each respective basis mask in the set of basis masks: extracting a subset of the input data for processing based on the respective basis mask; and computing the sum-pooled output for the respective basis mask based on the subset of the input data for the respective basis mask are mathematical calculations based on specification paragraphs [0051]-[0061] and [0099].
Step 2A Prong 2 and Step 2B: The claim does not recite any additional elements which would integrate the abstract ideas into a practical application or which, in combination with the abstract ideas, would be sufficient to amount to significantly more than the abstract ideas. The claim is not patent eligible.
Claim 26 incorporates the rejection of claim 22.
Step 2A Prong 1: The abstract ideas of claim 22 are incorporated. The sum-pooled output is associated with a first stride of a structured convolution, and the convolution layer output is associated with the first stride of the structured convolution are mathematical calculations.
Generating a second sum-pooled output associated with a second stride of the structured convolution
Step 2A Prong 2 and Step 2B: Using the ESU and the VMU amount to generic computer hardware components for applying the abstract ideas on a generic computer under MPEP 2106.05(f). The claim is not patent eligible.
Claim 27 incorporates the rejection of claim 22.
Step 2A Prong 1: The abstract ideas of claim 22 are incorporated.
Step 2A Prong 2 and Step 2B: Configuring the ESU based on a structure of each basis mask in the set of basis masks amounts to a field of use and technological environment under MPEP 2106.05(h). The claim is not patent eligible.
Claim 28 incorporates the rejection of claim 27.
Step 2A Prong 1: The abstract ideas of claim 27 are incorporated.
Step 2A Prong 2 and Step 2B: Configuring the VMU based on a number of basis masks in the set of basis masks amounts to a field of use and technological environment under MPEP 2106.05(h). The claim is not patent eligible.
Claim 29 incorporates the rejection of claim 22.
Step 2A Prong 1: The abstract ideas of claim 22 are incorporated. Generating the sum-pooled output comprises performing a cross-kernel sum sharing operation is a mathematical calculation based on specification paragraphs [0072]-[0078].
Step 2A Prong 2 and Step 2B: The claim does not recite any additional elements which would integrate the abstract ideas into a practical application or which, in combination with the abstract ideas, would be sufficient to amount to significantly more than the abstract ideas. The claim is not patent eligible.
Claim 30 incorporates the rejection of claim 22.
Step 2A Prong 1: The abstract ideas of claim 22 are incorporated. Generating the sum-pooled output comprises performing a cross-stride sum sharing operation is a mathematical calculation based on specification paragraphs [0079]-[0080].
Step 2A Prong 2 and Step 2B: The claim does not recite any additional elements which would integrate the abstract ideas into a practical application or which, in combination with the abstract ideas, would be sufficient to amount to significantly more than the abstract ideas. The claim is not patent eligible.
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.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1-3, 7-10, 14-17, 21-23, 27-30 are rejected under 35 U.S.C. 103 as being unpatentable over Qiu et al. (“DCFNet: Deep Neural Network with Decomposed Convolutional Filters”, cited in PTO-892 issued 09/06/2024) in view of Rastegari et al. (“XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks”, cited in PTO-892 issued 05/16/2025), Croxford (US 20190258306 A1, cited in PTO-892 issued 11/06/2025), and Hutel et al. (US 20210208224 A1, cited in PTO-892 issued 09/06/2024).
Regarding claim 1, Qiu teaches: A method of machine learning, comprising: generating a set of basis masks for a convolution layer of a machine learning model, … (Qiu’s Abstract, lines 1-8 teaches the claimed “generating a set of basis masks” by decomposing convolutional filters (mapped to the claimed “convolutional layer”) into bases (mapped to the claimed “basis masks”). On p. 3, § 2.2 from line 10 to the end of col. 1 discloses the same generating process. Here, the limitation “basis mask” corresponds to each of Qiu’s basis ψk. The limitation “a convolution layer” corresponds to Qiu’s convolutional filters Wλ’, λ(u) at a certain layer. Further, each of the basis ψk could be of L x L size according to Fig. 1 and its caption on p. 1, col. 2. These mappings appear consistent with Applicant’s specification paragraph [0039].)
