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 .
DETAILED ACTION
This action is responsive to the Application filed on 05/09/2024
Claim(s) 1-30 are pending in the case. Claim(s) 1, 13, 22 and 30 are independent claims.
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(s) 1-30 are rejected under 35 U.S.C. § 101 because the claimed invention is directed to an abstract idea without significantly more.
When considering subject matter eligibility under 35 U.S.C. 101, it must be determined whether the claim is directed to one of the four statutory categories of invention, i.e., process, machine, manufacture, or composition of matter (Step 1). If the claim does fall within one of the statutory categories, the second step in the analysis is to determine whether the claim is directed to a judicial exception (Step 2A). The Step 2A analysis is broken into two prongs. In the first prong (Step 2A, Prong 1), it is determined whether or not the claims recite a judicial exception (e.g., mathematical concepts, mental processes, certain methods of organizing human activity). If it is determined in Step 2A, Prong 1 that the claims recite a judicial exception, the analysis proceeds to the second prong (Step 2A, Prong 2), where it is determined whether or not the claims integrate the judicial exception into a practical application. If it is determined at step 2A, Prong 2 that the claims do not integrate the judicial exception into a practical application, the analysis proceeds to determining whether the claim is a patent-eligible application of the exception (Step 2B). If an abstract idea is present in the claim, any element or combination of elements in the claim must be sufficient to ensure that the claim integrates the judicial exception into a practical application, or else amounts to significantly more than the abstract idea itself. Applicant is advised to consult the 2019 PEG for more details of the analysis.
Step 1 Analysis: Is the claim to a process, machine, manufacture or composition of matter? See MPEP § 2106.03.
Claim(s) 1-12 are drawn to a method, claims 13-21 and 30 are drawn to a processing system and claims 22-29 are drawn to a non-transitory computer-readable medium, therefore each of these claim groups falls under one of four categories of statutory subject matter (machine/products/apparatus, process/method, manufactures and compositions of mater; Step 1). Nonetheless, the claims are directed to a judicially recognized exception of an abstract idea without significant more (Step 2A, see below). Independent claims 1, 10 and 17 are nonverbatim but similar in claim construction, hence share the same rationale that the claimed inventions are directed to non-statutory subject matter as follows:
Regarding claim 1:
Claim 1 recites: A computer-implemented method performed by a training system while training a machine learning model, comprising:
identifying oscillation of a parameter of the machine learning model during quantization-aware training of the machine learning model; and
applying an oscillation mitigation procedure during the quantization- aware training of the machine learning model in response to identifying the oscillation, the oscillation mitigation procedure comprising at least one of oscillation dampening or parameter freezing
Step 2A Prong One Analysis: Does the claim recite an abstract idea, law of nature, or natural phenomenon? See MPEP § 2106.04(II)(A)(1).
Claim 1 is directed to an abstract idea, specifically, a mental process – concepts performed in the human mind or by a human using a pen and paper (including an observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2)(III).
Independent claim 1 recites in part:
“identifying oscillation of a parameter of the machine learning model during quantization-aware training of the machine learning model”
The limitation above is broadly and reasonably interpreted as a mental process, as a form or mental evaluation or judgement. It encompasses observing or evaluating successive parameter value to determine whether the parameter exhibits oscillatory behavior. For example, suppose we have a list of values (0.48, 0.52, 0.54, 0.51). A person could determine based on the values moving back and forth around 0.50. Nothing more than observing the values and recognizing a pattern. See MPEP § 2106.04(a)(2)(III).
Step 2A Prong Two Analysis: Does the claim recite additional elements that integrate the judicial exception into a practical application? See MPEP § 2106.04(d).
Independent claim 1 recites in part:
A computer-implemented method performed by a training system while training a machine learning model, comprising:
applying an oscillation mitigation procedure during the quantization- aware training of the machine learning model in response to identifying the oscillation, the oscillation mitigation procedure comprising at least one of oscillation dampening or parameter freezing, as drafted, amount to additional element that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or merely uses a computer in its ordinary capacity as a tool to perform an existing process. See MPEP §§ 2106.04(d), 2106.05(f)(2).
Step 2B Analysis: Does the claim recite additional elements that amount to significantly more than the judicial exception? See MPEP § 2106.05.
First, the additional elements directed to generally linking the use of a judicial exception to a particular technological environment or field of use are deemed insufficient to transform the judicial exception to a patentable invention because the claimed limitations generally link the judicial exception to the technology environment, see MPEP 2106.05(h). However, they are included below for the sake of completeness.
Second, the additional elements mere application of the abstract idea or mere instructions to implement an abstract idea on a computer are deemed insufficient to transform the judicial exception to a patentable invention because the limitations generally apply the use of a generic computer and/or process with the judicial exception. See MPEP 2106.05(f). However, they are included below for the sake of completeness.
