DETAILED ACTION
This office action is in response to the application filed on March 22, 2024.
Claims 1-22 are pending and have been examined. Claims 1-22 are rejected.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Information Disclosure Statement
Acknowledgment is made of the information disclosure statements filed March 22, 2024 and March 29, 2024, which comply with 37 CFR 1.97. As such, the information disclosure statements have been placed in the application file and the information referred to therein has been considered by the examiner.
Claim Rejections - 35 USC § 101
35 U.S.C. § 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-22 are rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter.
According to the USPTO guidelines, a claim is directed to non-statutory subject matter if:
Step 1: The claim does not fall within one of the four statutory categories of invention (process, machine, manufacture, or composition of matter) – see MPEP 2106.03, or,
Step 2: The claim recites a judicial exception, e.g. an abstract idea, without reciting additional elements that amount to significantly more than the judicial exception, as determined using the following analysis – see MPEP 2106.04:
Step 2A, Prong 1: Does the claim recite an abstract idea, law of nature, or natural phenomenon?
Step 2A, Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
Step 2B: Does the claim recite additional elements that amount to significantly more than the judicial exception? - see MPEP 2106.05
MPEP 2106.04(a)(2)(I) states: “The mathematical concepts grouping is defined as mathematical relationships, mathematical formulas or equations, and mathematical calculations.”
MPEP 2106.04(a)(2)(III) states: “Accordingly, the “mental processes” abstract idea grouping is defined as concepts performed in the human mind, and examples of mental processes include observations, evaluations, judgements, and opinions.
Further, the MPEP states: “The courts do not distinguish between mental processes that are performed entirely in the human mind and mental processes that require a human to use a physical aid (e.g. pen and paper or a slide run) to perform the claim limitation.
Using the two-step inquiry, it is clear that Claims 1-22 are each directed to non-statutory subject matter as shown below:
With respect to Claims 1, 12, and 18:
Step 1: Claim 1 is directed to a method, also known as a process, which is one of the four statutory categories of patentable subject matter. Claim 12 corresponds to an article of manufacture, which is one of the four statutory categories of patentable subject matter. Claim 18 corresponds to an article of manufacture, which is one of the four statutory categories of patentable subject matter.
Step 2A, Prong 1: A judicial exception is recited in the claims as they recite mental processes, which are abstract ideas:
“generating a first compressed set of data by compressing particular data from the first set of data to a second set of dimensions, wherein the second set of dimensions has fewer dimensions than the first set of dimensions;” (Covers mental concepts that could be practically performed in the human mind or through the use of a pencil and paper, including observation, evaluation, and judgement - see MPEP 2106.04.)
“generating a first reconstructed set of data by decompressing the first compressed set of data to the first set of dimensions;” (Covers mental concepts that could be practically performed in the human mind or through the use of a pencil and paper, including observation, evaluation, and judgement - see MPEP 2106.04.)
“determining a first reconstruction loss between the first reconstructed set of data and the particular data based at least in part on differences between the first reconstructed set of data and the particular data along the first set of dimensions;” (Covers mental concepts that could be practically performed in the human mind or through the use of a pencil and paper, including observation, evaluation, and judgement - see MPEP 2106.04.)
“generating a second compressed set of data by compressing a second set of data to the second set of dimensions;” (Covers mental concepts that could be practically performed in the human mind or through the use of a pencil and paper, including observation, evaluation, and judgement - see MPEP 2106.04.)
“generating a second reconstructed set of data by decompressing the second compressed set of data to the first set of dimensions;” (Covers mental concepts that could be practically performed in the human mind or through the use of a pencil and paper, including observation, evaluation, and judgement - see MPEP 2106.04.)
“determining a second reconstruction loss between the second reconstructed set of data and the second set of data based at least in part on differences between the second reconstructed set of data and the second set of data along the first set of dimensions;” (Covers mental concepts that could be practically performed in the human mind or through the use of a pencil and paper, including observation, evaluation, and judgement - see MPEP 2106.04.)
“determining a drift difference between the first reconstruction loss and the second reconstruction loss, and including the drift difference in an aggregate drift difference;” (Covers mental concepts that could be practically performed in the human mind or through the use of a pencil and paper, including observation, evaluation, and judgement - see MPEP 2106.04.)
“and determining whether to retrain the particular machine learning model based at least in part on one or more conditions that are based at least in part on the aggregate drift difference.” (Covers mental concepts that could be practically performed in the human mind or through the use of a pencil and paper, including observation, evaluation, and judgement - see MPEP 2106.04.)
Step 2A, Prong 2: The claims do not recite additional elements that integrate the judicial exception into a practical application:
“A computer-implemented method comprising:
storing a first set of data and a particular machine learning model, (Storing a first set of data and a machine learning model is regarded as a generic computer function of storing data. Storing data is considered insignificant extra-solution activity – see MPEP 2106.05(g).)
wherein the particular machine learning model was trained using at least part of the first set of data to predict one or more values along a first set of dimensions, (A machine learning model trained using at least part of a first set of data to predict one or more values along a first set of dimensions only amounts to “apply it” and mere instructions to implement an abstract idea on a computer – see MPEP 2106.05(f)(1).)
wherein the first set of data comprises a plurality of combinations of value occurrences in the first set dimensions;” (A first set of data comprising a plurality of combination of value occurrences in the first set dimensions generally links the use of the abstract idea to a particular technological environment or field of use – see MPEP 2106.05(h).)