determining a set of scaling factors, wherein each scaling factor of the set of scaling factors corresponds to a basis mask in the set of basis masks; (On the right side of Equation (2) on p. 3, col. 1, each weight matrix (a λ’, λ)k corresponds to a basis ψk(u). The limitation “each scaling factor” constitutes each weight matrix and the limitation “a basis mask” constitutes the basis ψk(u).)
generating, λ’, λ(u).)
training the machine learning model,
the training process comprises performing a convolution operation based on the composite kernel; (P. 3, § 2.2 from line 10 to the end of col. 1)
the training process reduces a number of parameters of the machine learning model as a result of imposing the structured property associated with the composite kernel on the convolutional layer of the machine learning model; and (P. 1, col. 2, lines 12-15 and p. 7, col. 1, § 4.3 to col. 2, L. 8.)
the imposing of the structured property associated with the composite kernel on the convolutional layer of the machine learning model is accomplished by a sum-pooling operation,
However, Qiu does not explicitly teach (underlines indicate limitations not taught): wherein each basis mask comprises a binary mask;
generating, by an efficient variable-length vector multiplication unit (VMU) of a hardware accelerator, a composite kernel
training the machine learning model, using the hardware accelerator
a sum-pooling operation, performed by an extract sum unit (ESU) of the hardware accelerator, comprising a convolution with a Toeplitz matrix that is based on a kernel made of all ones.
But Rastegari teaches: a
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have incorporated Rastegari’s binary filters and pooling layer from its XNOR-Network into Qiu. Since Rastegari’s binary filters comprise only binary weights, Qiu’s basis mask would be a binary mask in the combination. A motivation for the combination is that “The binary weight filters reduce memory usage by a factor of ∼ 32× compared to single-precision filters.” (Rastegari, P. 5, § 3.1, lines 2-4 below Equation (1))
However, Qiu and Rastegari do not explicitly teach: generating, by an efficient variable-length vector multiplication unit (VMU) of a hardware accelerator, a composite kernel
training the machine learning model, using the hardware accelerator
a sum-pooling operation, performed by an extract sum unit (ESU) of the hardware accelerator, comprising a convolution with a Toeplitz matrix
But Croxford teaches: generating, by an efficient variable-length vector multiplication unit (VMU) of a hardware accelerator, a
training the machine learning model, using the hardware accelerator ([0035], lines 1-8 and final 7 lines discloses generating a kernel during neural network training using a neural network accelerator.)
a
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have trained Qiu and Rastegari’s convolutional neural network using Croxford’s neural network accelerator, and it would have been obvious to have performed the sum-pooling operation using Croxford’s addition unit. A motivation for the combination is that a neural network accelerator is suited to the high power consumption of classification of an image. (Croxford, [0038]).
However, Qiu, Rastegari, and Croxford do not explicitly teach: Toeplitz
But Hutel teaches: Toeplitz ([0035], lines 1-5)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have incorporated Hutel’s use of a Toeplitz matrix into the combination of Qiu, Rastegari, and Croxford. A motivation for the combination is to optimize convolution and/or reduce computation complexity (Hutel, [0035], lines 5-13). In the combination, Rastegari’s weight filter would contain all positive ones or all negative ones.
Regarding claim 2, the combination of Qiu, Rastegari, Croxford, and Hutel teaches: The method of Claim 1,
Qiu teaches: wherein performing the convolution operation based on the composite kernel comprises: receiving input data; (On p. 3 in col. 1, the Ψ-step below Equation (2) states “(Ψ-step) the input is convolved with each of the basis Ψk … The convolution for each input channel is independent from other channels, adding computational efficiency.” The limitation “input data” corresponds to Qiu’s input. Performing convolution for each input indicates the input data has been received.)
for each respective basis mask in the set of basis masks associated with the composite kernel: extracting a subset of the input data for processing based on the respective basis mask; (On p. 3 in col. 1, the Ψ-step below Equation (2) discloses the input is convolved at each input channel. The limitation “a subset of the input data” corresponds to each input channel while the claimed “input” comprises multiple disclosed input channels.)
