DETAILED ACTION
This action is responsive to the Request for Continued Examination filed on 02/26/2026. Claims 1-20 are pending in the case. Claims 1, 12 and 18 are independent claims. Claims 1, 12 and 18 are amended.
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 02/26/2026 has been entered.
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 .
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.
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.
Claim 1-20 rejected under 35 U.S.C. 101 because the claims are directed to an abstract idea or mental process.
Regarding claim 1:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim recites setting a scaling parameter and a clipping parameter…the scaling parameter being different than the clipping parameter which, under the broadest reasonable interpretation, covers performance of the limitation in the mind with or without a physical aid. The limitations encompass a user using judgement to evaluation to assign values. See 2106.04.(a)(2).III.C.
The claim recites generate outputs based on input training data which, under the broadest reasonable interpretation, covers performance of the limitation in the mind with or without a physical aid. The limitations encompass a user creating/choosing values based on a set of data. See 2106.04.(a)(2).III.C.
The claim recites generating a loss function based on the outputs, the loss function including a term for the scaling parameter and a term for the clipping parameter which, under the broadest reasonable interpretation, covers performance of the limitation in the mind with or without a physical aid. The limitations encompass a user judging a set of values and creating a formula based off said judgment. See 2106.04.(a)(2).III.C.
The claim recites modifying at least one of the scaling parameter or the clipping parameter using the loss function which, under the broadest reasonable interpretation, covers performance of the limitation in the mind with or without a physical aid. The limitations encompass a user changing a variable using a function. See 2106.04.(a)(2).III.C.
Subject Matter Eligibility Analysis Step 2A Prong 2:
initializing a machine learning model(merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
for at least one activation layer, of a plurality of activation layers of the machine learning model(merely specifies a particular technological environment in which the abstract idea is to take place, i.e. a field of use (see MPEP 2106.05(h)))
generating a trained machine learning model(merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
causing the machine learning model to (merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
outputting the trained machine learning model (recites insignificant extra-solution activity of transmitting data (see MPEP 2106.05(g))
Subject Matter Eligibility Analysis Step 2B:
Additional elements (a), (c) and (d) do not integrate the abstract idea into a practical application nor do the additional limitation provide significantly more than the abstract idea because the limitation amount to no more than mere instructions to apply the exception using a generic computer component. Please see MPEP §2106.05(f).
Additional elements (b) do not integrate the abstract idea into a practical application nor do the additional limitation provide significantly more than the abstract idea because the limitation merely specifies a field of use in which the abstract idea is to take place, i.e. a field of use (see MPEP 2106.05(h)).
Further, additional element (e) obtaining a network input is well understood, routine, and conventional activity of “transmitting or receiving data over a network" (see MPEP 2106.05(d)(II)(i) using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 ).
The additional element(s) (a) (b) (c) (d) and (e) in the claim does not include any additional elements , when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor significantly more than the judicial exception for the reasons set forth in step 2A prong 2 analysis above. The claim is not patent eligible.
Regarding claim 2:
The rejection of claim 1 is incorporated and further claim recites further additional elements/limitations:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim does not contain elements that would warrant a Step 2A Prong 1 analysis.
Subject Matter Eligibility Analysis Step 2A Prong 2:
wherein the machine learning model comprises a neural network including a plurality of neurons, each of the plurality of neurons configured to produce an output using a numerical representation of eight or fewer bits (merely specifies a particular technological environment in which the abstract idea is to take place, i.e. a field of use (see MPEP 2106.05(h)))
Subject Matter Eligibility Analysis Step 2B:
Additional elements (a) do not integrate the abstract idea into a practical application nor do the additional limitation provide significantly more than the abstract idea because the limitation merely specifies a field of use in which the abstract idea is to take place, i.e. a field of use (see MPEP 2106.05(h)).
The additional element(s) (a) in the claim does not include any additional elements , when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor significantly more than the judicial exception for the reasons set forth in step 2A prong 2 analysis above. The claim is not patent eligible.
Regarding claim 3:
The rejection of claim 1 is incorporated and further claim recites further additional elements/limitations:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim does not contain elements that would warrant a Step 2A Prong 1 analysis.
Subject Matter Eligibility Analysis Step 2A Prong 2:
wherein the machine learning model comprises a neural network including a plurality of neurons, each of the plurality of neurons being associated with one of the plurality of activation layers and configured to produce an output that is processed by the one of the plurality of activation layers to activate another one of the plurality of neurons. (merely specifies a particular technological environment in which the abstract idea is to take place, i.e. a field of use (see MPEP 2106.05(h)))
Subject Matter Eligibility Analysis Step 2B:
Additional elements (a) do not integrate the abstract idea into a practical application nor do the additional limitation provide significantly more than the abstract idea because the limitation merely specifies a field of use in which the abstract idea is to take place, i.e. a field of use (see MPEP 2106.05(h)).
