DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application is being examined under the pre-AIA first to invent provisions.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 07/23/2025 and 05/05/2026 are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Response to Arguments
101 Rejection Arguments
Applicant asserts:
Applicant argues, on page 11, that the amended claims can not be practically performed in the human mind.
Examiner response:
Examiner disagrees and notes that the cloud based storage, language model, and artificial intelligence model has not been interpreted as an abstract idea, but as a generic computer component merely used to implement mental steps associated with generate a plurality of embeddings, calculate amounts of information in the semantic graph, identify a second number of dimensions, fewer than first number of dimensions, projecting the semantic graph onto second embedding space, select a first projection from projected graph, measure a first distance between first projection and second projection and a second distance between first projection and third projection, compute a first likelihood that the first projection is assigned a first preassigned label and a second likelihood the first projection is assigned a second preassigned label. The displaying a recommendation step is interpreted as insignificant extra-solution. The claims do no describe how the cloud based storage, language model, and artificial intelligence model perform the mental steps. Examiner notes that the claims do not describe what occurs to the mental steps in a way that would differentiate this process from the way a person could perform them mentally.
The claimed do not provide an improvement to the technology, instead the examiner’s analysis is similar to Recentive Analytics because the claims use machine learning at a high level.
103 Rejection Arguments
Applicant asserts:
Applicant argues, on page 12-13, that the prior art does not teach the amended claim 1.
Examiner response:
Applicant’s arguments with respect to claim(s) 1 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
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 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
In reference to claim 1:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a manufacture
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“project the semantic graph into a second embedding space having the second number of dimensions to generate a projected graph” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could project the semantic graph into a second embedding space having the second number of dimensions.
“select a first projection from the projected graph, corresponding to a first datapoint in the semantic graph representing a first message from the second set of messages lacking a label” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could select a first projection corresponding to a first datapoint representing a first message lacking a label.
“measure a first distance between the first projection and a second projection from the projected graph corresponding to a second datapoint in the semantic graph representing a second message from the first set of messages having a first preassigned label and a second distance between the first projection and a third projection from the projected graph corresponding to a third datapoint in the semantic graph representing a third message from the first set of messages having a second preassigned label;” which is an abstract idea because it is directed to a mathematical relationships, mathematical formulas or equations, and mathematical calculations. (MPEP 2106.04(a)(2)(I)(C)).
“and compute a first likelihood that the first projection is to be assigned the first preassigned label based on the first distance and a second likelihood that the first projection is to be assigned the second preassigned label based on, the second distance” which is an abstract idea because it is directed to a mathematical relationships, mathematical formulas or equations, and mathematical calculations. (MPEP 2106.04(a)(2)(I)(C)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
“A system for propagating labels through a sparsely labeled dataset using a supervised projection of a semantic embedding, the system comprising: cloud-based storage that stores (i) a plurality of messages including a first set of messages having labels and a second set of messages, greater in quantity than the first set of messages, lacking labels, (ii) a language model trained using training data separate from the plurality of messages, and(iii)an artificial intelligence model trained to output optimal sets of dimensions for labeling tasks based on inputted semantic graphs;” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“cloud-based control circuitry configured to:” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“input the plurality of messages into the language model to generate a plurality of embeddings representing the plurality of messages in a first embedding space having a first number of dimensions and a semantic graph comprising a plurality of datapoints representing the plurality of messages connected based on a semantic similarity of each message to each other message, the first number of dimensions being less than or equal to a quantity of unique tokens included within the plurality of messages;” (insignificant extra-solution activity mere data gathering MPEP 2106.05(g))
“input the semantic graph and a labeling task into the artificial intelligence model to (a) calculate amounts of information in the semantic graph represented by each dimension in the first number of dimensions and (b) identify a second number of dimensions, fewer than the first number of dimensions, by selecting dimensions in the first number of dimensions that represent respective amounts of information exceeding a threshold amount, wherein the threshold amount represents a fraction of a total amount of information present in the semantic graph” (insignificant extra-solution activity mere data gathering MPEP 2106.05(g))
“and cloud-based input/output circuitry: that displays, on a user interface, a recommendation to assign the first preassigned label to the first projection based on the first likelihood being greater than the second likelihood or assign the second preassigned label to the first projection based on the second likelihood being greater than the first likelihood.” (insignificant extra-solution activity mere data gathering MPEP 2106.05(g))
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
“A system for propagating labels through a sparsely labeled dataset using a supervised projection of a semantic embedding, the system comprising: cloud-based storage that stores (i) a plurality of messages including a first set of messages having labels and a second set of messages, greater in quantity than the first set of messages, lacking labels, (ii) a language model trained using training data separate from the plurality of messages, and(iii)an artificial intelligence model trained to output optimal sets of dimensions for labeling tasks based on inputted semantic graphs;” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“cloud-based control circuitry configured to:” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“input the plurality of messages into the language model to generate a plurality of embeddings representing the plurality of messages in a first embedding space having a first number of dimensions and a semantic graph comprising a plurality of datapoints representing the plurality of messages connected based on a semantic similarity of each message to each other message, the first number of dimensions being less than or equal to a quantity of unique tokens included within the plurality of messages;” (well-understood, routine, conventional MPEP 2106.05(d))
“input the semantic graph and a labeling task into the artificial intelligence model to (a) calculate amounts of information in the semantic graph represented by each dimension in the first number of dimensions and (b) identify a second number of dimensions, fewer than the first number of dimensions, by selecting dimensions in the first number of dimensions that represent respective amounts of information exceeding a threshold amount, wherein the threshold amount represents a fraction of a total amount of information present in the semantic graph” (well-understood, routine, conventional MPEP 2106.05(d))
“and cloud-based input/output circuitry: that displays, on a user interface, a recommendation to assign the first preassigned label to the first projection based on the first likelihood being greater than the second likelihood or assign the second preassigned label to the first projection based on the second likelihood being greater than the first likelihood.” (well-understood, routine, conventional MPEP 2106.05(d))
In reference to claim 2:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“projecting the semantic graph into a second embedding space having the second number of dimensions to generate a projected graph” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could project the semantic graph into a second embedding space having the second number of dimensions.
“measuring (a) a first distance between (_i)a first projection from the projected graph corresponding to a first datapoint in the semantic graph representing a first message in the plurality of messages lacking a label and fii) a second projection in the projected graph corresponding to a second datapoint in the semantic graph representing a second message in the plurality of messages corresponding to, a first preassigned label, and (b) a second distance between the first projection and a third projection from the projected graph corresponding to a third datapoint in the semantic graph representing a third message in the plurality of messages having a second preassigned label;” which is an abstract idea because it is directed to a mathematical relationships, mathematical formulas or equations, and mathematical calculations. (MPEP 2106.04(a)(2)(I)(C)).