computing a basis sum for the respective basis mask based on the subset of the input data for the respective basis mask; and (The disclosed summing result after “the input is convolved with each of the basis Ψk” is mapped to the claimed “a basis sum for the respective basis mask,” part of the disclosed “intermediate output” that is later linearly transformed by an effectively fully-connected weight matrix (a λ’, λ)k as indicated by the summation in equation 2.)
computing a partial convolution layer output by applying a scaling factor corresponding to the respective basis mask to the basis sum; and (On p. 3 in col. 1, the a-step discloses “the intermediate output is linearly transformed by an effectively fully-connected weight matrix (a λ’, λ)k” as indicated by the summation in equation 2.)
generating a convolution layer output by summing each partial convolution layer output associated with each basis mask in the set of basis masks. (On the right side of Equation (2), the summation operation indicates summing each partial convolution layer output for k=1 … K.)
Regarding claim 3, the combination of Qiu, Rastegari, Croxford, and Hutel teaches: The method of Claim 1, wherein:
Qiu teaches: the composite kernel comprises a structured kernel; and (On p. 3 in col. 1, the summation at the right side of Equation (2) has been mapped to the claimed “composite kernel,” which has been structured by combining multiple bases Ψk (u). This mapping is consistent with the instant specification because paragraph [0033] discloses “This structure can be thought of as constructing the convolution kernel by super-imposing several lower-resolution kernels, which may be referred to as basis kernels, each defined by a basis mask and a scaling factor.”)
the convolution operation comprises a structured convolution. (On p. 3 in col. 1, the Ψ-step below Equation (2) discloses “the input is convolved with each of the basis Ψk.” Therefore, the convolution operation is structured by combining multiple convolutions based on K bases.)
Regarding claim 7, the combination of Qiu, Rastegari, Croxford, and Hutel teaches: The method of Claim 1, further comprising:
Qiu teaches: applying a structural decomposition to the convolution layer to generate a decomposed convolution layer; and (On p. 1, caption for Fig. 1 teaches decomposing an L x L x M’ x M convolutional layer into a decomposed convolution layer. The caption states, “The basis can carry prior (explainable) structure if available.” Therefore, Qiu’s decomposition is a structural decomposition.)
training the machine learning model using the decomposed convolution layer and
However, Qiu, Rastegari, and Croxford do not explicitly teach: a task loss function.
But Hutel teaches: a task loss function (In [0035], equation 2 is a cost function that minimizes the L2 norm between the original data X and its reconstruction X̂ (or ground truth).)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have trained Qiu, Rastegari, Croxford, and Hutel’s model using the cost function. A motivation for the combination is that the model is trained via back-propagation by deriving the corresponding gradient for the cost function. (Hutel, [0035])
Claim 8 recites a system which implements the same features as the method of claim 1 and is therefore rejected for at least the same reasons.
However, Qiu does not explicitly teach: A processing system for machine learning, comprising: at least one memory comprising computer-executable instructions; and one or more processors configured to execute the computer-executable instructions and cause the processing system to:
But Rastegari teaches: A processing system for machine learning, comprising: at least one memory comprising computer-executable instructions; and (P. 2, lines 20-24)
one or more processors configured to execute the computer-executable instructions and cause the processing system to: (P. 2, lines 20-24)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have implemented the method of claim 1 using Rastegari’s memory and CPU. A motivation for the combination is to execute a convolutional neural network in the real world.
Croxford teaches processors and storage at [0028], lines 1-3 and [0030], lines 1-4.
Claims 9-10 and 14 each recites a system which implements the same features as the method of claims 2-3 and 7, respectively, and are therefore rejected for at least the same reasons.
Claim 15 recites a product which implements the same features as the method of claim 1 and is therefore rejected for at least the same reasons.
However, Qiu does not explicitly teach: A non-transitory computer-readable medium comprising instructions that, when executed by one or more processors of a processing system, cause the processing system to perform a method of machine learning
But Rastegari teaches: A non-transitory computer-readable medium comprising instructions that, when executed by one or more processors of a processing system, cause the processing system to perform a method of machine learning (P. 2, lines 20-24)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have implemented the method of claim 1 using Rastegari’s memory and CPU. A motivation for the combination is to execute a convolutional neural network in the real world.
Croxford teaches processors and storage at [0028], lines 1-3 and [0030], lines 1-4.