The additional element(s) (a) in the claim does not include any additional elements , when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor significantly more than the judicial exception for the reasons set forth in step 2A prong 2 analysis above. The claim is not patent eligible.
Regarding claim 4:
The rejection of claim 1 is incorporated and further claim recites further additional elements/limitations:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim recites wherein the scaling parameter defines a degree by which a numerical value within a range of numerical values is to be amplified relative to zero which is an abstract idea (Mathematical Relationships (see MPEP 2106.04(a)(2)(I)(A)))).
Subject Matter Eligibility Analysis Step 2A Prong 2:
The claim does not contain elements that would warrant a Step 2A Prong 2 analysis.
Subject Matter Eligibility Analysis Step 2B:
The claim does not include any additional element, when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor to significantly more than the judicial exception. The claim is not patent eligible.
Regarding claim 5:
The rejection of claim 4 is incorporated and further claim recites further additional elements/limitations:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim recites wherein the scaling parameter causes linear amplification of the range of numerical values relative to zero which is an abstract idea (Mathematical Relationships (see MPEP 2106.04(a)(2)(I)(A)))).
Subject Matter Eligibility Analysis Step 2A Prong 2:
The claim does not contain elements that would warrant a Step 2A Prong 2 analysis.
Subject Matter Eligibility Analysis Step 2B:
The claim does not include any additional element, when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor to significantly more than the judicial exception. The claim is not patent eligible.
Regarding claim 6:
The rejection of claim 4 is incorporated and further claim recites further additional elements/limitations:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim recites wherein the scaling parameter causes nonlinear amplification of the range of numerical values relative to zero which is an abstract idea (Mathematical Relationships (see MPEP 2106.04(a)(2)(I)(A)))).
Subject Matter Eligibility Analysis Step 2A Prong 2:
The claim does not contain elements that would warrant a Step 2A Prong 2 analysis.
Subject Matter Eligibility Analysis Step 2B:
The claim does not include any additional element, when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor to significantly more than the judicial exception. The claim is not patent eligible.
Regarding claim 7:
The rejection of claim 1 is incorporated and further claim recites further additional elements/limitations:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim recites wherein the clipping parameter defines a threshold numerical value and causes numerical values … that satisfy the threshold numerical value to be expressed as the threshold numerical value which, under the broadest reasonable interpretation, covers performance of the limitation in the mind with or without a physical aid. The limitations encompass a user seeing whether a value is higher than a threshold value and choosing to use the threshold value. See 2106.04.(a)(2).III.C.
Subject Matter Eligibility Analysis Step 2A Prong 2:
output by the plurality of activation layers(merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
Subject Matter Eligibility Analysis Step 2B:
Additional elements (a) do not integrate the abstract idea into a practical application nor do the additional limitation provide significantly more than the abstract idea because the limitation amount to no more than mere instructions to apply the exception using a generic computer component. Please see MPEP §2106.05(f).
The additional element(s) (a) in the claim does not include any additional elements , when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor significantly more than the judicial exception for the reasons set forth in step 2A prong 2 analysis above. The claim is not patent eligible.
Regarding claim 8:
The rejection of claim 1 is incorporated and further claim recites further additional elements/limitations:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim recites generate the outputs based on the input training data which, under the broadest reasonable interpretation, covers performance of the limitation in the mind with or without a physical aid. The limitations encompass a user creating/choosing values based on a set of data. See 2106.04.(a)(2).III.C.
The claim recites the generating the loss function based on the outputs which, under the broadest reasonable interpretation, covers performance of the limitation in the mind with or without a physical aid. The limitations encompass a user judging a set of values and creating a formula based off said judgment. See 2106.04.(a)(2).III.C.
The claim recites modifying the at least one of the scaling parameter or the clipping parameter using the loss function which, under the broadest reasonable interpretation, covers performance of the limitation in the mind with or without a physical aid. The limitations encompass a user changing a variable using a function. See 2106.04.(a)(2).III.C.
Subject Matter Eligibility Analysis Step 2A Prong 2:
wherein training the machine learning model is performed over a plurality of training iterations(merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
and comprises performing the causing the machine learning model to… during each of the plurality of training iteration(merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
Subject Matter Eligibility Analysis Step 2B:
Additional elements (a) and (b) do not integrate the abstract idea into a practical application nor do the additional limitation provide significantly more than the abstract idea because the limitation amount to no more than mere instructions to apply the exception using a generic computer component. Please see MPEP §2106.05(f).
The additional element(s) (a) and (b) in the claim does not include any additional elements , when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor significantly more than the judicial exception for the reasons set forth in step 2A prong 2 analysis above. The claim is not patent eligible.
Regarding claim 9:
The rejection of claim 1 is incorporated and further claim recites further additional elements/limitations:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim recites determining that the outputs generated during training satisfy a threshold difference from ground truth information for the input training data which, under the broadest reasonable interpretation, covers performance of the limitation in the mind with or without a physical aid. The limitations encompass a user evaluating and judging whether a value is satisfactory and choosing an action based on the evaluation. See 2106.04.(a)(2).III.C.