“using the particular machine learning model to make a prediction for data along the first set of dimensions;” (Using a particular machine learning model to make a prediction for data along a first set of dimensions only amounts to “apply it” and mere instructions to implement an abstract idea on a computer – see MPEP 2106.05(f)(1).)
“and storing the aggregate drift difference in association with the particular machine learning model” (Storing the aggregate drift difference in association with a particular machine learning model is regarded as a generic computer function of storing data. Storing data is considered insignificant extra-solution activity – see MPEP 2106.05(g).)
Step 2B: The claim does not recite additional elements that amount to significantly more than the judicial exception. Storing a first set of data and a machine learning model is regarded as a generic computer function of storing data. Storing data is considered insignificant extra-solution activity – see MPEP 2106.05(g). A machine learning model trained using at least part of a first set of data to predict one or more values along a first set of dimensions only amounts to “apply it” and mere instructions to implement an abstract idea on a computer – see MPEP 2106.05(f)(1). A first set of data comprising a plurality of combination of value occurrences in the first set dimensions generally links the use of the abstract idea to a particular technological environment or field of use – see MPEP 2106.05(h). Using a particular machine learning model to make a prediction for data along a first set of dimensions only amounts to “apply it” and mere instructions to implement an abstract idea on a computer – see MPEP 2106.05(f)(1). Storing the aggregate drift difference in association with a particular machine learning model is regarded as a generic computer function of storing data. Storing data is considered insignificant extra-solution activity – see MPEP 2106.05(g).
With respect to Claims 2, 13, and 19:
Step 2A, Prong 1: Inherits the limitations and abstract ideas from Claims 1, 12, and 18, respectively. An additional judicial exception is recited in the claims as they recite mental processes, which are abstract ideas:
and wherein a second dimension of the second set of dimensions is selected to be orthogonal to the first dimension.” (Covers mental concepts that could be practically performed in the human mind or through the use of a pencil and paper, including observation, evaluation, and judgement - see MPEP 2106.04.)
Step 2A, Prong 2: The claim does not recite additional elements that integrate the judicial exception into a practical application:
“wherein at least a first dimension of the second set of dimensions comprises a distance from a hyperplane covering a selected combination of value occurrences of the first set of data; (A first dimension of a second set of dimensions comprising a distance from a hyperplane covering a selected combination of value occurrences of a first set of data generally links the use of the abstract idea to a particular technological environment or field of use – see MPEP 2106.05(h).)
Step 2B: The claim does not recite additional elements that amount to significantly more than the judicial exception. A first dimension of a second set of dimensions comprising a distance from a hyperplane covering a selected combination of value occurrences of a first set of data generally links the use of the abstract idea to a particular technological environment or field of use – see MPEP 2106.05(h).
With respect to Claims 3, 14, and 20: Step 2A, Prong 1: Inherits the limitations and abstract ideas from Claims 1, 12, and 18, respectively.
Step 2A, Prong 2: The claims do not recite additional elements that integrate the judicial exception into a practical application:
“wherein generating the first compressed set of data uses principal component analysis to compress the first set of data, (Generally links the use of the abstract idea to a particular technological environment or field of use – see MPEP 2106.05(h).)
and wherein generating the second compressed set of data uses the principal component analysis to compress the second set of data” (Generally links the use of the abstract idea to a particular technological environment or field of use – see MPEP 2106.05(h).)
Step 2B: The claims do not recite additional elements that amount to significantly more than the judicial exception. Using principal component analysis to compress a first set of data generally links the use of the abstract idea to a particular technological environment or field of use – see MPEP 2106.05(h). Using principal component analysis to compress a second set of data generally links the use of the abstract idea to a particular technological environment or field of use – see MPEP 2106.05(h).
With respect to Claims 4, 15, and 21: Step 2A, Prong 1: Inherits the limitations and abstract ideas from Claims 1, 12, and 18, respectively.
Step 2A, Prong 2: The claims do not recite additional elements that integrate the judicial exception into a practical application:
“wherein the second set of dimensions is different from the first set of dimensions, (Generally links the use of the abstract idea to a particular technological environment or field of use – see MPEP 2106.05(h).)
wherein generating the first compressed set of data uses a neural network to compress the first set of data based on one or more feature embedding vectors that describe the first set of data, (Generally links the use of the abstract idea to a particular technological environment or field of use – see MPEP 2106.05(h).)
and wherein generating the second compressed set of data uses the neural network to compress the second set of data based on one or more feature embedding vectors that describe the second set of data.” (Generally links the use of the abstract idea to a particular technological environment or field of use – see MPEP 2106.05(h).)
Step 2B: The claims do not recite additional elements that amount to significantly more than the judicial exception. A second set of dimensions different from a first set of dimensions generally links the use of the abstract idea to a particular technological environment or field of use – see MPEP 2106.05(h). Using a neural network to compress a first set of data based on one or more feature embedding vectors that describe the first set of data generally links the use of the abstract idea to a particular technological environment or field of use – see MPEP 2106.05(h). Using a neural network to compress a second set of data based on one or more feature embedding vectors that describe the second set of data generally links the use of the abstract idea to a particular technological environment or field of use – see MPEP 2106.05(h).