“and computing a first likelihood that the first projection is to be assigned the first preassigned label based on the first distance and a second likelihood that the first projection is to be assigned the second preassigned label based on, the second distance” which is an abstract idea because it is directed to a mathematical relationships, mathematical formulas or equations, and mathematical calculations. (MPEP 2106.04(a)(2)(I)(C)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
“A method inputting a plurality of messages into a language model to generate embeddings in a first embedding space having a first number of dimensions and a semantic graph comprising datapoints representing the plurality of messages connected based on a semantic similarity” (insignificant extra-solution activity mere data gathering MPEP 2106.05(g))
“inputting the semantic graph and a labeling task into the artificial intelligence model to identify a second number of dimensions that is less than the first number of dimensions, wherein each dimension of the second number of dimensions represents a portion of a total amount of information present in the semantic graph exceeding a threshold amount of information;” (insignificant extra-solution activity mere data gathering MPEP 2106.05(g))
“based on the first likelihood being greater than the second likelihood, storing, in electronic storage, the first preassigned label in association with the first message represented by the first projection.” (insignificant extra-solution activity mere data gathering MPEP 2106.05(g))
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
“A method inputting a plurality of messages into a language model to generate embeddings in a first embedding space having a first number of dimensions and a semantic graph comprising datapoints representing the plurality of messages connected based on a semantic similarity” (well-understood, routine, conventional MPEP 2106.05(d))
“inputting the semantic graph and a labeling task into the artificial intelligence model to identify a second number of dimensions that is less than the first number of dimensions, wherein each dimension of the second number of dimensions represents a portion of a total amount of information present in the semantic graph exceeding a threshold amount of information;” (well-understood, routine, conventional MPEP 2106.05(d))
“based on the first likelihood being greater than the second likelihood, storing, in electronic storage, the first preassigned label in association with the first message represented by the first projection.” (well-understood, routine, conventional MPEP 2106.05(d))
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 3:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The method of claim 2, further comprising: determining corresponding variances for each dimension of the first number of dimensions, wherein a variance represents a measure of shared information between a respective dimension and the first number of dimensions; and” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could determine corresponding variances for each dimension of the first number of dimensions.
“comparing the corresponding variances to a variance threshold related to the threshold amount of information to identify the second number of dimensions.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could compare the corresponding variances to a variance threshold related to the threshold amount of information to identify the second number of dimensions.
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
In reference to claim 4:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The method of claim 2, further comprising: in response to the first likelihood being greater than the second likelihood, generating for display, on a user interface, a recommendation to assign the first preassigned label in association with the first message represented by the first projection; ” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)). For example, a person could generate a recommendation to assigned the first preassigned label in association with the first message represented by the first projection.
“and based on a review received by the user interface, assigning the first preassigned label in association with the first message represented by the first projection.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)). For example, a person could assign the first preassigned label in association with the first message represented by the first projection.
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 5:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The method of claim 2, wherein computing the second likelihood that the first projection is to be assigned the second preassigned label comprises: determining a third likelihood that the second projection is to be assigned the second preassigned label;” which is an abstract idea because it is directed to a mathematical relationships, mathematical formulas or equations, and mathematical calculations. (MPEP 2106.04(a)(2)(I)(C)).
“and determining the second likelihood based on the third likelihood.” which is an abstract idea because it is directed to a mathematical relationships, mathematical formulas or equations, and mathematical calculations. (MPEP 2106.04(a)(2)(I)(C)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
In reference to claim 6:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The method of claim 2, further comprising: determining a composite likelihood from the first likelihood and the second likelihood;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could determine a composite likelihood from the first and second likelihood.
“determining a first entropy score from the composite likelihood, wherein the first entropy score indicates randomness of label components in the composite likelihood;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could determine a first entropy score from the composite likelihood.
“comparing the first entropy score to a first threshold entropy score;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could compare the first entropy score to a first threshold entropy score.
“in response to comparing the first entropy score to the first threshold entropy score, generating for display, on a user interface, a recommendation to review the first preassigned label being assigned to the first projection.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)). For example, a person could generate for display a recommendation to review the first preassigned label being assigned to the first projection.
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 7:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The method of claim 2, further comprising: determining a first predicted label for the first projection;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could determine a first predicted label for the first projection.
“determining a first consistency score for the first projection based on a comparison of the first predicted label and the first preassigned label, wherein the first consistency score indicates a degree of consensus between the first predicted label and the first preassigned label;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could determine a first consistency score the first projection.
“comparing the first consistency score to a first threshold consistency score;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could compare the first consistency score to a first threshold consistency score.
“in response to comparing the first consistency score to the first threshold consistency score, filtering the first projection to a first group;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could filter the first projection to a first group based on the comparison.
“generating for display, on a user interface, a recommendation to use the first group as a training sample for a supervised learning task.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could generate for display, on a user interface, a recommendation to use the first group as a training sample for a supervised learning task.
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 8:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The method of claim 2, further comprising: determining a first outlier score based on the first distance;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could determine a first outlier score based on the first distance.
“comparing the first outlier score to a first threshold outlier score;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could compare the first outlier score to a first threshold outlier score.
“and selecting a recommendation from a plurality of recommendations based on comparing the first outlier score to a first threshold outlier score.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could select a recommendation based on the comparison of the first outlier score.
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
In reference to claim 9:
Claim 9 is directed to a judicial exception from claim(s) depended on and does not recite additional elements that integrate the judicial exception into a practical application and amount to significantly more than the judicial exception.
In reference to claim 10:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The method of claim 2, further comprising based on the first likelihood being greater than the second likelihood, assigning the first preassigned label to the first projection” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could assign the first label to the first projection.
“assigning the first preassigned label to the first datapoint” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could assign the first label to the first datapoint.
“and assigning the first preassigned label to the first message” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could assign the first label to the first message.
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
In reference to claim 11:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The method of claim 9, wherein the first subset comprises the first message having the first preassigned label, and wherein the method further comprises: comparing the first preassigned label to a corresponding ground truth label from the given labels of the first subset;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could compare the first label to a corresponding given label from the given labels of the first subset.
“and determining an evaluation of the artificial intelligence model in response to comparing the first preassigned label to the corresponding ground truth label.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could determine an evaluation of the artificial intelligence model in response to the comparison between first label and corresponding given label.
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
In reference to claim 12:
Claim 12 is directed to a judicial exception from claim(s) depended on and does not recite additional elements that integrate the judicial exception into a practical application and amount to significantly more than the judicial exception.
In reference to claim 13:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The method of claim 12, further comprising: determining the total amount of information present in the semantic graph, wherein the total amount of information is unevenly distributed among the original dimensions;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could determine an amount of information present in the semantic graph.
“determining a first dimension from the original dimensions, wherein the first dimension has a first amount of information, comprising a portion of the total amount of information present in the semantic graph;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could determine a first optimal dimension from the original dimensions.
“comparing the first amount of information to the threshold amount of information;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could compare the first amount of information to a cut-off score.
“and in response to comparing the first amount of information to the threshold amount of information, including the first dimension in the second number of dimensions.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could include/group the first optimal dimension in the optimal set of dimensions.
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
In reference to claim 14:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The method of claim 13, wherein determining the first dimension from the original dimensions comprises: determining an original vector for each original dimension of the original dimensions;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could determine an original vector for each original dimension.
“determining a correlation between the original vector and all other original vectors, wherein the correlation comprises a measure of shared information;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could determine a correlation between the original vector and all other original vectors.
“determining a first vector based on the correlation;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could determine first optimal vector based on the correlation.
“and determining the first dimension from the first vector.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could determine the first optimal dimension from the first optimal vector.
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
In reference to claim 15:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The method of claim 13, further comprising: determining a optimal dimension from the original dimensions, wherein the second dimension has a second amount of information, and wherein the second amount of information is less than the first amount of information;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could determine a second optimal dimension from the original dimensions.
“combining the first amount of information and the second amount of information into a running total amount of information;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could combine the first and second amount of information into a running total amount of information.
“comparing the running total amount of information to the threshold amount of information;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could compare the running total amount of information to the cut-off score.
“and in response to comparing the running total amount of information to the threshold amount of information, discarding the second dimension.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could discard/not consider the second optimal dimension.
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
In reference to claim 16:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a process
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“and determining the threshold amount of information based on the first user input.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could determine the cut-off score based on the input.
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
“The method of claim 13, further comprising: receiving a first user input;” (insignificant extra-solution activity mere data gathering MPEP 2106.05(g))
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
“The method of claim 13, further comprising: receiving a first user input;” (well-understood, routine, conventional MPEP 2106.05(d))
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 17:
Claim 17 is directed to a judicial exception from claim(s) depended on and does not recite additional elements that integrate the judicial exception into a practical application and amount to significantly more than the judicial exception.