Claims 16-17 and 21 each recites a product which implements the same features as the method of claims 2-3 and 7, respectively, and are therefore rejected for at least the same reasons.
Regarding claim 22, Qiu teaches: A method, comprising: generating a set of basis masks for a convolution layer of a machine learning model, … (Qiu’s Abstract teaches the claimed “generating a set of basis masks” by decomposing convolutional filters (mapped to the claimed “convolutional layer”) into bases (mapped to the claimed “basis masks”), disclosing, “In this paper, we suggest to decompose convolutional filters in CNN as a truncated expansion with pre-fixed bases, namely the Decomposed Convolutional Filters network (DCFNet), where the expansion coefficients remain learned from data.” On p. 3, § 2.2 from line 10 to the end of col. 1 discloses the same generating process. Here, the limitation “basis mask” corresponds to each of Qiu’s basis ψk. The limitation “a convolution layer” corresponds to Qiu’s convolutional filters Wλ’, λ(u) at a certain layer. Further, each of the basis ψk could be of L x L size according to Fig. 1 and its caption on p. 1, col. 2. These mappings appear consistent with Applicant’s specification paragraph [0039].)
determining a set of scaling factors, wherein each scaling factor of the set of scaling factors corresponds to a basis mask in the set of basis masks; (On the right side of Equation (2) on p. 3, col. 1, each weight matrix (a λ’, λ)k corresponds to a basis ψk(u). The limitation “each scaling factor” constitutes each weight matrix and the limitation “a basis mask” constitutes the basis ψk(u).)
generating a sum-pooled output based on input data to the convolution layer of the machine learning model; and (P. 3 in col. 1, equation (2) and ψ-step and a-step below it. The summing result, part of the intermediate output, for each basis Ψk(u) after the input is convolved with each of the basis Ψk is linearly transformed by (a λ’, λ)k as indicated by the summation in equation 2. The limitation “a sum-pooled output” as claimed is the result of the disclosed summing process.)
training,
the training process comprises generating a convolution layer output based on the sum-pooled output and the set of scaling factors; (P. 3, § 2.2 from line 10 to the end of col. 1.)
the training process reduces a number of parameters of the machine learning model as a result of imposing the structured property associated with the composite kernel on the convolutional layer of the machine learning model; and (P. 1, col. 2, lines 12-15 and p. 7, col. 1, § 4.3 to col. 2, L. 8.)
the imposing of the structured property associated with the composite kernel on the convolutional layer of the machine learning model is accomplished by a sum-pooling operation,
However, Qiu does not explicitly teach (underlines indicate limitations not taught): wherein each basis mask comprises a binary mask;
training, using a hardware accelerator, the machine learning model through a training process that imposes a structured property associated with a composite kernel, generated by an efficient variable-length vector multiplication unit (VMU) of the hardware accelerator
a sum-pooling operation, performed by an extract sum unit (ESU) of the hardware accelerator, comprising a convolution with a Toeplitz matrix that is based on a kernel made of all ones to produce the sum-pooled output.
But Rastegari teaches: a
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have incorporated Rastegari’s binary filters and pooling layer from its XNOR-Network into Qiu. Since Rastegari’s binary filters comprise only binary weights, Qiu’s basis mask would be a binary mask in the combination. A motivation for the combination is that “The binary weight filters reduce memory usage by a factor of ∼ 32× compared to single-precision filters.” (Rastegari, P. 5, § 3.1, lines 2-4 below Equation (1))
However, Qiu and Rastegari do not explicitly teach: training, using a hardware accelerator, the machine learning model through a training process that imposes a structured property associated with a composite kernel, generated by an efficient variable-length vector multiplication unit (VMU) of the hardware accelerator
a sum-pooling operation, performed by an extract sum unit (ESU) of the hardware accelerator, comprising a convolution with a Toeplitz matrix
But Croxford teaches: training, using a hardware accelerator, the machine learning model through a training process, and a kernel, generated by an efficient variable-length vector multiplication unit (VMU) of the hardware accelerator ([0035], lines 1-8 and final 7 lines, and [0054] discloses generating a kernel during neural network training using a neural network accelerator which comprises a multiplication unit 132.)
a
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have trained Qiu and Rastegari’s convolutional neural network using Croxford’s neural network accelerator, and it would have been obvious to have performed the sum-pooling operation using Croxford’s addition unit. A motivation for the combination is that a neural network accelerator is suited to the high power consumption of classification of an image. (Croxford, [0038]).