A computer-implemented method performed by a training system while training a machine learning model, comprising:
applying an oscillation mitigation procedure during the quantization- aware training of the machine learning model in response to identifying the oscillation, the oscillation mitigation procedure comprising at least one of oscillation dampening or parameter freezing, as drafted, amount to additional element that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or merely uses a computer in its ordinary capacity as a tool to perform an existing process. See MPEP §§ 2106.04(d), 2106.05(f)(2).
Regarding claim 13:
Claim 13 recites: A processing system, comprising:
a memory comprising computer-executable instructions; and
a processor configured to execute the computer-executable instructions and cause the processing system to perform an operation comprising:
identifying oscillation of a parameter of a machine learning model during quantization-aware training of the machine learning model; and
applying an oscillation mitigation procedure during the quantization-aware training of the machine learning model in response to identifying the oscillation, the oscillation mitigation procedure comprising at least one of oscillation dampening or parameter freezing
Step 2A Prong One Analysis: Does the claim recite an abstract idea, law of nature, or natural phenomenon? See MPEP § 2106.04(II)(A)(1).
Claim 13 is directed to an abstract idea, specifically, a mental process – concepts performed in the human mind or by a human using a pen and paper (including an observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2)(III).
Independent claim 13 recites in part:
“identifying oscillation of a parameter of the machine learning model during quantization-aware training of the machine learning model”
The limitation above is broadly and reasonably interpreted as a mental process, as a form or mental evaluation or judgement. It encompasses observing or evaluating successive parameter value to determine whether the parameter exhibits oscillatory behavior. For example, suppose we have a list of values (0.48, 0.52, 0.54, 0.51). A person could determine based on the values moving back and forth around 0.50. Nothing more than observing the values and recognizing a pattern. See MPEP § 2106.04(a)(2)(III).
Step 2A Prong Two Analysis: Does the claim recite additional elements that integrate the judicial exception into a practical application? See MPEP § 2106.04(d).
Independent claim 13 recites in part:
A processing system, comprising:
a memory comprising computer-executable instructions; and
a processor configured to execute the computer-executable instructions and cause the processing system to perform an operation comprising, as drafted, amount to additional elements that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, such generic computing components recited at a high-level of generality (i.e., as a generic processor performing data gathering and mathematical calculations), which the courts have recognized as well‐understood, routine, and conventional when they are claimed in a generic manner. See MPEP § 2106.05(d)(II). MPEP §§ 2106.04(d), 2106.05(f)(2).
applying an oscillation mitigation procedure during the quantization- aware training of the machine learning model in response to identifying the oscillation, the oscillation mitigation procedure comprising at least one of oscillation dampening or parameter freezing, as drafted, amount to additional element that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or merely uses a computer in its ordinary capacity as a tool to perform an existing process. See MPEP §§ 2106.04(d), 2106.05(f)(2).
Step 2B Analysis: Does the claim recite additional elements that amount to significantly more than the judicial exception? See MPEP § 2106.05.
First, the additional elements directed to generally linking the use of a judicial exception to a particular technological environment or field of use are deemed insufficient to transform the judicial exception to a patentable invention because the claimed limitations generally link the judicial exception to the technology environment, see MPEP 2106.05(h). However, they are included below for the sake of completeness.
Second, the additional elements mere application of the abstract idea or mere instructions to implement an abstract idea on a computer are deemed insufficient to transform the judicial exception to a patentable invention because the limitations generally apply the use of a generic computer and/or process with the judicial exception. See MPEP 2106.05(f). However, they are included below for the sake of completeness.
A processing system, comprising:
a memory comprising computer-executable instructions; and
a processor configured to execute the computer-executable instructions and cause the processing system to perform an operation comprising as drafted, amount to additional elements that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, such generic computing components recited at a high-level of generality (i.e., as a generic processor performing data gathering and mathematical calculations), which the courts have recognized as well‐understood, routine, and conventional when they are claimed in a generic manner. See MPEP § 2106.05(d)(II). MPEP §§ 2106.04(d), 2106.05(f)(2).
applying an oscillation mitigation procedure during the quantization- aware training of the machine learning model in response to identifying the oscillation, the oscillation mitigation procedure comprising at least one of oscillation dampening or parameter freezing, as drafted, amount to additional element that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or merely uses a computer in its ordinary capacity as a tool to perform an existing process. See MPEP §§ 2106.04(d), 2106.05(f)(2).
Thus, considering the additional elements individually and in combination and the claims as a whole, the additional elements do not provide significantly more than the abstract idea. The claims are not eligible subject matter.
Therefore, in examining elements as recited by the limitations individually and as an ordered combination, as a whole the independent claim limitations do not recite what have the courts have identified as “significantly more”.