In reference to claim 18:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a machine
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“projecting the semantic graph into a second embedding space having the second number of dimensions to generate a projected graph” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could project the semantic graph into a second embedding space having the second number of dimensions.
“measuring (a) a first distance between (_i)a first from the projected graph representing a first message in the messages lacking a label and fii) a second projection in the projected graph representing a second message in the messages corresponding to a first preassigned label, and (b) a second distance between the first projection and a third projection from the projected graph representing a third message in the messages having a second preassigned label;” which is an abstract idea because it is directed to a mathematical relationships, mathematical formulas or equations, and mathematical calculations. (MPEP 2106.04(a)(2)(I)(C)).
“and computing a first likelihood that the first projection is to be assigned the first preassigned label based on the first distance and a second likelihood that the first projection is to be assigned the second preassigned label based on, the second distance” which is an abstract idea because it is directed to a mathematical relationships, mathematical formulas or equations, and mathematical calculations. (MPEP 2106.04(a)(2)(I)(C)).
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
“One or more non-transitory, computer readable media storing computer instructions which, when executed by one or more computer processors, cause the one or more computer processors to:” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“inputting messages into a language model to generate embeddings in a first embedding space having a first number of dimensions and;” (insignificant extra-solution activity mere data gathering MPEP 2106.05(g))
“inputting the semantic graph and a labeling task into an artificial intelligence model to identify a second number of dimensions from the first number of dimensions, wherein each dimension of the second number of dimensions comprises a portion of information present in the semantic graph exceeding a threshold;” (insignificant extra-solution activity mere data gathering MPEP 2106.05(g))
“based on comparing the first likelihood and the second likelihood, storing, in electronic storage, the first preassigned label in association with the first message represented by the first projection.” (insignificant extra-solution activity mere data gathering MPEP 2106.05(g))
The claim does not include additional elements that are integrated into a practical application.
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
“One or more non-transitory, computer readable media storing computer instructions which, when executed by one or more computer processors, cause the one or more computer processors to:” is merely reciting the words "apply it" (or an equivalent) with the judicial exception, or merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
“inputting messages into a language model to generate embeddings in a first embedding space having a first number of dimensions and;” (well-understood, routine, conventional MPEP 2106.05(d))
“inputting the semantic graph and a labeling task into an artificial intelligence model to identify a second number of dimensions from the first number of dimensions, wherein each dimension of the second number of dimensions comprises a portion of information present in the semantic graph exceeding a threshold;” (well-understood, routine, conventional MPEP 2106.05(d))
“based on comparing the first likelihood and the second likelihood, storing, in electronic storage, the first preassigned label in association with the first message represented by the first projection.” (well-understood, routine, conventional MPEP 2106.05(d))
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
In reference to claim 19:
Claim 19 is directed to a judicial exception from claim(s) depended on and does not recite additional elements that integrate the judicial exception into a practical application and amount to significantly more than the judicial exception.
In reference to claim 20:
Step 1 - Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is directed to a machine
Step 2A Prong 1 - Does the claim recite an abstract idea, law of nature, or natural phenomenon?
“The one or more non-transitory, computer readable media of claim 19, further comprising: determining a total amount of information present in the semantic graph, wherein the total amount of information is unevenly distributed among the original dimensions;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could determine an amount of information present in the semantic graph.
“determining a first dimension from the original dimensions, wherein the first dimension has a first amount of information, comprising a portion of the information present in the semantic graph;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could determine a first optimal dimension form the original dimensions.
“comparing the first amount of information to the threshold;” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could compare the first amount of information to a cut-off score.
“and in response to comparing the first amount of information to the threshold, including the first dimension in the second number of dimensions.” which is an abstract idea because it is directed to a mental process, an observation, evaluation, judgement, or opinion. The limitation as drafted, and under a broadest reasonable interpretation, can be performed in the human mind, or by a human using a pen and paper (MPEP 2106.04(a)(2)(Ill)(c)). For example, a person could include/group the first optimal dimension in the optimal set of dimensions.
Step 2A Prong 2 - Does the claim recite additional elements that integrate the judicial exception into a practical application?
No
Step 2B - Does the claim recite additional elements that amount to significantly more than the judicial exception?
No
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1 is rejected under 35 U.S.C. 103 as being unpatentable over Kun Cao et al; US 12566973 B1 filed on Sep 27, 2021 (hereinafter “Cao”) in view of Alexander S Polichroniadis et al; US 12112519 B1 filed on Feb 23, 2022 (hereinafter “Polichroniadis”) in further view of bradleyboehmke; “Chapter 17 Principal Component Analysis” available online on Dec 23, 2022 (hereinafter “Bradley”) in further view of Stanislav Olegovich Ponomarev; US 20210042471 A1 filed on Jul 27, 2020 (hereinafter “Ponomarev”) in further view of Ravish Raj; “K-Nearest Neighbors (KNN) Algorithm in Machine Learning” available online Dec 27, 2022 (hereinafter “Raj”) in further view of Sewak; US 20220414137 A1 filed on Apr 1, 2022 (hereinafter “Sewak”) in further view of Rui Zhong et al; US 20230206255 A1 filed on Dec 27, 2021 (hereinafter “Zhong”) in further view of Ahmet et al; “Label Propagation for Deep Semi-Supervised learning” published on 2019 (hereinafter “Ahmet”) in further view of Dane et al; US 20250022615 A1 filed on Nov 14, 2022 (hereinafter “Dane”)
Regarding claim 1, Cao teaches [cloud-based control circuitry configured to:] input the plurality of messages into the language model to generate a plurality of embeddings representing the plurality of messages in a first embedding space having a first number of dimensions and a semantic graph comprising a plurality of datapoints representing the plurality of messages connected based on a semantic similarity of each message to each other message, [the first number of dimensions being less than or equal to a quantity of unique tokens included within the plurality of messages]; (Cao Column 3 Line 3; “Semantics-based tagging can consider the semantic similarity between different instances of contents using one or more language models (LMs).” Cao Fig 5 and Column 7 Line 8; “this process includes at least some amount of textual content. At least an alphanumeric or textual portion of the content can be processed 504 using a language model to infer one or more semantic embeddings for the content, where a semantic embedding can include a numerical vector including values corresponding to various semantic features inferred for the input content.” Examiner notes that a plurality of messages (input content/some amount of textual content) is input into a language model to generate embeddings in a first embedding space having a first number of dimensions (language model generates embeddings corresponding to various semantic features/first embedding space having a first number of dimensions); plurality of datapoints are connected based on a semantic similarity (Semantics-based tagging can consider the semantic similarity between different instances of contents))
Cao does not teach and (iii)an artificial intelligence model trained to output sets of dimensions for labeling tasks based on inputted semantic graphs;
input the semantic graph and a labeling task into the artificial intelligence model to (a) [calculate amounts of information in the semantic graph represented by each dimension in the first number of dimensions and] (b) identify a second number of dimensions, fewer than the first number of dimensions, [by selecting dimensions in the first number of dimensions that represent respective amounts of information exceeding a threshold amount, wherein the threshold amount represents a fraction of a total amount of information present in the semantic graph]
However, Polichroniadis does teach and (iii)an artificial intelligence model trained to output sets of dimensions for labeling tasks based on inputted semantic graphs; (Polichroniadis Column 6 Line 26; “A machine learning method may include machine learning algorithms (such as PCA, clustering algorithms, etc.) and/or machine learning models (such as neural network-based approaches).” Polichroniadis Column 6 Line 39; “One or more machine learning methods may be used to classify objects, include, or add semantic information to a 3D reconstruction. In some implementations, a machine learning method is used to generate semantic labels for pixels of one or more images of the physical environment.” Polichroniadis Column 7 Line 4; “a machine learning method is used to estimate relationships between objects of the semantic mesh or graph representing the semantic mesh. A machine learning method may include machine learning algorithms (such as PCA, clustering algorithms, etc.) and/or machine learning models (such as neural network-based approaches).” Examiner notes that artificial intelligence model trained (machine learning methods) to output sets of dimensions for labeling tasks (generate semantic labels) based on inputted semantic graphs (semantic mesh or graph))
input the semantic graph and a labeling task into the artificial intelligence model to (a) [calculate amounts of information in the semantic graph represented by each dimension in the first number of dimensions and] (b) identify a second number of dimensions, fewer than the first number of dimensions, [by selecting dimensions in the first number of dimensions that represent respective amounts of information exceeding a threshold amount, wherein the threshold amount represents a fraction of a total amount of information present in the semantic graph] (Polichroniadis Column 6 Line 39; “One or more machine learning methods may be used to classify objects, include, or add semantic information to a 3D reconstruction. In some implementations, a machine learning method is used to generate semantic labels for pixels of one or more images of the physical environment.” Polichroniadis Column 7 Line 4; “a machine learning method is used to estimate relationships between objects of the semantic mesh or graph representing the semantic mesh. A machine learning method may include machine learning algorithms (such as PCA, clustering algorithms, etc.) and/or machine learning models (such as neural network-based approaches).” Examiner notes that the semantic graph (graph representing the semantic mesh) is input into an artificial intelligence model (machine learning models) to identify a second number of dimensions that is less than the first number of dimensions (PCA is a dimensionality reduction technique); a labeling task (generate semantic labels) is input into the AI model (machine learning method))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao and Polichroniadis. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. One of ordinary skill would have motivation to combine Cao and Polichroniadis to improve the systems accuracy or efficiency “the machine learning method to further improve its accuracy or efficiency” (Polichroniadis Column 7 Line 30).