Subject Matter Eligibility Analysis Step 2A Prong 2:
wherein outputting the machine learning model is performed responsive to(merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
Subject Matter Eligibility Analysis Step 2B:
Additional element (a) do not integrate the abstract idea into a practical application nor do the additional limitation provide significantly more than the abstract idea because the limitation amount to no more than mere instructions to apply the exception using a generic computer component. Please see MPEP §2106.05(f).
The additional element(s) (a) in the claim does not include any additional elements , when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor significantly more than the judicial exception for the reasons set forth in step 2A prong 2 analysis above. The claim is not patent eligible
Regarding claim 10:
The rejection of claim 1 is incorporated and further claim recites further additional elements/limitations:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim recites wherein generating the loss function comprises comparing the outputs to ground truth information for the input training data which is an abstract idea (Mathematical Relationships (see MPEP 2106.04(a)(2)(I)(A)))).
Subject Matter Eligibility Analysis Step 2A Prong 2:
The claim does not contain elements that would warrant a Step 2A Prong 2 analysis.
Subject Matter Eligibility Analysis Step 2B:
The claim does not include any additional element, when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor to significantly more than the judicial exception. The claim is not patent eligible.
Regarding claim 11:
The rejection of claim 1 is incorporated and further claim recites further additional elements/limitations:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim does not contain elements that would warrant a Step 2A Prong 1 analysis.
Subject Matter Eligibility Analysis Step 2A Prong 2:
further comprising producing an output that classifies one or more features of input data by providing the input data as input to the trained machine learning model(merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
Subject Matter Eligibility Analysis Step 2B:
Additional element (a) do not integrate the abstract idea into a practical application nor do the additional limitation provide significantly more than the abstract idea because the limitation amount to no more than mere instructions to apply the exception using a generic computer component. Please see MPEP §2106.05(f).
The additional element(s) (a) in the claim does not include any additional elements , when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor significantly more than the judicial exception for the reasons set forth in step 2A prong 2 analysis above. The claim is not patent eligible
Regarding claim 12:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim recites to produce an output that classifies one or more features of input data which, under the broadest reasonable interpretation, covers performance of the limitation in the mind with or without a physical aid. The limitations encompass a user creating/choosing values based on a set of data. See 2106.04.(a)(2).III.C.
The claim recites that processes a numerical value output…using a scaling parameter and a clipping parameter, the scaling parameter being different than the clipping parameter which is an abstract idea (Mathematical Calculations (see MPEP 2106.04(a)(2)(I)(C))).
The claim recites processing the numerical value output by… using the scaling parameter and the clipping parameter which, under the broadest reasonable interpretation, covers performance of the limitation in the mind with or without a physical aid. The limitations encompass a user using judgement to evaluation to assign values. See 2106.04.(a)(2).III.C. Alternatively, with the broadest reasonable interpretation the claim can also be categorized as Mathematical Calculations (see MPEP 2106.04(a)(2)(I)(C)).
Subject Matter Eligibility Analysis Step 2A Prong 2:
obtaining a machine learning model that includes a neural network comprising a plurality of neurons (recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g))
by inputting the input data to the machine learning model(merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
the one of the plurality of neurons(merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
at least one of the plurality of neurons being associated with an activation layer (merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
causing the neural network to produce an output that classifies one or more features of input data (merely specifies a particular technological environment in which the abstract idea is to take place, i.e. a field of use (see MPEP 2106.05(h)))
the output that classifies the one or more features of input data being generated based on a result of the activation layer(merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
Subject Matter Eligibility Analysis Step 2B:
Further, additional element (a) obtaining a network input is well understood, routine, and conventional activity of “transmitting or receiving data over a network" (see MPEP 2106.05(d)(II)(i) using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 ).
Additional elements (b), (c), (d) and (f) do not integrate the abstract idea into a practical application nor do the additional limitation provide significantly more than the abstract idea because the limitation amount to no more than mere instructions to apply the exception using a generic computer component. Please see MPEP §2106.05(f).
Additional elements (e) do not integrate the abstract idea into a practical application nor do the additional limitation provide significantly more than the abstract idea because the limitation merely specifies a field of use in which the abstract idea is to take place, i.e. a field of use (see MPEP 2106.05(h)).
The additional element(s) (a) (b) (c) (d) (e) and (f) in the claim does not include any additional elements , when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor significantly more than the judicial exception for the reasons set forth in step 2A prong 2 analysis above. The claim is not patent eligible.
Regarding claim 13:
Regarding claim 13, The rejection of claim 12 incorporated in claim 13, and further, claim 13 is rejected under the same 101 rationale as set forth in the rejection of claim 4.