Regarding claim 22:
Claim 22 recites: A non-transitory computer-readable medium comprising computer- executable instructions that, when executed by a processor of a processing system, cause the processing system to perform an operation comprising:
identifying oscillation of a parameter of a machine learning model during quantization-aware training of the machine learning model; and
applying an oscillation mitigation procedure during the quantization- aware training of the machine learning model in response to identifying the oscillation, the oscillation mitigation procedure comprising at least one of oscillation dampening or parameter freezing
Step 2A Prong One Analysis: Does the claim recite an abstract idea, law of nature, or natural phenomenon? See MPEP § 2106.04(II)(A)(1).
Claim 22 is directed to an abstract idea, specifically, a mental process – concepts performed in the human mind or by a human using a pen and paper (including an observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2)(III).
Independent claim 22 recites in part:
“identifying oscillation of a parameter of the machine learning model during quantization-aware training of the machine learning model”
The limitation above is broadly and reasonably interpreted as a mental process, as a form or mental evaluation or judgement. It encompasses observing or evaluating successive parameter value to determine whether the parameter exhibits oscillatory behavior. For example, suppose we have a list of values (0.48, 0.52, 0.54, 0.51). A person could determine based on the values moving back and forth around 0.50. Nothing more than observing the values and recognizing a pattern. See MPEP § 2106.04(a)(2)(III).
Step 2A Prong Two Analysis: Does the claim recite additional elements that integrate the judicial exception into a practical application? See MPEP § 2106.04(d).
Independent claim 22 recites in part:
A non-transitory computer-readable medium comprising computer- executable instructions that, when executed by a processor of a processing system, cause the processing system to perform an operation comprising, as drafted, amount to additional elements that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, such generic computing components recited at a high-level of generality (i.e., as a generic processor performing data gathering and mathematical calculations), which the courts have recognized as well‐understood, routine, and conventional when they are claimed in a generic manner. See MPEP § 2106.05(d)(II). MPEP §§ 2106.04(d), 2106.05(f)(2).
applying an oscillation mitigation procedure during the quantization- aware training of the machine learning model in response to identifying the oscillation, the oscillation mitigation procedure comprising at least one of oscillation dampening or parameter freezing, as drafted, amount to additional element that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or merely uses a computer in its ordinary capacity as a tool to perform an existing process. See MPEP §§ 2106.04(d), 2106.05(f)(2).
Step 2B Analysis: Does the claim recite additional elements that amount to significantly more than the judicial exception? See MPEP § 2106.05.
First, the additional elements directed to generally linking the use of a judicial exception to a particular technological environment or field of use are deemed insufficient to transform the judicial exception to a patentable invention because the claimed limitations generally link the judicial exception to the technology environment, see MPEP 2106.05(h). However, they are included below for the sake of completeness.
Second, the additional elements mere application of the abstract idea or mere instructions to implement an abstract idea on a computer are deemed insufficient to transform the judicial exception to a patentable invention because the limitations generally apply the use of a generic computer and/or process with the judicial exception. See MPEP 2106.05(f). However, they are included below for the sake of completeness.
A non-transitory computer-readable medium comprising computer- executable instructions that, when executed by a processor of a processing system, cause the processing system to perform an operation comprising, as drafted, amount to additional elements that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, such generic computing components recited at a high-level of generality (i.e., as a generic processor performing data gathering and mathematical calculations), which the courts have recognized as well‐understood, routine, and conventional when they are claimed in a generic manner. See MPEP § 2106.05(d)(II). MPEP §§ 2106.04(d), 2106.05(f)(2).
applying an oscillation mitigation procedure during the quantization- aware training of the machine learning model in response to identifying the oscillation, the oscillation mitigation procedure comprising at least one of oscillation dampening or parameter freezing, as drafted, amount to additional element that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or merely uses a computer in its ordinary capacity as a tool to perform an existing process. See MPEP §§ 2106.04(d), 2106.05(f)(2).
Thus, considering the additional elements individually and in combination and the claims as a whole, the additional elements do not provide significantly more than the abstract idea. The claims are not eligible subject matter.
Therefore, in examining elements as recited by the limitations individually and as an ordered combination, as a whole the independent claim limitations do not recite what have the courts have identified as “significantly more”.
Regarding claim 30:
Claim 30 recites: A processing system, comprising:
means for identifying oscillation of a parameter of a machine learning model during quantization-aware training of the machine learning model; and
means for applying an oscillation mitigation procedure during the quantization-aware training of the machine learning model in response to identifying the oscillation, the oscillation mitigation procedure comprising at least one of oscillation dampening or parameter freezing
Step 2A Prong One Analysis: Does the claim recite an abstract idea, law of nature, or natural phenomenon? See MPEP § 2106.04(II)(A)(1).
Claim 30 is directed to an abstract idea, specifically, a mental process – concepts performed in the human mind or by a human using a pen and paper (including an observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2)(III).