Cao in view of Polichroniadis does not teach calculate amounts of information in the semantic graph represented by each dimension in the first number of dimensions and
by selecting dimensions in the first number of dimensions that represent respective amounts of information exceeding a threshold amount, wherein the threshold amount represents a fraction of a total amount of information present in the semantic graph
However, Bradley does teach calculate amounts of information in the semantic graph represented by each dimension in the first number of dimensions and (Bradley Fig 17.6 and Section 17.5.2 Paragraph 1; “The proportion of variance explained (PVE) identifies the optimal number of PCs to keep based on the total variability that we would like to account for. Mathematically, the PVE for the m-th PC is calculated as” Examiner notes that determining corresponding variances for each dimension (calculate PVE for the m-th PC) of the first number of dimensions (all 42 PCs), wherein a variance represents a measure of shared information between respective dimension and a first number of dimensions (Fig 17.6 shows variance as measured of shared information between respective PC of all 42 PCs in PVE graph))
by selecting dimensions in the first number of dimensions that represent respective amounts of information exceeding a threshold amount, wherein the threshold amount represents a fraction of a total amount of information present in the semantic graph (Bradley Section 17.5.2 Paragraph 2; “The first PCt in our example explains 5.46% of the feature variability, and the second principal component explains 5.17%. Together, the first two PCs explain 10.63% of the variability. Thus, if an analyst desires to choose the number of PCs required to explain at least 75% of the variability in our original data then they would choose the first 27 components.” Examiner notes that the corresponding variances (feature variability of PCs) is compared to a variance threshold related to the threshold amount of information (choose the number of PCs required to explain at least 75% of the variability in our original data) to identify the second number of dimensions (choose the first 27 components.))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, and Bradley. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. One of ordinary skill would have motivation to combine Cao, Polichroniadis, and Bradley to find low-dimensional representations of a data set for more efficient data processing “Principal components analysis (PCA) is a method for finding low-dimensional representations of a data set that retain as much of the original variation as possible.” (Bradley Paragraph 1).
Cao in view of Polichroniadis in further view of Bradley does not teach project the semantic graph into a second embedding space having the second number of dimensions to generate a projected graph
However, Ponomarev does teach project the semantic graph into a second embedding space having the second number of dimensions to generate a projected graph (Ponomarev Paragraph 0032; "At operation 304, the semantic graph embedding system 116 constructs a Markov chain. The Markov chain comprises a plurality of nodes. Each node in the plurality of nodes represents a data value in the dataset." Malden Paragraph 0035; "At operation 310, the distance (e.g., path cost) determined in operation 308 is stored as a dimension in a vector… then the vector representation of the target node would be: [A,B,C,D,E]… the semantic graph embedding system 116 reduces the vector (e.g., an n-dimensional vector) to a three-dimensional vector using principal component analysis (PCA)." Examiner notes that the projected graph (three-dimensional vector) is determined by projection (reduction by semantic embedding) the semantic graph into the optimal set of dimension (vector))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, and Ponomarev. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, and Ponomarev to extract the meaning of data in different contexts for more accurate data processing “The semantic graph embedding system may extract the meaning of data in the dataset when the data is used in one particular context versus another context.” (Ponomarev Paragraph 0012).
Cao in view of Polichroniadis in further view of Bradley in further view of Ponomarev does not teach select a first projection from the projected graph, corresponding to a first datapoint in the semantic graph representing a first message from the second set of messages lacking a label
measure a first distance between the first projection and a second projection from the projected graph corresponding to a second datapoint in the semantic graph representing a second message from the first set of messages having a first preassigned label
and a second distance between the first projection and a third projection from the projected graph corresponding to a third datapoint in the semantic graph representing a third message from the first set of messages having a second preassigned label;
and compute a first likelihood that the first projection is to be assigned the first preassigned label based on the first distance and a second likelihood that the first projection is to be assigned the second preassigned label based on, the second distance
[and cloud-based input/output circuitry: that displays, on a user interface, a recommendation] to assign the first preassigned label to the first projection based on the first likelihood being greater than the second likelihood or assign the second preassigned label to the first projection based on the second likelihood being greater than the first likelihood.