However, Qiu, Rastegari, and Croxford do not explicitly teach: Toeplitz
But Hutel teaches: Toeplitz ([0035], lines 1-5)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have incorporated Hutel’s use of a Toeplitz matrix into the combination of Qiu, Rastegari, and Croxford. A motivation for the combination is to optimize convolution and/or reduce computation complexity (Hutel, [0035], lines 5-13). In the combination, Rastegari’s weight filter would contain all positive ones or all negative ones.
Regarding claim 23, the combination of Qiu, Rastegari, Croxford, and Hutel teaches: The method of Claim 22,
Qiu teaches: generating the sum-pooled output based on the input data to the convolution layer comprises: for each respective basis mask in the set of basis masks: extracting a subset of the input data for processing based on the respective basis mask; and (On p. 3, § 2.2, Equation (2) and ψk(u) are mapped to the claimed basis mask. In the Ψ-step below Equation (2), each input channel is mapped to the claimed “a subset of the input data” while the claimed “input” comprises multiple disclosed input channels. The Ψ-step discloses “the input is convolved with each of the basis ψk.)
computing the sum-pooled output for the respective basis mask based on the subset of the input data for the respective basis mask. (The claimed “sum-pooled output” is mapped to the output after “the input is convolved with each of the basis ψk” according to the Ψ-step below Equation (2).)
Regarding claim 27, the combination of Qiu, Rastegari, Croxford, and Hutel teaches: The method of Claim 22, further comprising
Qiu teaches: [performing operations]
However, Qiu, Rastegari, and Hutel do not explicitly teach: configuring the ESU
But Croxford teaches: configuring the ESU ([0057] on p. 7, col. 1, lines 6-9)
A motivation for the combination is the same as the motivation given for claim 22.
Regarding claim 28, the combination of Qiu, Rastegari, Croxford, and Hutel teaches: The method of Claim 27, further comprising
Qiu teaches: [perform operations]
However, Qiu, Rastegari, and Hutel do not explicitly teach: configuring the VMU
But Croxford teaches: configuring the VMU ([0035], lines 1-8 and final 7 lines, and [0054] discloses configuring a multiplication unit of a neural network accelerator to perform operations based on kernels.)
A motivation for the combination is the same as the motivation given for claim 22.
Regarding claim 29, the combination of Qiu, Rastegari, Croxford, and Hutel teaches: The method Claim 22,
Qiu teaches generating the sum-pooled output. However, Qiu, Rastegari, and Hutel do not explicitly teach: wherein generating the sum-pooled output comprises performing a cross-kernel sum sharing operation.
But Croxford teaches: wherein generating the pooling outputs of the convolution layer 120b, and all of [0035] discloses a convolution involves a multiply-add operation between a 3 x 3 kernel and a 3 x 3 image patch. The convolution output (i.e., the weighted sum) is shared by all 9 pixels of the image patch. The kernel is slid across the image by 1 pixel at a time after each multiply-add operation, and this corresponds to the limitation “cross-kernel” as claimed.)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have shared a convolution output by all pixels of an image patch in the combination of Qiu, Rastegari, Croxford, and Hutel. A motivation for the combination is to classify image data. (Croxford, [0035], lines 1-4)
Regarding claim 30, the combination of Qiu, Rastegari, Croxford, and Hutel teaches: The method Claim 22,
Qiu teaches sum-pooling. However, Qiu, Rastegari, and Hutel do not explicitly teach: wherein generating the sum-pooled output comprises performing a cross-stride sum sharing operation.
But Croxford teaches: wherein generating the
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have shared a convolution output by all pixels of an image patch in the combination of Qiu, Rastegari, Croxford, and Hutel. A motivation for the combination is to classify image data. (Croxford, [0035], lines 1-4)
Claims 5, 12, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Qiu et al. (“DCFNet: Deep Neural Network with Decomposed Convolutional Filters”, cited in PTO-892 issued 09/06/2024) in view of Rastegari et al. (“XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks”, cited in PTO-892 issued 05/16/2025), Croxford (US 20190258306 A1, cited in PTO-892 issued 11/06/2025), Hutel et al. (US 20210208224 A1, cited in PTO-892 issued 09/06/2024), and Stajner (US 20170293857 A1, cited in PTO-892 issued 11/06/2025).