Regarding claim 14:
Regarding claim 14, The rejection of claim 13 incorporated in claim 14, and further, claim 14 is rejected under the same 101 rationale as set forth in the rejection of claim 5.
Regarding claim 15:
Regarding claim 15, The rejection of claim 13 incorporated in claim 15, and further, claim 15 is rejected under the same 101 rationale as set forth in the rejection of claim 6.
Regarding claim 16:
Regarding claim 16, The rejection of claim 12 incorporated in claim 16, and further, claim 16 is rejected under the same 101 rationale as set forth in the rejection of claim 7
Regarding claim 17:
Regarding claim 17, The rejection of claim 12 incorporated in claim 17, and further, claim 17 is rejected under the same 101 rationale as set forth in the rejection of claim 2
Regarding claim 18:
Subject Matter Eligibility Analysis Step 2A Prong 1:
The claim recites assigning a scaling parameter and a clipping parameter…the scaling parameter being different than the clipping parameter which, under the broadest reasonable interpretation, covers performance of the limitation in the mind with or without a physical aid. The limitations encompass a user using judgement to evaluation to assign values. See 2106.04.(a)(2).III.C.
The claim recites generate training outputs based on input training data which, under the broadest reasonable interpretation, covers performance of the limitation in the mind with or without a physical aid. The limitations encompass a user creating/choosing values based on a set of data. See 2106.04.(a)(2).III.C.
The claim recites generating a loss function based on the training outputs, the loss function including a term for the scaling parameter and a term for the clipping parameter which, under the broadest reasonable interpretation, covers performance of the limitation in the mind with or without a physical aid. The limitations encompass a user judging a set of values and creating a formula based off said judgment. See 2106.04.(a)(2).III.C.
The claim recites modifying at least one of the scaling parameter or the clipping parameter using the loss function which, under the broadest reasonable interpretation, covers performance of the limitation in the mind with or without a physical aid. The limitations encompass a user changing a variable using a function. See 2106.04.(a)(2).III.C.
Subject Matter Eligibility Analysis Step 2A Prong 2:
an initialization circuitry to initialize a machine learning model (merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
an activation layer of the machine learning model(merely specifies a particular technological environment in which the abstract idea is to take place, i.e. a field of use (see MPEP 2106.05(h)))
training circuitry to generate a trained machine learning model by(merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
causing the machine learning model to (merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
prediction circuitry to generate an output(recites insignificant extra-solution activity of data gathering (see MPEP 2106.05(g))
processing input data using the trained machine learning model(merely recites a generic computer on which to perform the abstract idea, e.g. "apply it on a computer" (see MPEP 2106.05(f)))
Subject Matter Eligibility Analysis Step 2B:
Additional elements (a), (c), (d) and (f) do not integrate the abstract idea into a practical application nor do the additional limitation provide significantly more than the abstract idea because the limitation amount to no more than mere instructions to apply the exception using a generic computer component. Please see MPEP §2106.05(f).
Additional elements (b) do not integrate the abstract idea into a practical application nor do the additional limitation provide significantly more than the abstract idea because the limitation merely specifies a field of use in which the abstract idea is to take place, i.e. a field of use (see MPEP 2106.05(h)).
Further, additional element (e) obtaining a network input is well understood, routine, and conventional activity of “transmitting or receiving data over a network" (see MPEP 2106.05(d)(II)(i) using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 ).
The additional element(s) (a) (b) (c) (d) (e) and (f) in the claim does not include any additional elements , when considered separately and in combination, that amount to an integration of the judicial exception into a practical application, nor significantly more than the judicial exception for the reasons set forth in step 2A prong 2 analysis above. The claim is not patent eligible.
Regarding claim 19:
Regarding claim 19, The rejection of claim 18 incorporated in claim 19, and further, claim 19 is rejected under the same 101 rationale as set forth in the rejection of claim 2.
Regarding claim 20:
Regarding claim 20, The rejection of claim 18 incorporated in claim 20, and further, claim 20 is rejected under the same 101 rationale as set forth in the rejection of claim 7.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claim(s) 1-5, 7-14 and 16-20 is/are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Choi et al.(“PACT: PARAMETERIZED CLIPPING ACTIVATION FOR QUANTIZED NEURAL NETWORKS”, henceforth known as Choi)
Regarding claim 1:
Choi discloses initializing a machine learning model by setting a scaling parameter and a clipping parameter for at least one activation layer, of a plurality of activation layers, of the machine learning model, the scaling parameter being different than the clipping parameter(Choi, Page 2, Paragraph 2, “We introduce a new parameter α that is used to represent the clipping level in the activation function and is learned via back-propagation. α sets the quantization scale smaller than ReLU to reduce the quantization error, but larger than a conventional clipping activation function” where α being a parameter that is used to find the scale of the quantization scale is considered a parameter and Choi, Page 4, Equation 2,
PNG
media_image1.png
57
274
media_image1.png
Greyscale
, where ((2k-1)/α) corresponds a scaling parameter as it scales the y value)
Choi discloses generating a trained machine learning model(Choi, Page 13, Paragraph 1, “For PACT experiments, we only replace ReLU into PACT but the same hyper-parameters are used. All the time the networks are trained from scratch.”)