However, Raj does teach select a first projection from the projected graph, corresponding to a first datapoint in the semantic graph representing a first message from the second set of messages lacking a label (Raj Section “Working of KNN”; “We need to find the K neighbors based on any distance metric… We will have the test sample on which we want the prediction” Examiner notes that a first projection corresponding to a first datapoint representing a first message lacking a label (test sample on which we want the prediction) is selected)
measure a first distance between the first projection and a second projection from the projected graph corresponding to a second datapoint in the semantic graph representing a second message from the first set of messages having a first preassigned label (Raj Section Working of KNN; “We need to find the K neighbors based on any distance metric… We will have the test sample on which we want the prediction... Among the selected K neighbors, we need to count how many neighbors are from different classes” Examiner notes that a first distance is measured between a first projection from the projected graph corresponding to a first data point represent a first message lacking a label (test sample to be predicted) and a second projection corresponding to a second data point representing a second message corresponding to a first preassigned label (neighbor associated with a class))
and a second distance between the first projection and a third projection from the projected graph corresponding to a third datapoint in the semantic graph representing a third message from the first set of messages having a second preassigned label; (Examiner refers to previous mapping to show a second distance is measured between the first projection (test sample) and a third projection corresponding to a third datapoint representing a third message having a second preassigned label (neighbor associated with different class))
and compute a first likelihood that the first projection is to be assigned the first preassigned label based on the first distance and a second likelihood that the first projection is to be assigned the second preassigned label based on, the second distance (Raj Section Working of KNN; “Among the selected K neighbors, we need to count how many neighbors are from the different classes… Now, we have to assign the test data sample to the class for which the count of neighbors was maximum” Examiner notes that a first likelihood that the first projection is to be assigned the first preassigned label is computed based on the first distance (using the distance metrics, closest neighbors are selected and counted, then out of those neighbors, how many are associated to class 1 are counted; EX 6/10 are class 1) and a second likelihood that the first projection is to be assigned the second preassigned label based on the second distance (using the distance metrics, closest neighbors are selected and counted, then out of those neighbors, how many are associated to class 2 are counted; EX 3/10 are class 2))
[and cloud-based input/output circuitry: that displays, on a user interface, a recommendation] to assign the first preassigned label to the first projection based on the first likelihood being greater than the second likelihood or assign the second preassigned label to the first projection based on the second likelihood being greater than the first likelihood. (Examiner refers to previous mapping to show that based on the first likelihood being greater than the second likelihood (count of neighbors for class 1 is greater than count of neighbors for class 2), storing the first preassigned label in the association with the first message represented by the first projection (test data sample is assigned to class for which the count of neighbors was maximum))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, Ponomarev, and Raj. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. Raj teaches KNN algorithm. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, Ponomarev, and Raj to leverage KNN strengths for label classification “KNN is a very famous algorithm because of its simplicity, so let’s understand the key strengths… zero training time…sample
Cao in view of Polichroniadis in further view of Bradley in further view of Ponomarev in further view of Raj does not teach A system for propagating labels through a sparsely labeled dataset using a supervised projection of a semantic embedding, the system comprising: cloud-based storage that stores: (i) a plurality of messages [including a first set of messages having labels and a second set of messages, greater in quantity than the first set of messages, lacking labels], (ii) a language model [trained using training data separate from the plurality of messages, and (iii)an artificial intelligence model trained to output sets of dimensions for labeling tasks based on inputted semantic graphs;]
and cloud-based input/output circuitry: that displays, on a user interface, a recommendation
However, Sewak does teach A system for propagating labels through a sparsely labeled dataset using a supervised projection of a semantic embedding, the system comprising: cloud-based storage that stores: (i) a plurality of messages [including a first set of messages having labels and a second set of messages, greater in quantity than the first set of messages, lacking labels], (ii) a language model [trained using training data separate from the plurality of messages, and (iii)an artificial intelligence model trained to output sets of dimensions for labeling tasks based on inputted semantic graphs;] (Sewak Fig 1 shows labelling system that comprises cloud-based storage circuitry configure to store (Cloud Service 199 contains storage 180); Sewak Paragraph 0082; "An NLG model taken from repository 162 and employed by labeling service 142 to perform a step in a label scoring service 168 is generally trained over a natural language corpus that is unlabeled." Sewak Paragraph 0083; "method 300 begins at step 303 when the labeling service 142 serves a display page to labeling application 110. At step 305, labeling service 142 receives a text string defining candidate text from a document in corpus 154 or from the labeling application 110. At step 310, the labeling service 142 receives a text string defining a label, e.g. from labeling application 110." Examiner notes that cloud-based circuitry (Cloud service) receives text string defining candidate text and text string defining label meaning it has capabilities to store a plurality of messages; language model (NLG model))
and cloud-based input/output circuitry: that displays, on a user interface, a recommendation (Sewak Paragraph 0056; "A labeling application 110 in the operating environment 100 may present a prompt to the user on a display 120." Examiner notes that in response to the comparing, paragraph 0056 and Fig 1 shows display 120 (user interface) used to display recommendation (prompt))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, and Sewak. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. Raj teaches KNN algorithm. Sewak teaches a method for automatic labeling of text data. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, and Sewak to improve the success rate of classification, while maintaining improved efficiency “The success rate of the classification can be improved, while maintaining this improved efficiency, by obtaining a second generative result from a generative model and estimating label probability using the second generative result.” (Sewak Paragraph 0008).
Cao in view of Polichroniadis in further view of Bradley in further view of Ponomarev in further view of Raj in further view of Sewak does not teach a language model trained using training data separate from the plurality of messages
However, Zhong does teach a language model trained using training data separate from the plurality of messages (Zhong Paragraph 0027; “the natural language processing model 270 is trained on training data 251 which includes historical sentiment data 250, 250H obtained from a sentiment data store 252. As discussed above, the sentiment data 250 may be received as interaction data 120 indicative of a number of interactions 119 between the user 10 and the entity 12 and may include textual feedback 121 as well as non-textual metadata 122.” Examiner notes that a language model (natural language processing model) is trained using training data (training data 251 which includes historical sentiment data) separate from the plurality of messages (received interaction data 120))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, Sewak, and Zhong. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. Raj teaches KNN algorithm. Sewak teaches a method for automatic labeling of text data. Zhong teaches a method for predicting a customer trust target metric. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, Sewak, and Zhong to train the language processing model to make accurate predictions from large sets of data “The model 270 is capable of analyzing large sets of user interaction data 120 characterizing sentiment data 250 from numerous users 10 in order to accurately predict a customer trust target metric for each user 10. In order to achieve the intended functionality, the language processing model 270 is trained to analyze large data sets and recognize and group similar interactions 119.” (Zhong Paragraph 0026).
Cao in view of Polichroniadis in further view of Bradley in further view of Ponomarev in further view of Raj in further view of Sewak in further view of Zhong does not teach including a first set of messages having labels and a second set of messages, greater in quantity than the first set of messages, lacking labels
However, Ahmet does teach including a first set of messages having labels and a second set of messages, greater in quantity than the first set of messages, lacking labels (Ahmet Page 5075 Paragraph 3; "The training set consists of 50k images coming from 10 classes, while the test set consists of 10k images from the same 10 classes. All images have resolution 32 × 32. Evaluation is performed with 50, 100, 200, and 400 labeled images per classes, corresponding to l = 500, 1k, 2k, and 4k label images in total." Examiner notes that first subset (ground truth labeled images) contains 500 labeled images and second subset (unlabeled images) contains 50000 - 500 = 49500 unlabeled images; second subset makes up 90 percent of the dataset 49500/50000 = 0.99; Examiner interprets images as messages because people can message each other with pictures)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, Sewak, Zhong, and Ahmet. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. Raj teaches KNN algorithm. Sewak teaches a method for automatic labeling of text data. Zhong teaches a method for predicting a customer trust target metric. Ahmet teaches a method for label propagation for deep semi-supervised learning. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, Sewak, Zhong, and Ahmet to perform the method on more sparse labeled data for a larger benefit “The proposed approach performs the best out of the pseudo-label based approaches on CIFAR-10. Results in Figure 6 show that our benefit is larger when the number of labels is reduced” (Ahmet Page 5077 Paragraph 4).
Cao in view of Polichroniadis in further view of Bradley in further view of Ponomarev in further view of Raj in further view of Sewak in further view of Zhong in further view of Ahmet does not teach the first number of dimensions being less than or equal to a quantity of unique tokens included within the plurality of messages
However, Dane does teach the first number of dimensions being less than or equal to a quantity of unique tokens included within the plurality of messages (Dane Fig 4 and Paragraph 0111; "The word piece tokens 403 and the positional embeddings 407 may simply be summed to form the input representation of the input text sequence 401… The classification layer 409 is trained to output a probability 410 for each of the possible unique biological target identifiers 411" Examiner notes that plurality of messages (input text sequence 401) has a number of unique tokens (word piece tokens), wherein the semantic graph has original dimensions (possible unique biological target identifier), and wherein the original dimensions have a number less than or equal to the number of unique tokens (10 tokens > 3 identifiers))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, Sewak, Zhong, Ahmet, and Dane. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. Raj teaches KNN algorithm. Sewak teaches a method for automatic labeling of text data. Zhong teaches a method for predicting a customer trust target metric. Ahmet teaches a method for label propagation for deep semi-supervised learning. Dane teaches using knowledge graphs to predict new biological targets. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, Sewak, Zhong, Ahmet, and Dane to leverage the advantages of a knowledge graph to enhance the ability of the model “This method therefore benefits from the advantages associated with knowledge graph inference and language models to further enhance the ability of the model to determine biological entities of interest for a given user-specified biological context.” (Dane Paragraph 0026).