With respect to Claim 5:
Step 2A, Prong 1: Inherits the limitations and abstract ideas from Claim 1. An additional judicial exception is recited in the claim as it recites mental processes, which are abstract ideas:
“wherein each dimension of the second set of dimensions is selected to account for a maximum remaining variance in the first set of data.” (Covers mental concepts that could be practically performed in the human mind or through the use of a pencil and paper, including observation, evaluation, and judgement - see MPEP 2106.04.)
Step 2A, Prong 2: The claim does not recite additional elements that integrate the judicial exception into a practical application.
Step 2B: The claim does not recite additional elements that amount to significantly more than the judicial exception.
With respect to Claims 6 and 16: Step 2A, Prong 1: Inherits the limitations and abstract ideas from Claims 1 and 12, respectively. An additional judicial exception is recited in the claims as they recite mental processes, which are abstract ideas:
“wherein performing said generating the first compressed set of data, said generating the first reconstructed set of data, and determining the first reconstruction loss is performed automatically in response to training the particular machine learning model.” (Covers mental concepts that could be practically performed in the human mind or through the use of a pencil and paper, including observation, evaluation, and judgement - see MPEP 2106.04.)
Step 2A, Prong 2: The claims do not recite additional elements that integrate the judicial exception into a practical application:
“receiving a request to train a machine learning model on the first set of data;” (Receiving a request to train a machine learning model on a first set of data is regarded as a generic computer function of receiving data. Receiving data is considered insignificant extra-solution activity – see MPEP 2106.05(g).)
“in response to the request, training the particular machine learning model;” (Training a particular machine learning model in response to a request only amounts to “apply it” and mere instructions to implement an abstract idea on a computer – see MPEP 2106.05(f)(1).)
Step 2B: The claims do not recite additional elements that amount to significantly more than the judicial exception. Receiving a request to train a machine learning model on a first set of data is regarded as a generic computer function of receiving data. Receiving data is considered insignificant extra-solution activity – see MPEP 2106.05(g). Training a particular machine learning model in response to a request only amounts to “apply it” and mere instructions to implement an abstract idea on a computer – see MPEP 2106.05(f)(1).
With respect to Claim 7:
Step 2A, Prong 1: Inherits the limitations and abstract ideas from Claim 1. An additional judicial exception is recited in the claim as it recites mental processes, which are abstract ideas:
“determining, based at least in part on the aggregate drift difference, that the one or more conditions are not satisfied, and, without retraining the particular machine learning model, outputting a retraining score that indicates how close the one or more conditions are to being satisfied.” (Determining, based at least in part on an aggregate drift difference, that one or more conditions are not satisfied covers mental concepts that could be practically performed in the human mind or through the use of a pencil and paper, including observation, evaluation, and judgement - see MPEP 2106.04.)
Step 2A, Prong 2: The claim does not recite additional elements that integrate the judicial exception into a practical application:
“determining, based at least in part on the aggregate drift difference, that the one or more conditions are not satisfied, and, without retraining the particular machine learning model, outputting a retraining score that indicates how close the one or more conditions are to being satisfied.” (Outputting a retraining score that indicates how close the one or more conditions are to be satisfied is regarded as a generic computer function of outputting information. Outputting information is considered insignificant extra-solution activity – see MPEP 2106.05(g).)
Step 2B: The claim does not recite additional elements that amount to significantly more than the judicial exception. Outputting a retraining score that indicates how close the one or more conditions are to be satisfied is regarded as a generic computer function of outputting information. Outputting information is considered insignificant extra-solution activity – see MPEP 2106.05(g).
With respect to Claims 8, 17, and 22:
Step 2A, Prong 1: Inherits the limitations and abstract ideas from Claim 1, 12, and 18, respectively. An additional judicial exception is recited in the claims as they recite mental processes, which are abstract ideas:
“determining, based at least in part on the aggregate drift difference, that the one or more conditions are not satisfied, and, without retraining the particular machine learning model, outputting an aggregate drift difference specific to one or more of the first set of dimensions” (Determining, based at least in part on an aggregate drift difference, that one or more conditions are not satisfied covers mental concepts that could be practically performed in the human mind or through the use of a pencil and paper, including observation, evaluation, and judgement - see MPEP 2106.04.)
Step 2A, Prong 2: The claims do not recite additional elements that integrate the judicial exception into a practical application:
“determining, based at least in part on the aggregate drift difference, that the one or more conditions are not satisfied, and, without retraining the particular machine learning model, outputting an aggregate drift difference specific to one or more of the first set of dimensions” (Outputting an aggregate drift difference specific to one or more of the first set of dimensions is regarded as a generic computer function of outputting information. Outputting information is considered insignificant extra-solution activity – see MPEP 2106.05(g).)
Step 2B: The claims do not recite additional elements that amount to significantly more than the judicial exception. Outputting an aggregate drift difference specific to one or more of the first set of dimensions is regarded as a generic computer function of outputting information. Outputting information is considered insignificant extra-solution activity – see MPEP 2106.05(g).
With respect to Claim 9:
Step 2A, Prong 1: Inherits the limitations and abstract ideas from Claim 1. An additional judicial exception is recited in the claim as it recites mental processes, which are abstract ideas:
“determining, based at least in part on the aggregate drift difference, that the one or more conditions are satisfied;” (Covers mental concepts that could be practically performed in the human mind or through the use of a pencil and paper, including observation, evaluation, and judgement - see MPEP 2106.04.)