Choi discloses causing the machine learning model to generate outputs based on input training data(Choi, Page 13, Paragraph 2, “The CIFAR10 dataset (Krizhevsky & Hinton (2010)) is an image classification benchmark containing 32 × 32 pixel RGB images” where one of the CNN’s used to test is CIFAR10 which classifies images based on input which is considered generated predicted outputs based on input training data(See also: Choi, Page 4, Paragraph 3, “Fig. 2 illustrates how the value of α changes during full-precision training of CIFAR10-ResNet20 starting with an initial value of 10 and using the L2-regularizer”))
Choi discloses generating a loss function based on the outputs(Choi, page 5, Paragraph 2, “Finally, in Fig. 3c, we show the total training loss including both the cross-entropy discussed above and the cost from α regularization” where cross-entropy is the measure between the predicted result and the true label)
Choi discloses the loss function including a term for the scaling parameter and a term for the clipping parameter(Choi, Page 4, Paragraphs 2, “With this new activation function, α is a variable in the loss function, whose value can be optimized during training” where α is a term for clipping and α is a term for the scaling parameter)
Choi discloses modifying at least one of the scaling parameter or the clipping parameter using the loss function(Choi, Page 3, Paragraphs 5, “α is dynamically adjusted via gradient descent-based training with the objective of minimizing the accuracy degradation arising from quantization.”
Choi discloses outputting the trained machine learning model(Choi, Page 7, Paragraph 3, “As can be seen, with 4 bit precision for both weights and activation, PACT achieves full-precision accuracy consistently across the networks tested” where the full-precision and partial precision being used to inference after training converges is considered outputting the trained machine learning model)
Regarding claim 2:
The rejection of claim 1 with prior art Choi is incorporated and further:
Choi discloses wherein the machine learning model comprises a neural network including a plurality of neurons, each of the plurality of neurons configured to produce an output using a numerical representation of eight or fewer bits(Choi, Page 7, Paragraph 3, “As can be seen, with 4 bit precision for both weights and activation, PACT achieves full-precision accuracy consistently across the networks tested.”)
Regarding claim 3:
The rejection of claim 1 with prior art Choi is incorporated and further:
Choi discloses wherein the machine learning model comprises a neural network including a plurality of neurons(Choi, Page 7, Paragraph 3, “Empirically, we show that: (a) for extremely low bit-precision (≤ 2-bits for weights and activations), PACT achieves the highest model accuracy compared to all published schemes and (b) 4-bit quantized CNNs based on PACT achieve accuracies similar to single-precision floating point representations” where a CNN is considered a neural network that includes neurons, layers and produces an output), each of the plurality of neurons being associated with one of the plurality of activation layers and configured to produce an output(Choi, Page 7, Paragraph 3, “Empirically, we show that: (a) for extremely low bit-precision (≤ 2-bits for weights and activations), PACT achieves the highest model accuracy compared to all published schemes and (b) 4-bit quantized CNNs based on PACT achieve accuracies similar to single-precision floating point representations” where a CNN is considered a neural network that includes neurons, layers and produces an output) that is processed by the one of the plurality of activation layers to activate another one of the plurality of neurons(Choi, Page 3, Equation 1, where the paper defines y = PACT(x), where x is the pre-activation neuron output(See also, Choi, Page 5, Figure 3, where Figure 3 shows multiple layers with act0 to act6 layers))
Regarding claim 4:
The rejection of claim 1 with prior art Choi is incorporated and further:
Choi discloses wherein the scaling parameter defines a degree by which a numerical value within a range of numerical values is to be amplified relative to zero”(Choi, Page 4, Equation 2,
PNG
media_image1.png
57
274
media_image1.png
Greyscale
, where ((2k-1)/α) amplifying y corresponds to amplifying relative to zero a numerical value within a range of numerical values as both y and ((2k-1)/α) have numerical ranges .)
Regarding claim 5:
The rejection of claim 4 with prior art Choi is incorporated and further:
Choi discloses wherein the scaling parameter causes linear amplification of the range of numerical values relative to zero(Choi, Page 4, Equation 2,
PNG
media_image1.png
57
274
media_image1.png
Greyscale
, where ((2k-1)/α) corresponds to linear with respect to y and causes amplification relative to zero.)