Independent claim 30 recites in part:
“means for identifying oscillation of a parameter of a machine learning model during quantization-aware training of the machine learning model”
The limitation above is broadly and reasonably interpreted as a mental process, as a form or mental evaluation or judgement. It encompasses observing or evaluating successive parameter value to determine whether the parameter exhibits oscillatory behavior. For example, suppose we have a list of values (0.48, 0.52, 0.54, 0.51). A person could determine based on the values moving back and forth around 0.50. Nothing more than observing the values and recognizing a pattern. See MPEP § 2106.04(a)(2)(III).
Step 2A Prong Two Analysis: Does the claim recite additional elements that integrate the judicial exception into a practical application? See MPEP § 2106.04(d).
Independent claim 30 recites in part:
A processing system, comprising:
means for applying an oscillation mitigation procedure during the quantization-aware training of the machine learning model in response to identifying the oscillation, the oscillation mitigation procedure comprising at least one of oscillation dampening or parameter freezing, as drafted, amount to additional element that amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or merely uses a computer in its ordinary capacity as a tool to perform an existing process. See MPEP §§ 2106.04(d), 2106.05(f)(2).
Thus, considering the additional elements individually and in combination and the claims as a whole, the additional elements do not provide significantly more than the abstract idea. The claims are not eligible subject matter.
Therefore, in examining elements as recited by the limitations individually and as an ordered combination, as a whole the independent claim limitations do not recite what have the courts have identified as “significantly more”.
Furthermore, regarding dependent claims 2-12 are dependent on claim 1, claims 14-21 are dependent on claim 13, claims 23-29 are dependent on claim 22, the claims are directed to a judicial exception without significantly more as highlighted below in the claim limitations by evaluating the claim limitations under Step 2A and 2B:
Claims 2, 14 and 23 incorporates the rejection of independent claims 1, 13 and 22 respectively and does not integrate the judicial exception into a practical application.
Claims 3, 15 and 24 incorporates the rejection of dependent claims 2, 14 and 23 respectively, and incorporates a mental process, as a form of mental evaluation or judgement, and or by a human using a pen and paper. See MPEP § 2106.04(a)(2)(III).
Claims 4, 16 and 25 incorporated the rejection of dependent claims 3, 15 and 24 respectively, and does not integrate the judicial exception into a practical application.
Claims 5, 17 and 26 incorporates the rejection of dependent claims 3, 15 and 24 respectively, and does not integrate the judicial exception into a practical application.
Claims 6, 18 and 27 incorporates the rejection of dependent claims 5, 17 and 26 respectively, and does not integrate the judicial exception into a practical application.
Claims 7, 19 and 28 incorporates the rejection of dependent claims 5, 17 and 26 respectively, and as drafted, amount to additional element that amounts to adding the words “apply it” (or an equivalent) with the judicial exception and reciting only the idea of a solution or outcome, i.e., the claim fails to recite details of how a solution to a problem is accomplished because it is unclear how the “oscillation mitigation” is used nor the specification makes it clear how these actions are performed.
Claims 8, 20 and 29 incorporates the rejection of dependent claims 5, 17 and 26 respectively, and incorporates a mental process, as a form of mental evaluation or judgement, and or by a human using a pen and paper. See MPEP § 2106.04(a)(2)(III).
Claims 9 and 21 incorporates the rejection of dependent claims 3 and 15 respectively, and does not integrate the judicial exception into a practical application.
Claims 10-12 incorporates the rejection of dependent claim 9 respectively, and incorporates a mathematical formula or equation, as directed to “a claim that recites a numerical formula or equation will be considered as falling within the "mathematical concepts" grouping. In addition, there are instances where a formula or equation is written in text format that should also be considered as falling within this grouping”. See MPEP § 2106.04(a)(2)(I)(B).
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.
Claims 1-3, 5, 7-8, 13-15, 17, 19-20, 22, 23-24, 26 and 28-30 are rejected under 35 U.S.C 102 as being unpatentable over “ A Novel Training Method for sub-4-bit Mobile Net Model”, https://arxiv.org/abs/2008.04693, Park et al, 08/11/2020, hereinafter referred to as Park.
With respect to claim 1, Park disclose:
A computer-implemented method performed by a training system while training a machine learning model, comprising: identifying oscillation of a parameter of the machine learning model during quantization-aware training of the machine learning model (On pages 4–6 (3.1 Observation & 3.2 Activation Instability Metric), Park teaches monitoring activation instability during quantization-aware training by computing an AIWQ metric that quantifies instability of network behavior. Park further teaches that the measured instability causes training values to fluctuate and uses the measured instability to determine continued training. One ordinary skill in the art would have understood the disclosed fluctuating instability during AT to represent oscillatory behavior of the training process and would have found it obvious to identify oscillatory parameter behavior as the basis for applying the disclosed parameter-freezing mitigation.)