“based at least in part on determining that the one or more conditions are satisfied, scheduling a retraining of the particular machine learning model based at least in part on a workload that uses the particular machine learning model;” (Covers mental concepts that could be practically performed in the human mind or through the use of a pencil and paper, including observation, evaluation, and judgement - see MPEP 2106.04.)
“and retraining the particular machine learning model based at least in part on determining which particular dimensions to include from a superset of dimensions that includes the first set of dimensions and one or more other dimensions.” (Covers mental concepts that could be practically performed in the human mind or through the use of a pencil and paper, including observation, evaluation, and judgement - see MPEP 2106.04.)
Step 2A, Prong 2: The claim does not recite additional elements that integrate the judicial exception into a practical application.
Step 2B: The claim does not recite additional elements that amount to significantly more than the judicial exception.
With respect to Claim 10:
Step 2A, Prong 1: Inherits the limitations and abstract ideas from Claim 1. An additional judicial exception is recited in the claim as it recites mental processes, which are abstract ideas:
“wherein at least the step of determining the drift difference between the first reconstruction loss and the second reconstruction loss is performed asynchronously with using the particular machine learning model to make a prediction for data along the first set of dimensions.” (Covers mental concepts that could be practically performed in the human mind or through the use of a pencil and paper, including observation, evaluation, and judgement - see MPEP 2106.04.)
Step 2A, Prong 2: The claim does not recite additional elements that integrate the judicial exception into a practical application.
Step 2B: The claim does not recite additional elements that amount to significantly more than the judicial exception.
With respect to Claim 11:
Step 2A, Prong 1: Inherits the limitations and abstract ideas from Claim 1. An additional judicial exception is recited in the claim as it recites mental processes, which are abstract ideas:
“wherein at least the step of determining the drift difference between the first reconstruction loss and the second reconstruction loss is performed in response to a request to use the particular machine learning model to make a prediction for data along the first set of dimensions.” (Covers mental concepts that could be practically performed in the human mind or through the use of a pencil and paper, including observation, evaluation, and judgement - see MPEP 2106.04.)
Step 2A, Prong 2: The claim does not recite additional elements that integrate the judicial exception into a practical application.
Step 2B: The claim does not recite additional elements that amount to significantly more than the judicial exception.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e. changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
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 non-obviousness.
Claims(s) 1-2, 4-8, 11-13, 15-19, and 21-22 are rejected under 35 U.S.C. 103 as being unpatentable over Mopur et al., (Patent Application No. US20220215289A1 filed on April 22, 2021, hereinafter “Mopur”), in view of Sitaramagiridharganesh et al., (Patent Application No. US20230376825A1 filed on May 18, 2022, hereinafter “Sitaramagiridharganesh”).
With respect to Claims 1, 12, and 18:
Mopur teaches:
“A computer-implemented method comprising:
storing a first set of data and a particular machine learning model, (Paragraph 0016 recites an autoencoder, also known as an artificial neural network (particular machine learning model), is trained at a device hosting a training environment, such as a cloud server, using the image training data stored in the training environment (storing a first set of data).)
wherein the particular machine learning model was trained using at least part of the first set of data to predict one or more values along a first set of dimensions, (Paragraph 0016 recites an autoencoder, also known as an artificial neural network (particular machine learning model), is trained at a device hosting a training environment, such as a cloud server, using the image training data (at least part of the first set of data) used to train the machine learning (ML) model without any anomalies. The autoencoder is then trained until it is able to reconstruct expected output with minimum losses or reconstructions errors. Paragraph 0013 recites that once trained, the ML models are then deployed to predicted events and/or values associated with the events. It is understood in machine learning that for each image, a prediction value is outputted (to predict one or more values along a first set of dimensions).)
wherein the first set of data comprises a plurality of combinations of value occurrences in the first set dimensions;” (Paragraph 0013 recites that labeled image data (first set of data) is used to train the ML model that outputs prediction values (first set of dimensions). It is understood that labeled data usually consists of many training instances represented by multiple feature values (a plurality of combinations of value occurrences).
“generating a first compressed set of data by compressing particular data from the first set of data to a second set of dimensions, (Paragraph 0027 recites an autoencoder that is pre-trained on the cloud server to compress and encode image data (first set of dimensions) into fewer dimensions, wherein an encoded representation of the image data (second set of dimensions) with fewer dimensions is then outputted in a latent space (generating a first compressed set of data by compressing particular data from the first set of data to a second set of dimensions).)
wherein the second set of dimensions has fewer dimensions than the first set of dimensions;” (Paragraph 0027 recites during the process of compression, the autoencoder learns to compress the initial image data (first set of dimensions) into fewer dimensions, wherein an encoded representation of the image data (second set of dimensions) with fewer dimensions is then outputted in a latent space.)
“generating a first reconstructed set of data by decompressing the first compressed set of data to the first set of dimensions;” (Paragraph 0027 recites after the process of compression, the autoencoder decompresses and reconstructs the image data back from its compressed and encoded representation (first compressed set of data). The image data is reconstructed such that is it as close as possible to the image data provided to the autoencoder (to the first set of dimensions).)