Regarding claim 7:
The rejection of claim 1 with prior art Choi is incorporated and further:
Choi discloses wherein the clipping parameter defines a threshold numerical value and causes numerical values output by the plurality of activation layers that satisfy the threshold numerical value to be expressed as the threshold numerical value(Choi, Page 4, Equation 1,
PNG
media_image2.png
94
535
media_image2.png
Greyscale
, where the activation function with clipping parameter α limiting the value between 0 and α corresponds clipping parameter defining a threshold numerical value and causes numerical values output by the plurality of activation layers that satisfy the threshold numerical value to be expressed as the threshold numerical value(See also Choi, Page 4, Paragraph 1, “where α limits the range of activation to [0,α]”))
Regarding claim 8:
The rejection of claim 1 with prior art Choi is incorporated and further:
Choi discloses wherein training the machine learning model is performed over a plurality of training iterations and comprises performing the causing the machine learning model to generate the outputs based on the input training data(Choi, Page 4, Paragraph 3, “It can be observed that α converges to values much smaller than the initial value as the training epochs proceed, thereby limiting the dynamic range of activations and minimizing quantization loss” where training epochs are considered a plurality of training iterations that generate output based on training data(See also: Choi, page 4, Figure 2)), the generating the loss function based on the outputs(Choi, page 5, Paragraph 2, “Finally, in Fig. 3c, we show the total training loss including both the cross-entropy discussed above and the cost from α regularization” where cross-entropy is the measure between the predicted result(predicted output) and the true label), and the modifying the at least one of the scaling parameter or the clipping parameter using the loss function during each of the plurality of training iterations(Choi, Page 4, Paragraphs 2, “With this new activation function, α is a variable in the loss function, whose value can be optimized during training. For back-propagation, gradient ∂yq/∂α can be computed using the Straight-Through Estimator (STE)” where α being updated using gradients from the activation function is considered modifying clipping using the loss function on every iteration (See also: Choi, Page 2, Paragraphs 2, “In addition, regularization is applied to α in the loss function to enable faster convergence”))
Regarding claim 9:
The rejection of claim 1 with prior art Choi is incorporated and further:
Choi discloses wherein outputting the machine learning model is performed responsive to determining that the outputs generated during training(Choi, Page 3, Paragraph 5, “α is dynamically adjusted via gradient descent-based training with the objective of minimizing the accuracy degradation arising from quantization”) satisfy a threshold difference from ground truth information(See Choi, page 5, Paragraph 2, “Finally, in Fig. 3c, we show the total training loss including both the cross-entropy” where cross-entropy is considered measuring difference between a ground truth (actual labels) and a predicted probability distribution of a model) for the input training data(Choi, Page 7, Paragraph 1, “The final validation error has less than 1% difference relative to the full-precision validation error for all cases when the activation bit-precision is at least 4-bits” where the threshold is within 1% of full-precision accuracy is considered a threshold)
Regarding claim 10:
The rejection of claim 1 with prior art Choi is incorporated and further:
Choi discloses wherein generating the loss function comprises comparing the outputs to ground truth information for the input training data(Choi, page 5, Paragraph 2, “Finally, in Fig. 3c, we show the total training loss including both the cross-entropy discussed above and the cost from α regularization” where the cross entropy is considered comparing the predicted outputs to ground truth information as it quantifies the difference between the ground truth (actual labels) and a predicted probability distribution of a model)
Regarding claim 11:
The rejection of claim 1 with prior art Choi is incorporated and further:
Choi discloses further comprising producing an output that classifies one or more features of input data by providing the input data as input to the trained machine learning model(Choi, page 7, Figure 5, where the use of PACT for image classification models is considered producing an output that classifies one or more features of input data by providing input data as input to the trained learning model)
Regarding claim 12:
Choi discloses accessing a machine learning model that includes a neural network comprising a plurality of neurons, at least one of the plurality of neurons being associated with an activation layer(Choi, Page 3, Equation 1, where the paper defines y = PACT(x), where x is the pre-activation neuron output and part of a machine learning model being accessed(See also, Choi, Page 5, Figure 3, where Figure 3 shows multiple layers with act0 to act6 layers)) that processes a numerical value output by the one of the plurality of neurons(Choi , Page 3, Equation 1, where the PACT(x) function clipping activation output of x is considered processing a numerical value output of neurons(See also: Page 4, Paragraph 1, “where α limits the range of activation to [0, α]”)) using a scaling parameter and a clipping parameter, the scaling parameter being different than the clipping parameter(Choi, Page 2, Paragraph 2, “We introduce a new parameter α that is used to represent the clipping level in the activation function and is learned via back-propagation. α sets the quantization scale smaller than ReLU to reduce the quantization error, but larger than a conventional clipping activation function” where α being a parameter that is used to find the scale of the quantization scale is considered a parameter and Choi, Page 4, Equation 2,
PNG
media_image1.png
57
274
media_image1.png
Greyscale
, where ((2k-1)/α) corresponds a scaling parameter as it scales the y value)
Choi discloses causing the neural network to produce an output that classifies one or more features of input data by inputting the input data to the machine learning model(Choi, page 7, Figure 5, where the use of PACT for image classification models is considered producing an output that classifies one or more features of input data by providing input data as input to the trained learning model), the output that classifies the one or more features of input data being generated based on a result of the activation layer processing the numerical value output by the one of the plurality of neurons using the scaling parameter and the clipping parameter(Choi, equation 1(Page 3) and equation 2(Page 4), where activation y uses α and where quantization using α in Equation 2 directly influences network output).