Applying an oscillation mitigation procedure during the quantization- aware training of the machine learning model in response to identifying the oscillation (On page 8 (Quantization for Asymmetric Distributions), Park teaches computing an activation instability metric (AIWQ) during quantization-aware training to identify layers exhibiting instability. Based on the computed instability metric, Park schedules and applies progressive weight freezing to the identified layers while training continues. Thus, Park teaches applying a mitigation procedure during quantization-aware training in response to an identified training instability, wherein the mitigation procedure comprises parameter freezing. )
The oscillation mitigation procedure comprising at least one of oscillation dampening or parameter freezing (The examiner selects: parameter freezing. On pages 5–6 (Activation Instability Metric), Park discloses utilizing the per-layer sensitivity when determining the order of freezing the weights during training.)
Regarding claim 2, Park disclose the elements of claim 1. In addition, Park disclose:
The method of Claim 1, wherein the parameter of the machine learning model comprises a quantized weight of the machine learning model (In Fig. 3 and Pages 4–5 (3.1 Observation), Park teaches quantized weight values.)
Regarding claim 3, Park disclose the elements of claim 2. In addition, Park disclose:
The method of Claim 2, wherein identifying the oscillation comprises determining an integer weight value oscillation frequency associated with the quantized weight of the machine learning model (On page 5 (3.2 Activation Instability Metric), Park teaches determining the order of freezing the weights during training )
Regarding claim 5, Park disclose the elements of claim 3. In addition, Park disclose:
The method of Claim 3, wherein the oscillation mitigation procedure comprises freezing the integer value of the quantized weight at a set value for any remaining iterations during the quantization-aware training of the machine learning model (On page 2, Park teaches freezing iterative training that minimizes the effects of AIWQ by progressively freezing the weights sensitive to AIWQ during training.)
Regarding claim 7, Park disclose the elements of claim 5. In addition, Park disclose:
The method of Claim 5, further comprising applying the oscillation mitigation procedure based on the integer weight value oscillation frequency exceeding an oscillation frequency threshold (On page 9-10 (4.2 Proposed Method: Due), Park teaches oneAn additional advantage is that DuQ utilizes all the gradients across the entire activation.Data. PACT only utilizes the gradients from the truncation interval (the valuerange larger than the truncation threshold), while OIL only utilizes the gradientsfrom the quantization interval (between the minimum and maximum quantizationLevels).)
Regarding claim 8, Park disclose the elements of claim 5. In addition, Park disclose:
The method of Claim 5, further comprising determining the set value based on a rounded exponential moving average value of the quantized weight (On page 11 (5. Experiments), Park teaches using an exponential moving average of parameters with a momentum of 0.9997, and all networks were trained using PROFIT and Due (with negative padding if applicable).)
With respect to claim 13, Park disclose:
A processing system, comprising: a memory comprising computer-executable instructions (Park disclose operation such as training a neural network, quantization-aware training, computing gradients, updating weights, etc. All these operation necessarily require a processor (e.g., CPU, GPU, TPU, or other computing hardware).)
A processor configured to execute the computer-executable instructions and cause the processing system to perform an operation comprising: identifying oscillation of a parameter of a machine learning model during quantization-aware training of the machine learning model (On pages 4–6 (3.1 Observation & 3.2 Activation Instability Metric), Park teaches monitoring activation instability during quantization-aware training by computing an AIWQ metric that quantifies instability of network behavior. Park further teaches that the measured instability causes training values to fluctuate and uses the measured instability to determine continued training. One ordinary skill in the art would have understood the disclosed fluctuating instability during AT to represent oscillatory behavior of the training process and would have found it obvious to identify oscillatory parameter behavior as the basis for applying the disclosed parameter-freezing mitigation.)
Applying an oscillation mitigation procedure during the quantization-aware training of the machine learning model in response to identifying the oscillation (On page 8 (Quantization for Asymmetric Distributions), Park teaches computing an activation instability metric (AIWQ) during quantization-aware training to identify layers exhibiting instability. Based on the computed instability metric, Park schedules and applies progressive weight freezing to the identified layers while training continues. Thus, Park teaches applying a mitigation procedure during quantization-aware training in response to an identified training instability, wherein the mitigation procedure comprises parameter freezing. )
The oscillation mitigation procedure comprising at least one of oscillation dampening or parameter freezing (The examiner selects: parameter freezing. On pages 5–6 (Activation Instability Metric), Park discloses utilizing the per-layer sensitivity when determining the order of freezing the weights during training.)
Regarding claim 14, Park disclose the elements of claim 13. In addition, Park disclose:
The processing system of Claim 13, wherein the parameter of the machine learning model comprises a quantized weight of the machine learning model (In Fig. 3 and Pages 4–5 (3.1 Observation), Park teaches quantized weight values.)