“determining a first reconstruction loss between the first reconstructed set of data and the particular data based at least in part on differences between the first reconstructed set of data and the particular data along the first set of dimensions;” (Paragraph 0028 recites that during the reconstruction of the image (first reconstructed data), a “reconstruction error capturing unit” captures reconstruction error losses occurring during the reconstruction of each of the original images received by the edge device (particular data along the first set of dimensions) over a period of time (determining a first reconstruction loss between the first reconstructed set of data and the particular data).)
“using the particular machine learning model to make a prediction for data along the first set of dimensions;” (Paragraph 0013 recites an autoencoder, also known as an artificial neural network (particular machine learning model), that once trained, is then deployed to predict events and/or values associated with the events, operating on the same input image data (first set of dimensions) on which it was trained (to make a prediction for data along the first set of dimensions).)
“generating a second compressed set of data by compressing a second set of data to the second set of dimensions;” (Paragraph 0023 recites the existence of new images (second set of data) that include varying information compared to the initial images (first set of data) used to train the ML model. The new images are received from an image source by the edge device in repeatable cycles and may be provided to an autoencoder for reconstruction. Paragraph 0028 further clarifies that the autoencoder learns to compress the image data into fewer dimensions (generating a second compressed set of data by compressing a second set of data), wherein the encoded representation of the image data is present in a latent space (to the second set of dimensions).)
“generating a second reconstructed set of data by decompressing the second compressed set of data to the first set of dimensions;” (Paragraph 0027 recites after the process of compression, the autoencoder decompresses and reconstructs the image data back from its compressed and encoded representation (generating a second reconstructed set of data by decompressing the second compressed set of data to the first set of dimensions).)
“determining a second reconstruction loss between the second reconstructed set of data and the second set of data based at least in part on differences between the second reconstructed set of data and the second set of data along the first set of dimensions;” (Paragraph 0028 recites that during the reconstruction of the image (second reconstructed data), a “reconstruction error capturing unit” captures reconstruction error losses occurring during the reconstruction of each of the original images received by the edge device (second set of data along the first set of dimensions) over a period of time (determining a second reconstruction loss between the second reconstructed set of data and the second set of data).)
“determining a drift difference between the first reconstruction loss and the second reconstruction loss,” (Paragraph 0016 recites an autoencoder outputting data comprising stabilized error (loss) values after training within the watermarks called baseline data (first reconstruction loss). Paragraph 0017 further recites that during the operation of the autoencoder, data losses occurring during reconstruction of the images are captured as reconstruction errors (second reconstruction loss). Paragraph 0036 recites the data drift detection unit detects data drift by assessing densities of the clusters in a temporal manner, where a change in density of cluster with reference to the baseline data, for a period of time, is indicative of data drift. Paragraph 0038 further recites outputs obtained through auto-correlation is analyzed with reference to set threshold values in order to determine the data drift (determining a drift difference).)
Mopur does not appear to explicitly disclose:
“and including the drift difference in an aggregate drift difference;”
“and storing the aggregate drift difference in association with the particular machine learning model, and determining whether to retrain the particular machine learning model based at least in part on one or more conditions that are based at least in part on the aggregate drift difference.”
However, Sitaramagiridharganesh teaches:
“and including the drift difference in an aggregate drift difference;” (Paragraph 0033 recites aggregated data drift scores where an overall drift score consists of aggregated values over a period.)
“and storing the aggregate drift difference in association with the particular machine learning model, and determining whether to retrain the particular machine learning model based at least in part on one or more conditions that are based at least in part on the aggregate drift difference.” (Paragraph 0061 recites a deployed AI model, “M0”, configured for adaptive training, where drift algorithms are applied on the model and data, resulting in various computed aggregated data drift scores. The computation of drift scores also gives data drift flags, including a “retraining” flag, where based on the drift scores and drift flags, retraining is required or is not required.)
It would have been obvious to a person having ordinary skill in the art (PHOSITA) to combine the teachings of Mopur with the teachings of Sitaramagiridharganesh, which are both in the same field of invention. A PHOSITA would have been motivated to combine the autoencoder aided reconstruction loss comparison technique from Mopur with the aggregation and threshold retraining framework from Sitaramagiridharganesh to base retraining decisions on aggregated drift measurements over multiple samples instead of a single reconstruction loss reading, which would in turn create a more computationally efficient system and decrease using resources by avoiding unnecessary retraining of machine learning models.
With respect to Claims 2, 13, and 19:
Mopur and Sitaramagiridharganesh combined teach:
“wherein at least a first dimension of the second set of dimensions comprises a distance from a hyperplane covering a selected combination of value occurrences of the first set of data; (Paragraph 0027 from Mopur recites an autoencoder that is pre-trained on the cloud server to compress and encode image data, where during the process of compression, encoding, and reconstruction, the autoencoder learns to compress the image data into fewer dimensions, wherein the encoded representation of the image data is present in a latent space. It is understood that a hyperplane is a linear subspace defined by a linear equation that symbolizes the linear transformation performed by a neural network layer prior to the application of a nonlinear activation function. The autoencoder computes a weighted linear combination of input values at each layer. Under the broadest reasonable interpretation, the output of the linear transformation would be a value relative to a hyperplane defined by the layer’s weights and biases (distance from a hyperplane) derived from the weights input values (selected combination of value occurrences) of the first set data).)
and wherein a second dimension of the second set of dimensions is selected to be orthogonal to the first dimension.” (Paragraph 0027 from Mopur recites an autoencoder that is pre-trained on the cloud server to compress and encode image data, where during the process of compression, encoding, and reconstruction, the autoencoder learns to compress the image data into fewer dimensions, wherein the image is reconstructed such that it is as close as possible to the image data provided to the autoencoder, and the encoded representation of the image data is present in a latent space. It is understood that the autoencoder is trained to preserve the maximum amount of information using the fewest possible dimensions, meaning learned dimensions are required to be as independent and non-overlapping as possible. Under the broadest reasonable interpretation, orthogonality includes independence between the compressed dimensions, akin to the second dimension of the second set of dimensions selected to be orthogonal to the first dimension.)