Regarding claim 13:
Regarding claim 13, The rejection of claim 12 incorporated in claim 13, and further, claim 13 is rejected under the same rationale as set forth in the rejection of claim 4.
Regarding claim 14:
Regarding claim 14, The rejection of claim 13 incorporated in claim 14, and further, claim 14 is rejected under the same rationale as set forth in the rejection of claim 5.
Regarding claim 16:
Regarding claim 16, The rejection of claim 12 incorporated in claim 16, and further, claim 16 is rejected under the same rationale as set forth in the rejection of claim 7
Regarding claim 17:
Regarding claim 17, The rejection of claim 12 incorporated in claim 17, and further, claim 17 is rejected under the same rationale as set forth in the rejection of claim 2.
Regarding claim 18:
Choi discloses an initialization circuitry(Choi, Page 8, Paragraph 2, “First, using real hardware implementations in a state of the art technology (14 nm CMOS), we accurately estimate the reduction in the MAC area achieved by aggressively scaling bit precision” where the initialization module is interpreted as a generic computer/processor and real hardware implementations is considered an initialization module) to initialize a machine learning model by assigning a scaling parameter and a clipping parameter to an activation layer of the machine learning model, the scaling parameter being different than the clipping parameter(Choi, Page 2, Paragraph 2, “We introduce a new parameter α that is used to represent the clipping level in the activation function and is learned via back-propagation. α sets the quantization scale smaller than ReLU to reduce the quantization error, but larger than a conventional clipping activation function” where α being a parameter that is used to find the scale of the quantization scale is considered a parameter and Choi, Page 4, Equation 2,
PNG
media_image1.png
57
274
media_image1.png
Greyscale
, where ((2k-1)/α) corresponds a scaling parameter as it scales the y value)
Choi discloses training circuitry to generate a trained machine learning model by: causing the machine learning model to generate training outputs based on input training data(Choi, Page 13, Paragraph 2, “The CIFAR10 dataset (Krizhevsky & Hinton (2010)) is an image classification benchmark containing 32 × 32 pixel RGB images” where one of the CNN’s used to test is CIFAR10 which classifies images based on input which is considered generated predicted outputs based on input training data(See also: Choi, Page 4, Paragraph 3, “Fig. 2 illustrates how the value of α changes during full-precision training of CIFAR10-ResNet20 starting with an initial value of 10 and using the L2-regularizer”))
Choi discloses generating a loss function based on the training outputs(Choi, page 5, Paragraph 2, “Finally, in Fig. 3c, we show the total training loss including both the cross-entropy discussed above and the cost from α regularization” where cross-entropy is the measure between the predicted result and the true label), the loss function including a term for the scaling parameter and a different term for the clipping parameter(Choi, Page 4, Paragraphs 2, “With this new activation function, α is a variable in the loss function, whose value can be optimized during training” where α is a term for clipping and α is a term for the scaling parameter)
Choi discloses modifying at least one of the scaling parameter or the clipping parameter using the loss function(Choi, Page 3, Paragraphs 5, “α is dynamically adjusted via gradient descent-based training with the objective of minimizing the accuracy degradation arising from quantization.”) and prediction circuitry to generate an output by processing input data using the trained machine learning model(Choi, Page 7, Paragraph 3, “As can be seen, with 4 bit precision for both weights and activation, PACT achieves full-precision accuracy consistently across the networks tested” where the full-precision and partial precision being used to inference after training converges is considered outputting the trained machine learning model)
Regarding claim 19:
Regarding claim 19, The rejection of claim 18 incorporated in claim 19, and further, claim 19 is rejected under the same rationale as set forth in the rejection of claim 2.
Regarding claim 20:
Regarding claim 20, The rejection of claim 18 incorporated in claim 20, and further, claim 20 is rejected under the same rationale as set forth in the rejection of claim 7.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
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.
Claim(s) 6 and 15 is/are rejected under 35 U.S.C. 103 as being unpatentable over Choi et al.(“PACT: PARAMETERIZED CLIPPING ACTIVATION FOR QUANTIZED NEURAL NETWORKS”, henceforth known as Choi) and further in view of Gong et al.(“Differentiable Soft Quantization:Bridging Full-Precision and Low-Bit Neural Networks” henceforth known as Gong)
Regarding claim 6:
The rejection of claim 4 with prior art Choi is incorporated and further:
Choi does not teach, however Gong discloses wherein the scaling parameter causes nonlinear amplification of the range of numerical values relative to zero(Gong, Page 4854, Equation 3 and Equation 4, where scaling parameter s with tanh corresponds to non-linear amplification relative to zero(See also, Gong, Page 4854, Col. 1, Paragraph 5, “The scaling parameter s guarantees that tanh functions of ϕ for the adjacent intervals can be smoothly connected (see Figure 2(a)).”))