Claim(s) 2-4, 10, 13-16, 18, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Kun Cao et al; US 12566973 B1 filed on Sep 27, 2021 (hereinafter “Cao”) in view of Alexander S Polichroniadis et al; US 12112519 B1 filed on Feb 23, 2022 (hereinafter “Polichroniadis”) in further view of bradleyboehmke; “Chapter 17 Principal Component Analysis” available online on Dec 23, 2022 (hereinafter “Bradley”) in further view of Stanislav Olegovich Ponomarev; US 20210042471 A1 filed on Jul 27, 2020 (hereinafter “Ponomarev”) in further view of Ravish Raj; “K-Nearest Neighbors (KNN) Algorithm in Machine Learning” available online Dec 27, 2022 (hereinafter “Raj”)
Regarding claim 2, Cao teaches A method comprising: inputting a plurality of messages into a language model to generate embeddings in a first embedding space having a first number of dimensions (Cao Fig 5 and Column 7 Line 8; “this process includes at least some amount of textual content. At least an alphanumeric or textual portion of the content can be processed 504 using a language model to infer one or more semantic embeddings for the content, where a semantic embedding can include a numerical vector including values corresponding to various semantic features inferred for the input content.” Examiner notes that a plurality of messages (input content/some amount of textual content) is input into a language model to generate embeddings in a first embedding space having a first number of dimensions (language model generates embeddings corresponding to various semantic features/first embedding space having a first number of dimensions))
and a semantic graph comprising datapoints representing the plurality of messages connected based on a semantic similarity; (Cao Fig 5 and Column 7 Line 19; “provide semantic input to a graph model being co-trained with the language model. This can involve inserting 508 the semantic embedding(s) into a relevant knowledge graph as part of the upper layer of this unified framework. The knowledge graph, including the semantic information, can then be provided 510 as input to a graphing model.” Cao Column 8 Line 1; “One or more edge scores can be inferred 610, using this graphing model, between a content node, corresponding to the input and semantic embedding, and other nodes of the graph. One or more edge labels can then be determined 612 for these pairs of nodes, along with corresponding confidence values in at least one embodiment.” Examiner notes that a semantic graph (graph created from graphing model) is generated/inferred comprising datapoints representing a plurality of messages connected (edges inferred between content node) based on a semantic similarity (using semantic embedding))
Cao does not teach inputting the semantic graph and a labeling task into an artificial intelligence model to identify a second number of dimensions that is less than the first number of dimensions, [wherein each dimension of the second number of dimensions represents a portion of a total amount of information present in the semantic graph exceeding a threshold amount of information;]
However, Polichroniadis does teach inputting the semantic graph and a labeling task into an artificial intelligence model to identify a second number of dimensions that is less than the first number of dimensions, [wherein each dimension of the second number of dimensions represents a portion of a total amount of information present in the semantic graph exceeding a threshold amount of information;] (Polichroniadis Column 6 Line 39; “One or more machine learning methods may be used to classify objects, include, or add semantic information to a 3D reconstruction. In some implementations, a machine learning method is used to generate semantic labels for pixels of one or more images of the physical environment.” Polichroniadis Column 7 Line 4; “a machine learning method is used to estimate relationships between objects of the semantic mesh or graph representing the semantic mesh. A machine learning method may include machine learning algorithms (such as PCA, clustering algorithms, etc.) and/or machine learning models (such as neural network-based approaches).” Examiner notes that the semantic graph (graph representing the semantic mesh) is input into an artificial intelligence model (machine learning models) to identify a second number of dimensions that is less than the first number of dimensions (PCA is a dimensionality reduction technique); a labeling task (generate semantic labels) is input into the AI model (machine learning method))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao and Polichroniadis. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. One of ordinary skill would have motivation to combine Cao and Polichroniadis to improve the systems accuracy or efficiency “the machine learning method to further improve its accuracy or efficiency” (Polichroniadis Column 7 Line 30).
Cao in view of Polichroniadis does not teach [inputting the semantic graph and a labeling task into an artificial intelligence model to identify a second number of dimensions that is less than the first number of dimensions,] wherein each dimension of the second number of dimensions represents a portion of a total amount of information present in the semantic graph exceeding a threshold amount of information;
However, Bradley does teach [inputting the semantic graph and a labeling task into an artificial intelligence model to identify a second number of dimensions that is less than the first number of dimensions,] wherein each dimension of the second number of dimensions represents a portion of a total amount of information present in the semantic graph exceeding a threshold amount of information; (Bradley Section 17.5.2 Paragraph 2; “The first PCt in our example explains 5.46% of the feature variability, and the second principal component explains 5.17%. Together, the first two PCs explain 10.63% of the variability. Thus, if an analyst desires to choose the number of PCs required to explain at least 75% of the variability in our original data then they would choose the first 27 components.” Examiner notes that each dimension of the second number of dimensions (first 27 PCs) represents a portions of a total amount of information present in the semantic graph exceeding a threshold amount of information (number of PCs required to explain at least 75% of the variability in our original data))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, and Bradley. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. One of ordinary skill would have motivation to combine Cao, Polichroniadis, and Bradley to find low-dimensional representations of a data set for more efficient data processing “Principal components analysis (PCA) is a method for finding low-dimensional representations of a data set that retain as much of the original variation as possible.” (Bradley Paragraph 1).
Cao in view of Polichroniadis in further view of Bradley does not teach projecting the semantic graph into a second embedding space having the second number of dimensions to generate a projected graph;
However, Ponomarev does teach projecting the semantic graph into a second embedding space having the second number of dimensions to generate a projected graph; (Ponomarev Paragraph 0032; "At operation 304, the semantic graph embedding system 116 constructs a Markov chain. The Markov chain comprises a plurality of nodes. Each node in the plurality of nodes represents a data value in the dataset." Malden Paragraph 0035; "At operation 310, the distance (e.g., path cost) determined in operation 308 is stored as a dimension in a vector… then the vector representation of the target node would be: [A,B,C,D,E]… the semantic graph embedding system 116 reduces the vector (e.g., an n-dimensional vector) to a three-dimensional vector using principal component analysis (PCA)." Examiner notes that the projected graph (three-dimensional vector) is determined by projection (reduction by semantic embedding) the semantic graph into the optimal set of dimension (vector))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, and Ponomarev. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, and Ponomarev to extract the meaning of data in different contexts for more accurate data processing “The semantic graph embedding system may extract the meaning of data in the dataset when the data is used in one particular context versus another context.” (Ponomarev Paragraph 0012).
Cao in view of Polichroniadis in further view of Bradley in further view of Ponomarev does not teach measuring (a) a first distance between (i) a first projection from the projected graph corresponding to a first datapoint in the semantic graph representing a first message in the plurality of messages lacking a label and (ii) a second projection in the projected graph corresponding to a second datapoint in the semantic graph representing a second message in the plurality of messages corresponding to a first preassigned label,
and (b) a second distance between the first projection and a third projection from the projected graph corresponding to a third datapoint in the semantic graph representing a third message in the plurality of messages having a second preassigned label;
based on the first likelihood being greater than the second likelihood, storing, in electronic storage, the first preassigned label in association with the first message represented by the first projection.