With respect to Claims 4, 15, and 21:
Mopur and Sitaramagiridharganesh combined teach:
“wherein the second set of dimensions is different from the first set of dimensions, (Paragraph 0027 from Mopur recites during the process of compression, the autoencoder learns to compress the initial image data (first set of dimensions) into fewer dimensions, wherein an encoded representation of the image data (second set of dimensions) with fewer dimensions is then outputted in a latent space.)
wherein generating the first compressed set of data uses a neural network to compress the first set of data based on one or more feature embedding vectors that describe the first set of data, (Paragraph 0027 from Mopur recites an autoencoder (neural network) that is pre-trained on the cloud server to compress and encode image data (generating the first compressed set of data), wherein the encoded representation of the image data (first set of data) is present in a latent space. Under the broadest reasonable interpretation, a latent space representation produced by a neural network encoder consists of learned vectors that encode the features of the input from the first set of image data, akin to a feature embedding vector, wherein the vector is the compressed representation.)
and wherein generating the second compressed set of data uses the neural network to compress the second set of data based on one or more feature embedding vectors that describe the second set of data.” (Paragraph 0023 from Mopur recites the existence of new images (second set of data) that include varying information compared to the initial images (first set of data) used to train the ML model. The new images are received from an image source by the edge device in repeatable cycles and may be provided to an autoencoder for reconstruction. Paragraph 0028 from Mopur further clarifies that the autoencoder (neural network) learns to compress the image data into fewer dimensions (generating a second compressed set of data), wherein the encoded representation of the image data (second set of data) is present in a latent space. Under the broadest reasonable interpretation, a latent space representation produced by a neural network encoder consists of learned vectors that encode the features of the input from the second set of image data, akin to a feature embedding vector, wherein the vector is the compressed representation.)
With respect to Claim 5:
Mopur and Sitaramagiridharganesh combined teach:
“wherein each dimension of the second set of dimensions is selected to account for a maximum remaining variance in the first set of data.” (Paragraph 0027 recites during the process of compression, the autoencoder learns to compress the initial image data (first set of dimensions) into fewer dimensions, wherein an encoded representation of the image data (second set of dimensions) with fewer dimensions is then outputted in a latent space. It is understood that a dimensionality-reduction technique learns to select dimensions to preserve as much information from the original data as possible within a fixed, reduced number of dimensions. Generating minimum reconstruction errors at a fixed, reduced dimensionality is akin to capturing maximum variance that is obtainable from the first set of data (wherein each dimension of the second set of dimensions is selected to account for a maximum remaining variance in the first set of data).)
With respect to Claims 6 and 16:
Mopur and Sitaramagiridharganesh combined teach:
“receiving a request to train a machine learning model on the first set of data;” (Paragraph 0035 from Sitaramagiridharganesh recites a computing device or server selects the best artificial intelligence model for a the given dataset, and then trains the AI model using hyperparameter search. Under the broadest reasonable interpretation, a computing device or server selecting an AI model is equivalent to receiving a request to train a machine learning model on a first set of data.)
“in response to the request, training the particular machine learning model;” (Paragraph 0035 from Sitaramagiridharganesh recites a computing device or server selects the best artificial intelligence model for a the given dataset, and then in response, trains the AI model using hyperparameter search. Under the broadest reasonable interpretation, a computing device or server selecting an AI model and then training it in response is equivalent to training a machine learning model in response to the request.)
“wherein performing said generating the first compressed set of data, said generating the first reconstructed set of data, and determining the first reconstruction loss is performed automatically in response to training the particular machine learning model.” (Paragraph 0016 from Mopur recites training a machine learning model (autoencoder) with image training data for classification, where the autoencoder is trained and compresses the image training data (generating the first compressed set of data) until it can reconstruct expected output with minimum losses or reconstruction errors (generating the first reconstructed set of data) and further outputs Baseline data comprising stabilized reconstruction error (loss) values (determining the first reconstruction loss) after training and uses the Baseline data as a reference (performed automatically in response to training the particular machine learning model).)
With respect to Claim 7:
Mopur and Sitaramagiridharganesh combined teach:
“determining, based at least in part on the aggregate drift difference, that the one or more conditions are not satisfied, and, without retraining the particular machine learning model, outputting a retraining score that indicates how close the one or more conditions are to being satisfied.” (Paragraph 0038 from Sitaramagiridharganesh recites the computing of a retraining score based on aggregated drift scores and a retraining flag value of either True or False (determining, based at least in part on the aggregate drift difference, that one or more conditions are not satisfied). Paragraph 0063 from Sitaramagiridharganesh further recites an example where aggregated drift scores and an outputted retraining score trigger data drift flags equal to “False” values, meaning retraining of the machine learning model is not required (without retraining the particular machine learning model, outputting a retraining score that indicates how close the one or more conditions are to be satisfied).)