Regarding claim 5, the combination of Qiu, Rastegari, Croxford, and Hutel teaches: The method of Claim 1, further comprising
Qiu teaches: training the machine learning model (P. 7, col. 1, § 4.3 to col. 2, L. 8.)
However, Qiu, Rastegari, Croxford, and Hutel do not explicitly teach: with a structural regularization term.
But Stajner teaches: training the machine learning model with a structural regularization term. ([0037], lines 1-13, and page 5, col. 1, lines 2-7, and 13-18)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have trained Qiu, Rastegari, Croxford, and Hutel’s model using a regularized training error. A motivation for the combination is that the regularization term may be configured to penalize model complexity (Stajner, [0037]). This can improve its accuracy.
Claim 12 recites a system which implements the same features as the method of claim 5 and is therefore rejected for at least the same reasons.
Claim 19 recites a product which implements the same features as the method of claim 5 and is therefore rejected for at least the same reasons.
Claim 26 is rejected under 35 U.S.C. 103 as being unpatentable over Qiu et al. (“DCFNet: Deep Neural Network with Decomposed Convolutional Filters”, cited in PTO-892 issued 09/06/2024) in view of Rastegari et al. (“XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks”, cited in PTO-892 issued 05/16/2025), Croxford (US 20190258306 A1), Hutel et al. (US 20210208224 A1, cited in PTO-892 issued 09/06/2024), and Balasubramanian (US 20190102640 A1, cited in PTO-892 issued 11/06/2025).
Regarding claim 26, the combination of Qiu, Rastegari, Croxford, and Hutel teaches: The method Claim 22, wherein:
Qiu teaches sum-pooling and Croxford teaches the ESU and VMU. However, Qiu, Rastegari, Croxford, and Hutel do not explicitly teach, as a whole: the sum-pooled output is associated with a first stride of a structured convolution, the convolution layer output is associated with the first stride of the structured convolution, and the method further comprises generating a second sum-pooled output associated with a second stride of the structured convolution with the ESU concurrent with the VMU generating the convolution layer output associated with the first stride of the structured convolution.
But Balasubramanian teaches: the
the method further comprises generating a second
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have incorporated Balasubramanian’s pipelining process into the combination of Qiu, Rastegari, Croxford, and Hutel. A motivation for the combination is to accelerate CNN processing without increasing CPU frequency and without incurring more power consumption than conventional CNN processing (Balasubramanian, [0083], final 5 lines)
Response to Arguments
Applicant's arguments filed 01/28/2026 have been fully considered but they are not persuasive.
Applicant’s FIRST Arguments Under 35 U.S.C. 101 – Step 2A Prong Two: Pages 13-14 of the remarks argue that the features of training a machine learning model and using particular components of a hardware accelerator to reduce a number of parameters and reducing the number of operations reflects an improvement to the technical field of machine learning models. On page 15, the remarks argue that the improvements further illustrate an improvement to computer systems themselves.
Examiner’s Response: Applicant's arguments have been fully considered but they are not persuasive. Claim 1 as a whole is directed to performing a series of mathematical calculations on a convolution layer of a machine learning model, including convolution and training operations, and performing a judgement and evaluation mental process of determining a set of scaling factors. Specification paragraphs [0043]-[0044] and [0051]-[0061] disclose equations for generating a set of basis masks and generating a composite kernel. Specification paragraphs [0105]-[0106], [0109] disclose equations for training the machine learning model by optimizing a total loss function. This amounts to an optimization algorithm, which is a mathematical calculation and a mathematical concept. Specification paragraphs [0043]-[0044], [0051]-[0061] disclose equations for performing the convolution operation based on the composite kernel. Specification paragraph [0061] discloses mathematical concepts of reducing a number of parameters. Specification paragraph [0099] discloses an equation for the imposing of the structured property. In Step 2A Prong 2, the additional elements of using a VMU and an ESU of a hardware accelerator are recited in a generic manner as being implemented using generic tools, i.e. generic computer hardware components. The aspect of training is not a technical improvement, but rather a series of mathematical calculations and mathematical concepts.