However, Raj does teach measuring (a) a first distance between (i) a first projection from the projected graph corresponding to a first datapoint in the semantic graph representing a first message in the plurality of messages lacking a label and (ii) a second projection in the projected graph corresponding to a second datapoint in the semantic graph representing a second message in the plurality of messages corresponding to a first preassigned label, (Raj Section Working of KNN; “We need to find the K neighbors based on any distance metric… We will have the test sample on which we want the prediction... Among the selected K neighbors, we need to count how many neighbors are from different classes” Examiner notes that a first distance is measured between a first projection from the projected graph corresponding to a first data point represent a first message lacking a label (test sample to be predicted) and a second projection corresponding to a second data point representing a second message corresponding to a first preassigned label (neighbor associated with a class))
and (b) a second distance between the first projection and a third projection from the projected graph corresponding to a third datapoint in the semantic graph representing a third message in the plurality of messages having a second preassigned label; (Examiner refers to previous mapping to show a second distance is measured between the first projection (test sample) and a third projection corresponding to a third datapoint representing a third message having a second preassigned label (neighbor associated with different class))
computing a first likelihood that the first projection is to be assigned the first preassigned label based on the first distance and a second likelihood that the first projection is to be assigned the second preassigned label based on the second distance (Raj Section Working of KNN; “Among the selected K neighbors, we need to count how many neighbors are from the different classes… Now, we have to assign the test data sample to the class for which the count of neighbors was maximum” Examiner notes that a first likelihood that the first projection is to be assigned the first preassigned label is computed based on the first distance (using the distance metrics, closest neighbors are selected and counted, then out of those neighbors, how many are associated to class 1 are counted; EX 6/10 are class 1) and a second likelihood that the first projection is to be assigned the second preassigned label based on the second distance (using the distance metrics, closest neighbors are selected and counted, then out of those neighbors, how many are associated to class 2 are counted; EX 3/10 are class 2))
based on the first likelihood being greater than the second likelihood, storing, in electronic storage, the first preassigned label in association with the first message represented by the first projection. (Examiner refers to previous mapping to show that based on the first likelihood being greater than the second likelihood (count of neighbors for class 1 is greater than count of neighbors for class 2), storing the first preassigned label in the association with the first message represented by the first projection (test data sample is assigned to class for which the count of neighbors was maximum))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, Ponomarev, and Raj. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. Raj teaches KNN algorithm. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, Ponomarev, and Raj to leverage KNN strengths for label classification “KNN is a very famous algorithm because of its simplicity, so let’s understand the key strengths… zero training time…sample efficiency…explainable…easy to add and remove data… less sensitive to class imbalance” (Raj Section “Strengths of the KNN algorithm”).
Regarding claim 3, Cao does not teach The method of claim 2, further comprising:
determining corresponding variances for each dimension of the first number of dimensions, wherein a variance represents a measure of shared information between a respective dimension and the first number of dimensions; and
comparing the corresponding variances to a variance threshold related to the threshold amount of information to identify the second number of dimensions.
However, Bradely does teach The method of claim 2, further comprising:
determining corresponding variances for each dimension of the first number of dimensions, wherein a variance represents a measure of shared information between a respective dimension and the first number of dimensions; and (Bradley Fig 17.6 and Section 17.5.2 Paragraph 1; “The proportion of variance explained (PVE) identifies the optimal number of PCs to keep based on the total variability that we would like to account for. Mathematically, the PVE for the m-th PC is calculated as” Examiner notes that determining corresponding variances for each dimension (calculate PVE for the m-th PC) of the first number of dimensions (all 42 PCs), wherein a variance represents a measure of shared information between respective dimension and a first number of dimensions (Fig 17.6 shows variance as measured of shared information between respective PC of all 42 PCs in PVE graph))
PNG
media_image1.png
503
783
media_image1.png
Greyscale
comparing the corresponding variances to a variance threshold related to the threshold amount of information to identify the second number of dimensions. (Bradley Section 17.5.2 Paragraph 2; “The first PCt in our example explains 5.46% of the feature variability, and the second principal component explains 5.17%. Together, the first two PCs explain 10.63% of the variability. Thus, if an analyst desires to choose the number of PCs required to explain at least 75% of the variability in our original data then they would choose the first 27 components.” Examiner notes that the corresponding variances (feature variability of PCs) is compared to a variance threshold related to the threshold amount of information (choose the number of PCs required to explain at least 75% of the variability in our original data) to identify the second number of dimensions (choose the first 27 components.))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, and Bradley. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. One of ordinary skill would have motivation to combine Cao, Polichroniadis, and Bradley to find low-dimensional representations of a data set for more efficient data processing “Principal components analysis (PCA) is a method for finding low-dimensional representations of a data set that retain as much of the original variation as possible.” (Bradley Paragraph 1).
Regarding claim 4, Cao does not teach The method of claim 2, further comprising: in response to the first likelihood being greater than the second likelihood, generating for display, on a user interface,
However, Polichroniadis does teach The method of claim 2, further comprising: in response to the first likelihood being greater than the second likelihood, generating for display, on a user interface, (Polichroniadis Column 18 Line 39; “generate narrative display data 832 to present content within the 3D environment. In some implementations, the set of views is displayed on the device display 850 of a device (e.g., device 110 of FIG. 1).” Examiner notes that in response to the first likelihood being greater than the second likelihood, generating for display, on a user interface (display data is generated to be displayed on device display 850))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao and Polichroniadis. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. One of ordinary skill would have motivation to combine Cao and Polichroniadis to improve the systems accuracy or efficiency “the machine learning method to further improve its accuracy or efficiency” (Polichroniadis Column 7 Line 30).
Cao in view of Polichroniadis does not teach a recommendation to assign the first preassigned label in association with the first message represented by the first projection; and
based on a review received by the user interface, assigning the first preassigned label in association with the first message represented by the first projection.
However, Raj does teach a recommendation to assign the first preassigned label in association with the first message represented by the first projection; and (Raj Section Working of KNN; “Among the selected K neighbors, we need to count how many neighbors are from the different classes… Now, we have to assign the test data sample to the class for which the count of neighbors was maximum” Examiner notes that class for which the count of neighbors was maximum is a recommendation to assign the first preassigned label (class) in associated with the first message represented by the first projection (test sample))
based on a review received by the user interface, assigning the first preassigned label in association with the first message represented by the first projection. (Examiner refers to previous mapping to show that first preassigned label (class) is assigned in associated with the first message represented by the first projections (test sample))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, Ponomarev, and Raj. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. Raj teaches KNN algorithm. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, Ponomarev, and Raj to leverage KNN strengths for label classification “KNN is a very famous algorithm because of its simplicity, so let’s understand the key strengths… zero training time…sample efficiency…explainable…easy to add and remove data… less sensitive to class imbalance” (Raj Section “Strengths of the KNN algorithm”).
Regarding claim 10, Cao does not teach the method of claim 2, further comprising based on the first likelihood being greater than the second likelihood, assigning the first preassigned label to the first projection;
assigning the first preassigned label to the first datapoint;
and assigning the first label to the first message.
However, Raj does teach the method of claim 2, further comprising based on the first likelihood being greater than the second likelihood, assigning the first preassigned label to the first projection;
assigning the first preassigned label to the first datapoint;
and assigning the first label to the first message. (Raj Section “What are the common assumptions in KNN”; “Every sample in the training data is mapped to a real n-dimensional space, where each sample has the same number of attributes or dimensions” Raj Section Working of KNN; “Among the selected K neighbors, we need to count how many neighbors are from the different classes… Now, we have to assign the test data sample to the class for which the count of neighbors was maximum” Examiner notes that assigning the first preassigned label (class) corresponds to a first projection (test data sample mapped to n-dimensional space); first datapoint (test sample data) is associated to first projection so assigning to first data point is assigning to first projection as well; test sample data is interpreted as data pertaining to first message, so by assigning label to first projection, then so is the first message as well)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, Ponomarev, and Raj. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. Raj teaches KNN algorithm. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, Ponomarev, and Raj to leverage KNN strengths for label classification “KNN is a very famous algorithm because of its simplicity, so let’s understand the key strengths… zero training time…sample efficiency…explainable…easy to add and remove data… less sensitive to class imbalance” (Raj Section “Strengths of the KNN algorithm”).