With respect to Claims 8, 17, and 22:
Mopur and Sitaramagiridharganesh combined teach:
“determining, based at least in part on the aggregate drift difference, that the one or more conditions are not satisfied, and, without retraining the particular machine learning model, outputting an aggregate drift difference specific to one or more of the first set of dimensions” (Paragraph 0038 from Sitaramagiridharganesh recites the computing of a retraining score based on aggregated drift scores and a retraining flag value of either “True” or “False” (determining, based at least in part on the aggregate drift difference, that one or more conditions are not satisfied). Paragraph 0063 from Sitaramagiridharganesh further recites an example where an aggregated data drift score and an outputted retraining score trigger data drift flags equal to “False” values, meaning retraining of the machine learning model is not required. Paragraph 0032 from Sitaramagiridharganesh further clarifies that individual data drift scores are computed for each data drift function, where each one evaluates specific data dimensions. The individual scores are then inputted into an aggregation formula. Under the broadest reasonable interpretation, an aggregated drift difference value is computed (outputted) by passing the drift scores as inputs into an aggregation computation, wherein the value then exists within the computing system (without retraining the particular machine learning model, outputting an aggregate drift difference specific to one or more of the first set of dimensions).)
With respect to Claim 11:
Mopur and Sitaramagiridharganesh combined teach:
“wherein at least the step of determining the drift difference between the first reconstruction loss and the second reconstruction loss is performed in response to a request to use the particular machine learning model to make a prediction for data along the first set of dimensions.” (Paragraph 0061 from Mopur recites an autoencoder outputting data comprising stabilized error (loss) values after training within the watermarks called baseline data (first reconstruction loss). Paragraph 0017 from Mopur further recites that during the operation of the autoencoder, data losses occurring during reconstruction of the images are captured as reconstruction errors (second reconstruction loss). Paragraph 0036 from Mopur recites the data drift detection unit detects data drift by assessing densities of the clusters in a temporal manner, where a change in density of cluster with reference to the baseline data, for a period of time, is indicative of data drift. Paragraph 0038 from Mopur further recites outputs obtained through auto-correlation is analyzed with reference to set threshold values in order to determine the data drift (determining a drift difference). Paragraph 0035 from Sitaramagiridharganesh recites a computing device or server selects the best artificial intelligence model for the given dataset, and then trains the AI model using hyperparameter search. Under the broadest reasonable interpretation, a computing device or server selecting an AI model is equivalent to receiving a request to train a machine learning model on a first set of data. After training, the artificial intelligence model is then deployed to predict events and/or values associated with the events, operating on the same input image data (first set of dimensions) on which it was trained (to make a prediction for data along the first set of dimensions).)
Claim(s) 3, 14, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Mopur et al., (Patent Application No. US20220215289A1 filed on April 22, 2021, hereinafter “Mopur”), in view of Sitaramagiridharganesh et al., (Patent Application No. US20230376825A1 filed on May 18, 2022, hereinafter “Sitaramagiridharganesh”), in further view of Tang et al., (Patent Application No. US10635519B1 filed on November 30, 2017, hereinafter “Tang”).
With respect to Claims 3, 14, and 20:
Mopur and Sitaramagiridharganesh combined do not appear to explicitly disclose:
“wherein generating the first compressed set of data uses principal component analysis to compress the first set of data, and wherein generating the second compressed set of data uses the principal component analysis to compress the second set of data”
However, Tang teaches:
“wherein generating the first compressed set of data uses principal component analysis to compress the first set of data, and wherein generating the second compressed set of data uses the principal component analysis to compress the second set of data” (Column 4, Lines 33-34 recite applying principal component analysis (PCA) techniques to a set of training data vectors (generating a first compressed set of data). Column 19, Lines 5-7 recite transforming (through PCA) each observed data vector from the observed coordinate space to a transformed coordinate space (generating a second set of compressed data). Column 17, Lines 57-59 recite during the application of PCA, the platform may also reduce the dimensionality of the transformed coordinate space, which is akin to compressing a set of data.)
It would have been obvious to a PHOSITA before the effective filing date of the present application to implement a method that utilized the teachings of Mopur and Sitaramagiridharganesh with the teachings of Tang, which are all in the same field of invention. A PHOSITA would have been motivated to modify the compression and reconstruction loss techniques from Mopur and Sitaramagiridharganesh with the principal component analysis (PCA) pre-processing technique from Tang in order to transform and decorrelate data before compressing it into a reduced set of dimensions, which would produce a more accurate reduced-dimension representation for drift detection based on reconstruction loss.
Claim(s) 9-10 are rejected under 35 U.S.C. 103 as being unpatentable over Mopur et al., (Patent Application No. US20220215289A1 filed on April 22, 2021, hereinafter “Mopur”), in view of Sitaramagiridharganesh et al., (Patent Application No. US20230376825A1 filed on May 18, 2022, hereinafter “Sitaramagiridharganesh”), in further view of Patton et al., (Patent Application No. US10353685B2 filed on January 3, 2019, hereinafter “Patton”).