Regarding claim 15, Park disclose the elements of claim 14. In addition, Park disclose:
The processing system of Claim 14, wherein identifying the oscillation comprises determining an integer weight value oscillation frequency associated with the quantized weight of the machine learning model (On page 5 (3.2 Activation Instability Metric), Park teaches determining the order of freezing the weights during training )
Regarding claim 17, Park disclose the elements of claim 15. In addition, Park disclose:
The processing system of Claim 15, wherein the oscillation mitigation procedure comprises freezing the integer value of the quantized weight at a set value for any remaining iterations during the quantization-aware training of the machine learning model (On page 9-10 (4.2 Proposed Method: Due), Park teaches oneAn additional advantage is that DuQ utilizes all the gradients across the entire activation.Data. PACT only utilizes the gradients from the truncation interval (the valuerange larger than the truncation threshold), while OIL only utilizes the gradientsfrom the quantization interval (between the minimum and maximum quantizationLevels).)
Regarding claim 19 , Park disclose the elements of claim 17. In addition, Park disclose:
The processing system of Claim 17, the operation further comprising applying the oscillation mitigation procedure based on the integer weight value oscillation frequency exceeding an oscillation frequency threshold (On page 9-10 (4.2 Proposed Method: Due), Park teaches one An additional advantage is that DuQ utilizes all the gradients across the entire activation data PACT only utilizes the gradients from the truncation interval (the value range larger than the truncation threshold), while OIL only utilizes the gradientsfrom the quantization interval (between the minimum and maximum quantizationLevels).)
Regarding claim 20 , Park disclose the elements of claim 17. In addition, Park disclose:
The processing system of Claim 17, the operation further comprising determining the set value based on a rounded exponential moving average value of the quantized weight ((On page 11 (5. Experiments), Park teaches using an exponential moving average of parameters with a momentum of 0.9997, and all networks were trained using PROFIT and Due (with negative padding if applicable).)
With respect to claim 22, Park disclose:
A non-transitory computer-readable medium comprising computer- executable instructions that, when executed by a processor of a processing system, cause the processing system to perform an operation comprising: identifying oscillation of a parameter of a machine learning model during quantization-aware training of the machine learning model (Park disclose operation such as training a neural network, quantization-aware training, computing gradients, updating weights, etc. All these operation necessarily require a processor (e.g., CPU, GPU, TPU, or other computing hardware. On pages 4–6 (3.1 Observation & 3.2 Activation Instability Metric), Park teaches monitoring activation instability during quantization-aware training by computing an AIWQ metric that quantifies instability of network behavior. Park further teaches that the measured instability causes training values to fluctuate and uses the measured instability to determine continued training. One ordinary skill in the art would have understood the disclosed fluctuating instability during AT to represent oscillatory behavior of the training process and would have found it obvious to identify oscillatory parameter behavior as the basis for applying the disclosed parameter-freezing mitigation.)
Applying an oscillation mitigation procedure during the quantization- aware training of the machine learning model in response to identifying the oscillation (On page 8 (Quantization for Asymmetric Distributions), Park teaches computing an activation instability metric (AIWQ) during quantization-aware training to identify layers exhibiting instability. Based on the computed instability metric, Park schedules and applies progressive weight freezing to the identified layers while training continues. Thus, Park teaches applying a mitigation procedure during quantization-aware training in response to an identified training instability, wherein the mitigation procedure comprises parameter freezing. )
The oscillation mitigation procedure comprising at least one of oscillation dampening or parameter freezing (The examiner selects: parameter freezing. On pages 5–6 (Activation Instability Metric), Park discloses utilizing the per-layer sensitivity when determining the order of freezing the weights during training.)
Regarding claim 23, Park disclose the elements of claim 23. In addition, Park disclose:
The non-transitory computer-readable medium of Claim 22, wherein the parameter of the machine learning model comprises a quantized weight of the machine learning model (In Fig. 3 and Pages 4–5 (3.1 Observation), Park teaches quantized weight values.)
Regarding claim 24, Park disclose the elements of claim 23. In addition, Park disclose:
The non-transitory computer-readable medium of Claim 23, wherein identifying the oscillation comprises determining an integer weight value oscillation frequency associated with the quantized weight of the machine learning model (On page 5 (3.2 Activation Instability Metric), Park teaches determining the order of freezing the weights during training )
Regarding claim 26, Park disclose the elements of claim 24. In addition, Park disclose:
The non-transitory computer-readable medium of Claim 24, wherein the oscillation mitigation procedure comprises freezing the integer value of the quantized weight at a set value for any remaining iterations during the quantization-aware training of the machine learning model (On page 2, Park teaches freezing iterative training that minimizes the effects of AIWQ by progressively freezing the weights sensitive to AIWQ during training.)