Regarding claim 15:
Regarding claim 15, The rejection of claim 13 incorporated in claim 15, and further, claim 15 is rejected under the same rationale as set forth in the rejection of claim 6.
Response to Arguments
Applicant's arguments filed 2/26/2026 have been fully considered but they are not persuasive. A breakdown of the arguments can be found below.
101:
Applicant appears to argue Prong One on pages 9-10 that the limitations setting a scaling parameter and a clipping parameter for at least one activation layer, of a plurality of activation layers, of the machine learning model, the scaling parameter being different than the clipping parameter and generating a loss function based on the outputs, the loss function including a term for the scaling parameter and a term for the clipping parameter are not mental as they involve machine learning model initialization and training process that are performed by computers with automated learning involving mathematical computations that are performed by a machine learning system with mathematical calculations that cannot be performed in the human mind.
Examiner respectfully disagrees as the limitation is focused on setting a parameter, which Examiner finds to be a mental process as there are no technical details of assigning the parameters. The parameter being associated with an activation layer of a machine learning model is merely specifies the particular technological environment in which the abstract idea is to take place. For similar rational “generating a loss function based on the outputs, the loss function including a term for the scaling parameter and a term for the clipping” is considered as mental/abstract idea as there is no architecture or technological details for the generation. Additionally, the claims fail to integrate into “significantly more” as claims that require a computer may still recite a mental process (please see MPEP 2106.04(a)(2).III.C) and complexity of the computations is not taken into account and are categized into mathematical concepts grouping of abstract ideas(See Flook, 437 U.S. at 591-92, 198 USPQ2d at 198 (“the novelty of the mathematical algorithm is not a determining factor at all”)).
Applicant appears to argue Prong Two on pages 10-12 that claims provide an improvement to technology within the field of training a machine learning model by implementing reduced precision data to generate output of equal effective precision while significantly reducing a computational cost and training time required to generate the machine learning model.
Examiner respectfully disagrees as the claims as presented do not possess additional elements that result or highlight an improvement in neural networks or hardware processors. Examiner notes MPEP 2106.05(a) which provides the requirements for how an improvement to the functioning of a computer or to any other technology or technical field is evaluated. Further claims that require a computer may still recite a mental process (please see MPEP 2106.04(a)(2).III.C) and "claiming the improved speed or efficiency inherent with applying the abstract idea on a computer" does not integrate a judicial exception into a practical application or provide an inventive concept (MPEP 2106.05(f)(2)). Examiner understands the practical implementation that has been described as the neural networks and computers being used as tools to perform("apply it on a computer" (see MPEP 2106.05(f))) abstract ideas, which does not meet the MPEP requirement for practical application. At best Applicant’s example describes in an improvement provided by the claimed abstract idea of setting a scaling parameter and a clipping parameter and generating a loss function based on the outputs.
Applicant appears to argue Step 2B on Page 13 that the features in claim 1 are not well-understood, routine or conventional and the claimed combination of specific features recite an inventive concept.
Examiner respectfully disagrees as Applicant does not highlight a specific additional feature that was found to be well understood, routine and conventional. The only limitation that Examiner found to be well understood, routine and conventional(“transmitting or receiving data over a network") was outputting the trained machine learning model and Examiner maintains it is well understood, routine and conventional.
102/103:
Applicant appears to argue on pages 14-17 Choi does not teach or suggest in any way of incorporating a scaling parameter that is different from Choi’s clipping parameter. Additionally, Applicant appears to argue due to this lack of incorporating a scaling parameter it would be impossible for Choi to teach or suggest the claims or have a motivation to combine to Jin.
Examiner respectfully disagrees as Choi, Page 2, Paragraph 2, “We introduce a new parameter α that is used to represent the clipping level in the activation function” discusses a clipping parameter and Choi, Page 4, Equation 2,
PNG
media_image1.png
57
274
media_image1.png
Greyscale
uses ((2k-1)/α) to scale the y value. As the clipping parameter α and scaling parameter ((2k-1)/α) are not the same, Examiner comes to the conclusion they are different parameters despite α being a term in both. Additionally, Applicant’s arguments with respect to claim(s) 1-20 with respect to Jin have been considered but are moot because the new ground of rejection does not rely on Jin applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to CHARLES JEFFREY JONES JR whose telephone number is (703)756-1414. The examiner can normally be reached Monday - Friday 8:00 - 5:00 EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Kakali Chaki can be reached at 571-272-3719. 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.
/C.J.J./Examiner, Art Unit 2122
/KAKALI CHAKI/Supervisory Patent Examiner, Art Unit 2122
Prosecution Insights