Any improvements which may be recited in the claim, including those described in remarks page 14, lines 19-26, are merely improvements to a mental process, a mathematical calculation or a mathematical concept, which alone cannot provide the technical improvement. MPEP 2106.05(a) states, “It is important to note, the judicial exception alone cannot provide the improvement.” MPEP 2106.05(a), II. states, “it is important to keep in mind that an improvement in the abstract idea itself (e.g. a recited fundamental economic concept) is not an improvement in technology.”
In response to Applicant’s arguments on page 14, line 26 to page 15, line 8, the pending claims broadly recite performing mathematical calculations using a VMU and an ESU of the hardware accelerator at a high level. Again, any improvement which may be recited by the claim is an improvement to a mental process or a mathematical calculation, which alone cannot provide the technical improvement. The claim does not recite improvements to the hardware accelerator itself. The hardware accelerator, the VMU, and the ESU amount to generic computer hardware components for applying the abstract ideas on a computer. The claims do not explicitly show how these components reduce resource utilization and actually realize technical benefits in processing hardware beyond mere instructions to apply the abstract ideas on generic computer hardware components. In the remarks, the alleged improvement “(2) better utilization of a machine learning model for various computations” amounts to an improvement to an abstract idea. Increasing model efficiency by reducing the number of operations constitutes a mathematical calculation.
Applicant’s SECOND Arguments Under 35 U.S.C. 101 – Step 2A Prong Two: On page 15 of the remarks, Applicant argues that similar to the claims in Ex parte Desjardins, Applicant’s claims recite an improvement to machine learning technology such as model size reduction (e.g., by reducing parameter count) and increasing model computational efficiency (e.g., by reducing the number of operations). On page 16, Applicant argues each of the steps recited in Applicant’s claims relates directly to the training of a machine learning model. If reciting the steps involved in training a machine learning model in a manner that improves the operation of the model (as explained in Applicant's Specification) renders claims ineligible under Section 101, then any claim that recites training a machine learning model would be categorically excluded from patent eligibility. Thus, the rejection of Applicant’s claims under Section 101 is improper in light of Desjardins.
On pages 16-17, the remarks cite XY, LLC and argues that like this, Applicant’s claimed solution employs computations to achieve an improved result only when combined with the specific technical components recited in the claims.
Examiner’s Response: Applicant's arguments have been fully considered but they are not persuasive. Examiner agrees that pending claim 1 and Desjardins both relate to training a machine learning model. Pending claim 1 recites a series of mathematical calculations and mathematical concepts used in training, which alone cannot provide the technical improvement. The claim of Desjardin, when evaluated as a whole, provides a technical improvement to the problem of catastrophic forgetting. At least the limitation “adjust the first values of the plurality of parameters to optimize performance of the machine learning model on the second machine learning task while protecting performance of the machine learning model on the first machine learning task” reflected the improvement disclosed in the specification. Accordingly, the claims as a whole integrated what would otherwise be a judicial exception instead into a practical application at Step 2A Prong Two, and therefore the claims were deemed to be outside any specific, enumerated judicial exception (Step 2A: NO) (See MPEP 2106.04(d), subsection III). Pending claim 1 does not provide a technical solution to catastrophic forgetting, and pending claim 1 does not recite this same limitation which makes Desjardins eligible.
XY, LLC does not set a precedence and are not used for guidance in the 101 analysis, and the claims in this case recites different features from the pending claims.
Applicant’s Arguments Under 35 U.S.C. 101 – Step 2B: Pages 17-21 of the remarks argue that claimed features improve the technical field of machine learning model, and cites Cosmokey and BASCOM.
Examiner’s Response: Applicant's arguments have been fully considered but they are not persuasive. Any improvements which may be recited in the claim are merely improvements to a mental process, a mathematical calculation or a mathematical concept, which alone cannot provide the technical improvement, for the same reasons provided in the Examiner’s response to arguments under Step 2A Prong Two. The claim does not recite improvements to the hardware accelerator itself. The hardware accelerator, the VMU, and the ESU amount to generic computer hardware components for applying the abstract ideas on a computer.