With respect to Claim 9:
Mopur and Sitaramagiridharganesh combined teach:
“determining, based at least in part on the aggregate drift difference, that the one or more conditions are satisfied;” (Paragraph 0038 from Sitaramagiridharganesh recites the computing of a retraining score based on aggregated drift scores and a retraining flag value of either “True” or “False”, where “True” meaning the retraining of the AI model is required (determining, based at least in part on the aggregate drift difference, that one or more conditions are satisfied).)
Mopur and Sitaramagiridharganesh do not appear to explicitly disclose:
“based at least in part on determining that the one or more conditions are satisfied, scheduling a retraining of the particular machine learning model based at least in part on a workload that uses the particular machine learning model;”
“and retraining the particular machine learning model based at least in part on determining which particular dimensions to include from a superset of dimensions that includes the first set of dimensions and one or more other dimensions.”
However, Patton teaches:
“based at least in part on determining that the one or more conditions are satisfied, scheduling a retraining of the particular machine learning model based at least in part on a workload that uses the particular machine learning model;” (Column 15, Lines 31-34 recite the ability to build one or more models, wherein the models can be trained on different training data sets. Column 15, Lines 46-50 further clarifies that training of the model is performed when an execution condition is met (determining that one or more conditions are satisfied), such as when enough computing resources are available or anticipated to be available (based at least in part on a workload that uses the particular machine learning model).)
“and retraining the particular machine learning model based at least in part on determining which particular dimensions to include from a superset of dimensions that includes the first set of dimensions and one or more other dimensions.” (Column 15, Lines 31-33 recite candidate models can be built by (trained by) “S200”. Column 15, Lines 39-42 recite “S200” trains the model whenever an execution condition is met, meaning retraining of models occurs. Column 7, Lines 41-45 recite how building or training a model involves specifying the feature sets or raw data to use for training (determining which particular dimensions to include). Column 21, Lines 5-7 recite the data pool used to build or train a model can additionally include the data pool that was used to train a prior model (from a superset of dimensions that includes the first set of dimensions and one or more other dimensions).)
It would have been obvious to a PHOSITA before the effective filing date of the present application to implement a method that utilized the teachings of Mopur and Sitaramagiridharganesh with the teachings of Patton, which are all in the same field of invention. A PHOSITA would have been motivated to modify the drift-based and request-based retraining framework from Mopur and Sitaramagiridharganesh with the resource aware and multiple dimensionality retraining framework from Patton in order to time the execution of retraining AI models around the availability of computing resources and incorporate newly available dimensions into the retrained model, which in turn would improve accuracy of retrained models and decrease the cost associated with retraining models by using limited computational resources.
With respect to Claim 10:
Mopur and Sitaramagiridharganesh combined teach:
“wherein at least the step of determining the drift difference between the first reconstruction loss and the second reconstruction loss is performed asynchronously with using the particular machine learning model to make a prediction for data along the first set of dimensions.” (Paragraph 0061 recites an autoencoder outputting data comprising stabilized error (loss) values after training within the watermarks called baseline data (first reconstruction loss). Paragraph 0017 further recites that during the operation of the autoencoder, data losses occurring during reconstruction of the images are captured as reconstruction errors (second reconstruction loss). Paragraph 0036 recites the data drift detection unit detects data drift by assessing densities of the clusters in a temporal manner, where a change in density of cluster with reference to the baseline data, for a period of time, is indicative of data drift. Paragraph 0038 further recites outputs obtained through auto-correlation is analyzed with reference to set threshold values in order to determine the data drift (determining a drift difference).
Mopur and Sitaramagiridharganesh do not appear to explicitly disclose:
“wherein at least the step of determining the drift difference between the first reconstruction loss and the second reconstruction loss is performed asynchronously with using the particular machine learning model to make a prediction for data along the first set of dimensions.”
However, Patton teaches:
“wherein at least the step of determining the drift difference between the first reconstruction loss and the second reconstruction loss is performed asynchronously with using the particular machine learning model to make a prediction for data along the first set of dimensions.” (Column 2, Lines 60-65 recite the event detection models used (particular machine learning model to make a prediction for data along the first set of dimensions) can perform the detection of events asynchronously. Column 15, Lines 7-15 recite the evaluation system also includes a method for monitoring concept drift. Column 26, Lines 35-40 further recites that one or more instances of the method or processes described can be performed asynchronously together.)
It would have been obvious to a PHOSITA before the effective filing date of the present application to implement a method that utilized the teachings of Mopur and Sitaramagiridharganesh with the teachings of Patton, which are all in the same field of invention. A PHOSITA would have been motivated to modify the drift-based and request-based retraining framework from Mopur and Sitaramagiridharganesh with the asynchronous process execution process from Patton in order to perform the computation of drift difference separately from the model’s real-time predictions in order to avoid degrading prediction latency and performance while continuing to monitor drift in the background.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Vibha Bhat whose telephone number is (571)-272-7091. The examiner can normally be reached on Monday – Thursday from 8:00 AM to 5:00 PM EST and every other Friday from 8:00 AM to 4:00 PM EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. See MPEP § 713.01. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at https://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Mariela Reyes, can be reached at telephone number (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 an application 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://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 (572)-272-1000.
/Vibha Bhat/Examiner
Art Unit 2142
/Mariela Reyes/Supervisory Patent Examiner, Art Unit 2142
Prosecution Insights