Regarding claim 28, Park disclose the elements of claim 26. In addition, Park disclose:
The non-transitory computer-readable medium of Claim 26, the operation further comprising applying the oscillation mitigation procedure based on the integer weight value oscillation frequency exceeding an oscillation frequency threshold (On page 9-10 (4.2 Proposed Method: Due), Park teaches one An additional advantage is that DuQ utilizes all the gradients across the entire activation data PACT only utilizes the gradients from the truncation interval (the value range larger than the truncation threshold), while OIL only utilizes the gradients from the quantization interval (between the minimum and maximum quantization Levels).)
Regarding claim 29 , Park disclose the elements of claim 26. In addition, Park disclose:
The non-transitory computer-readable medium of Claim 26, the operation further comprising determining the set value based on a rounded exponential moving average value of the quantized weight ((On page 11 (5. Experiments), Park teaches using an exponential moving average of parameters with a momentum of 0.9997, and all networks were trained using PROFIT and Due (with negative padding if applicable).)
With respect to claim 30, Park disclose:
A processing system, comprising: means for identifying oscillation of a parameter of a machine learning model during quantization-aware training of the machine learning model (On pages 4–6 (3.1 Observation & 3.2 Activation Instability Metric), Park teaches monitoring activation instability during quantization-aware training by computing an AIWQ metric that quantifies instability of network behavior. Park further teaches that the measured instability causes training values to fluctuate and uses the measured instability to determine continued training. One ordinary skill in the art would have understood the disclosed fluctuating instability during AT to represent oscillatory behavior of the training process and would have found it obvious to identify oscillatory parameter behavior as the basis for applying the disclosed parameter-freezing mitigation.)
Means for applying an oscillation mitigation procedure during the quantization-aware training of the machine learning model in response to identifying the oscillation (On page 8 (Quantization for Asymmetric Distributions), Park teaches computing an activation instability metric (AIWQ) during quantization-aware training to identify layers exhibiting instability. Based on the computed instability metric, Park schedules and applies progressive weight freezing to the identified layers while training continues. Thus, Park teaches applying a mitigation procedure during quantization-aware training in response to an identified training instability, wherein the mitigation procedure comprises parameter freezing. )
The oscillation mitigation procedure comprising at least one of oscillation dampening or parameter freezing (The examiner selects: parameter freezing. On pages 5–6 (Activation Instability Metric), Park discloses utilizing the per-layer sensitivity when determining the order of freezing the weights during training.)
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.
Claim(s) 6, 9, 18, 21 and 27 are rejected under 35 U.S.C 103 as being unpatentable over Park in view of “Absolute-Cosine Regularization”, https://www.isca archive.org/interspeech_2020/nguyen20c_interspeech.pdf, Nguyen et al, 2020, hereinafter referred to as Nguyen.
Regarding claim 6, Park disclose the elements of claim 5. Park does not explicitly disclose:
The method of Claim 5, further comprising applying a second oscillation mitigation procedure during the quantization-aware training of the machine learning model, wherein the second oscillation mitigation procedure comprises updating the machine learning model based on a loss function including an oscillation dampening loss regularization term
However, Nguyen disclose the limitation (On page 2 (3. Absolute-Cosine Regularization), Nguyen teaches quantization-aware training, modifying the training objective, adding an absolute-cosine regularization and updating the neural network using combined loss.)
Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teaching of Park to include Nguyen’s, with generating the quantized weight value associated with the connection based on the determined minimum of the loss function as taught by Nguyen. The motivation for doing so would have been to improve the efficiency of post-training quantization (See (Page 2) of Nguyen).
Regarding claim 9, Park disclose the elements of claim 3. Park does not explicitly disclose:
The method of Claim 3, wherein the oscillation mitigation procedure comprises updating the machine learning model based on a loss function including an oscillation dampening loss regularization term
However, Nguyen disclose the limitation (On page 2 (3. Absolute-Cosine Regularization), Nguyen teaches updating the machine learning model based on a loss function that includes a regularization (ACosR).)
Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teaching of Park to include Nguyen’s, with generating the quantized weight value associated with the connection based on the determined minimum of the loss function as taught by Nguyen. The motivation for doing so would have been to improve the efficiency of post-training quantization (See (Page 2) of Nguyen).
Regarding claim 18, Park disclose the elements of claim 17. Park does not explicitly disclose:
The processing system of Claim 17, the operation further comprising applying a second oscillation mitigation procedure during the quantization-aware training of the machine learning model, wherein the second oscillation mitigation procedure comprises updating the machine learning model based on a loss function including an oscillation dampening loss regularization term
However, Nguyen disclose the limitation (On page 2 (3. Absolute-Cosine Regularization), Nguyen teaches quantization-aware training, modifying the training objective, adding an absolute-cosine regularization and updating the neural network using combined loss.)
Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teaching of Park to include Nguyen’s, with generating the quantized weight value associated with the connection based on the determined minimum of the loss function as taught by Nguyen. The motivation for doing so would have been to improve the efficiency of post-training quantization (See (Page 2) of Nguyen).