Cosmokey and BASCOM do not set a precedence and is not used for guidance in the 101 analysis, and the claims in these cases recite different features from the pending claims.
The claims in BASCOM describe the concept of filtering content, and were found to recite a judicial exception (method of organizing human behavior) in Step 2A Prong 1. The claims were found to be patent eligible in Step 2B because the additional elements amounted to significantly more than the recited abstract idea. However, the pending claims recite different features including different additional elements from the claims in BASCOM. In the pending claims, the additional elements do not amount to significantly more than the recited abstract ideas.
Applicant’s Arguments Under 35 U.S.C. 103 (Pages 21-24): Applicant traverses the rejection of at least claims 1 and 6. On page 22, Applicant submits that the Non-Final Office Action has failed to adequately show that Qiu and Croxford teaches or suggests the limitations of pending claim 1, lines 6-21. Neither Qiu nor Croxford disclose a VMU of a hardware accelerator or an ESU of the hardware accelerator. Qiu and Croxford are silent regarding “a convolution with a Toeplitz matrix that is based on a kernel made of all ones.”
Examiner’s Response: Applicant's arguments have been fully considered but they are not persuasive. Qiu teaches: a sum-pooling operation,
However, Qiu does not explicitly teach: generating, by an efficient variable-length vector multiplication unit (VMU) of a hardware accelerator, a composite kernel
training the machine learning model, using the hardware accelerator
a sum-pooling operation, performed by an extract sum unit (ESU) of the hardware accelerator, comprising a convolution with a Toeplitz matrix that is based on a kernel made of all ones.
But Rastegari teaches: a
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have incorporated Rastegari’s binary filters and pooling layer from its XNOR-Network into Qiu. Since Rastegari’s binary filters comprise only binary weights, Qiu’s basis mask would be a binary mask in the combination. A motivation for the combination is that “The binary weight filters reduce memory usage by a factor of ∼ 32× compared to single-precision filters.” (Rastegari, P. 5, § 3.1, lines 2-4 below Equation (1))
However, Qiu and Rastegari do not explicitly teach: generating, by an efficient variable-length vector multiplication unit (VMU) of a hardware accelerator, a composite kernel
training the machine learning model, using the hardware accelerator
a sum-pooling operation, performed by an extract sum unit (ESU) of the hardware accelerator, comprising a convolution with a Toeplitz matrix
But Croxford teaches: generating, by an efficient variable-length vector multiplication unit (VMU) of a hardware accelerator, a
a
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have trained Qiu and Rastegari’s convolutional neural network using Croxford’s neural network accelerator, and it would have been obvious to have performed the sum-pooling operation using Croxford’s addition unit. A motivation for the combination is that a neural network accelerator is suited to the high power consumption of classification of an image. (Croxford, [0038]).
However, Qiu, Rastegari, and Croxford do not explicitly teach: Toeplitz
But Hutel teaches: Toeplitz ([0035], lines 1-5)
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have incorporated Hutel’s use of a Toeplitz matrix into the combination of Qiu, Rastegari, and Croxford. A motivation for the combination is to optimize convolution and/or reduce computation complexity (Hutel, [0035], lines 5-13). In the combination, Rastegari’s weight filter would contain all positive ones or all negative ones.
Applicant's arguments regarding a VMU of a hardware accelerator, an ESU of the hardware accelerator, and a Toeplitz matrix fail to comply with 37 CFR 1.111(b) because they amount to a general allegation that the claims define a patentable invention without specifically pointing out how the language of the claims patentably distinguishes them from the references. Applicant has not explained how the combination of Qiu, Rastegari, Croxford, and Hutel specifically fails to teach a VSU of a hardware accelerator, an ESU of the hardware accelerator, and a Toeplitz matrix as recited by the claims.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Asher H. Jablon whose telephone number is (571)270-7648. The examiner can normally be reached Monday - Friday, 9:00 am - 6:00 pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Abdullah Al Kawsar can be reached at (571)270-3169. 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.
/A.H.J./Examiner, Art Unit 2127
/ABDULLAH AL KAWSAR/Supervisory Patent Examiner, Art Unit 2127