Regarding claim 21, Park disclose the elements of claim 15. Park does not explicitly disclose:
The processing system of Claim 15, wherein the oscillation mitigation procedure comprises updating the machine learning model based on a loss function including an oscillation dampening loss regularization term
However, Nguyen disclose the limitation (On page 2 (3. Absolute-Cosine Regularization), Nguyen teaches quantization-aware training, modifying the training objective, adding an absolute-cosine regularization and updating the neural network using combined loss.)
Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teaching of Park to include Nguyen’s, with generating the quantized weight value associated with the connection based on the determined minimum of the loss function as taught by Nguyen. The motivation for doing so would have been to improve the efficiency of post-training quantization (See (Page 2) of Nguyen).
Regarding claim 27, Park disclose the elements of claim 15. Park does not explicitly disclose:
The non-transitory computer-readable medium of Claim 26, the operation further comprising applying a second oscillation mitigation procedure during the quantization-aware training of the machine learning model, wherein the second oscillation mitigation procedure comprises updating the machine learning model based on a loss function including an oscillation dampening loss regularization term
However, Nguyen disclose the limitation (On page 2 (3. Absolute-Cosine Regularization), Nguyen teaches quantization-aware training, modifying the training objective, adding an absolute-cosine regularization and updating the neural network using combined loss.)
Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teaching of Park to include Nguyen’s, with generating the quantized weight value associated with the connection based on the determined minimum of the loss function as taught by Nguyen. The motivation for doing so would have been to improve the efficiency of post-training quantization (See (Page 2) of Nguyen).
Claim(s) 10-12 are rejected under 35 U.S.C 103 as being unpatentable over Park in view of Nguyen in view of Naumov et al . (US Patent No.: 11,468,313 B1), hereinafter referred to as Naumov.
Regarding claim 10, Park in view of Nguyen disclose the elements of claim 9. Park in view of Nguyen does not explicitly disclose:
PNG
media_image1.png
147
518
media_image1.png
Greyscale
However, Naumov disclose the limitation (In Col. 11, lines 5–20, Naumov discloses a regularization function,” “regularization term, “and/or a “regularizer” may include any term that may be included with a loss function that may impose a limit on the complexity of a function. A regularization term λR(w, x).)
Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teaching of Park and Nguyen to include Naumov, generating the quantized weight value associated with the connection based on the determined minimum of the loss function. The motivation for doing so would have been to improve the efficiency of quantization of artificial neural networks such that a quantization process may be efficiently implemented on computing hardware. (See (Col. 5, lines 40-42) of Naumov.)
Regarding claim 11, Park in view of Nguyen disclose the elements of claim 9. Park in view of Nguyen does not explicitly disclose:
PNG
media_image2.png
271
605
media_image2.png
Greyscale
However, Naumov disclose the limitation (In Col. 14, lines 1-5, Naumov teaches defining periodic quantization centers, adding a quantization regularization term to the loss, penalizing weights that are away from preferred quantization locations and optimizing the network, so weights converge toward quantization values.)
Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teaching of Park and Nguyen to include Naumov, generating the quantized weight value associated with the connection based on the determined minimum of the loss function. The motivation for doing so would have been to improve the efficiency of quantization of artificial neural networks such that a quantization process may be efficiently implemented on computing hardware. (See (Col. 5, lines 40-42) of Naumov.)
Regarding claim 12, Park in view of Nguyen disclose the elements of claim 9. Park in view of Nguyen does not explicitly disclose:
The method of Claim 9, wherein: updating the machine learning model based on the loss function including the oscillation dampening loss regularization term comprises determining
PNG
media_image3.png
118
566
media_image3.png
Greyscale
However, Naumov disclose the limitation (In Col. 14, lines 1-5, Naumov teaches defining periodic quantization centers, adding a quantization regularization term to the loss, penalizing weights that are away from preferred quantization locations and optimizing the network, so weights converge toward quantization values.)
Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teaching of Park and Nguyen to include Naumov, generating the quantized weight value associated with the connection based on the determined minimum of the loss function. The motivation for doing so would have been to improve the efficiency of quantization of artificial neural networks such that a quantization process may be efficiently implemented on computing hardware. (See (Col. 5, lines 40-42) of Naumov.)
Allowable Subject Matter
Claims 4, 16 and 25 are objected for being dependent upon a rejected base claim, but would be allowable if rewritten in independent form, including all the limitations of the base claim and any intervening claims.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to EVEL HONORE whose telephone number is (703)756-1179. The examiner can normally be reached Monday-Friday 8 a.m. -5:30 p.m.
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, Mariela D Reyes can be reached at (571) 270-1006. 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.
EVEL HONORE
Examiner
Art Unit 2142
/Mariela Reyes/Supervisory Patent Examiner, Art Unit 2142
Prosecution Insights