Regarding claim 13, Cao does not teach The method of claim 12, further comprising: determining the total amount of information present in the semantic graph, wherein the total amount of information is unevenly distributed among the original dimensions;
determining a first dimension from the original dimensions, wherein the first dimension has a first amount of information, comprising a portion of the total amount of information present in the semantic graph;
comparing the first amount of information to the threshold amount of information
and in response to comparing the first amount of information to the threshold amount of information, including the first dimension in the second number of dimensions.
However, Bradely does teach The method of claim 12, further comprising: determining the total amount of information present in the semantic graph, wherein the total amount of information is unevenly distributed among the original dimensions; (Bradely Figure 17.6 and Section 17.5.2 Paragraph 1; “The proportion of variance explained (PVE) identifies the optimal number of PCs to keep based on the total variability that we would like to account for. Mathematically, the PVE for the m-th PC is calculated as:” Examiner notes that calculating PVE for the m-th PVE is determining the total amount of information present in the semantic graph, wherein the total amount of information is unevenly distributed among the original dimensions (Fig 17.6 in PVE graph shows that each PC contributes a variance of information))
determining a first dimension from the original dimensions, wherein the first dimension has a first amount of information, comprising a portion of the total amount of information present in the semantic graph; (Bradely Section 17.5.2 Paragraph 2; “The first PCt in our example explains 5.46% of the feature variability, and the second principal component explains 5.17%. Together, the first two PCs explain 10.63% of the variability. Thus, if an analyst desires to choose the number of PCs required to explain at least 75% of the variability in our original data then they would choose the first 27 components.” Examiner notes that a first dimension from the original dimensions (first PCt) is determined, wherein the first dimension has a first amount of information present in the semantic graph (explains 5.46% of the feature variability))
comparing the first amount of information to the threshold amount of information
and in response to comparing the first amount of information to the threshold amount of information, including the first dimension in the second number of dimensions. (Examiner refers to previous mapping to show comparing the first amount of information (5.46% of the feature variability) to the threshold amount of information (explain at least 75% of the variability in our original data) and in response to comparing, including the first dimension in the second number of dimensions (first PC is included in the 27 PC needed to reach 75% variability))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, and Bradley. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. One of ordinary skill would have motivation to combine Cao, Polichroniadis, and Bradley to find low-dimensional representations of a data set for more efficient data processing “Principal components analysis (PCA) is a method for finding low-dimensional representations of a data set that retain as much of the original variation as possible.” (Bradley Paragraph 1).
Regarding claim 14, Cao does not teach The method of claim 13, wherein determining the first optimal dimension from the original dimensions comprises: determining an original vector for each original dimension of the original dimensions;
determining a correlation between the original vector and all other original vectors, wherein the correlation comprises a measure of shared information;
determining a first optimal vector based on the correlation;
And determining the first optimal dimension from the first optimal vector.
However, Bradley does teach The method of claim 13, wherein determining the first optimal dimension from the original dimensions comprises: determining an original vector for each original dimension of the original dimensions; (Bradley Section 17.3 Paragraph 1; “The first principal component of a set of features X1, X2, …, Xp is the linear combination of the features Z1… (17.1) that has the largest variance. Here ϕ1… is the loading vector for the first principal component… After the first principal component Z1 has been determined, we can find the second principal component Z2… This process proceeds until all p principal components are computed.” Examiner notes that an original vector (loading vector) is determined/computed for each original dimension of the original dimensions (all principal components PCs))
determining a correlation between the original vector and all other original vectors, wherein the correlation comprises a measure of shared information; (Bradley Section 17.3 Paragraph 1; “After the first principal component Z1 has been determined, we can find the second principal component Z2. The second principal component is the linear combination of X1,…,Xp that has maximal variance out of all linear combinations that are uncorrelated with Z1:” Examiner notes that a correlation between the original vectors and all other original vectors are determined (each PC has its variance computed without having the previously computed PC correlated with it))
determining a first optimal vector based on the correlation; (Bradley Section 17.3 Paragraph 1; “The first principal component of a set of features X1, X2, …, Xp is the linear combination of the features Z1=ϕ11X1+ϕ21X2+...+ϕp1Xp,(17.1) that has the largest variance. Here ϕ1=(ϕ11,ϕ21,…,ϕp1) is the loading vector for the first principal component.” Examiner notes that a first optimal vector is determined (loading vector) based on the correlation (linear combination of features))
And determining the first optimal dimension from the first optimal vector. (Examiner refers to previous mapping to show that the first optimal dimension (first principal component) is determined from the first optimal vector (loading vector))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, and Bradley. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. One of ordinary skill would have motivation to combine Cao, Polichroniadis, and Bradley to find low-dimensional representations of a data set for more efficient data processing “Principal components analysis (PCA) is a method for finding low-dimensional representations of a data set that retain as much of the original variation as possible.” (Bradley Paragraph 1).
Regarding claim 15, Cao does not teach The method of claim 13, further comprising: determining a second dimension from the original dimensions, wherein the second dimension has a second amount of information, and wherein the second amount of information is less than the first amount of information;
combining the first amount of information and the second amount of information into a running total amount of information;
comparing the running total amount of information to the threshold amount of information; and
and in response to comparing the running total amount of information to the threshold amount of information, discarding the second dimension.
However, Bradley does teach The method of claim 13, further comprising: determining a second dimension from the original dimensions, wherein the second dimension has a second amount of information, and wherein the second amount of information is less than the first amount of information; (Bradely Figure 17.6 and Section 17.5.2 Paragraph 1; “The proportion of variance explained (PVE) identifies the optimal number of PCs to keep based on the total variability that we would like to account for. Mathematically, the PVE for the m-th PC is calculated as:” Examiner notes that a second dimension from the original dimension is determined/calculated from the original dimensions, wherein the second dimension has a second amount of information, and wherein the second amount of information is less than the first amount of information (Figure 17.6 in PVE graph shows variance of information for the other 41 PC, each having less variance than first PC))
combining the first amount of information and the second amount of information into a running total amount of information; (Bradley Figure 17.6 and Section 17.5.2 Paragraph 1; “Mathematically, the PVE for the m-th PC is calculated … the cumulative variance explained (CVE)” Examiner notes that the first amount of information and the second amount of information is combine into a running total amount of information (Figure 17.6 in CVE graph shows how variance explained is increased with each PC added on))
comparing the running total amount of information to the threshold amount of information; and (Bradely Section 17.5.2 Paragraph 2; “The first PCt in our example explains 5.46% of the feature variability, and the second principal component explains 5.17%. Together, the first two PCs explain 10.63% of the variability. Thus, if an analyst desires to choose the number of PCs required to explain at least 75% of the variability in our original data then they would choose the first 27 components.” Examiner notes that the running total amount of information (CVE at PC included so far) is compared to threshold amount of information (75% of the variability in our original data))
and in response to comparing the running total amount of information to the threshold amount of information, discarding the second dimension. (Examiner refers to previous mapping to show that PC 28-42 is discarded/not considered because the first 27 PC explain at least 75% of the variability in the original data)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, and Bradley. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. One of ordinary skill would have motivation to combine Cao, Polichroniadis, and Bradley to find low-dimensional representations of a data set for more efficient data processing “Principal components analysis (PCA) is a method for finding low-dimensional representations of a data set that retain as much of the original variation as possible.” (Bradley Paragraph 1).
Regarding claim 16, Cao teaches The method of claim 13, further comprising: receiving a first user input; and (Cao Column 4 Line 45; “At inference time, received user input 102 can be provided to both the inner and outer layers.”)
Cao does not teach determining the threshold amount of information based on the first user input.
However, Bradley does teach determining the threshold amount of information based on the first user input. (Bradley Section 17.5.2 Paragraph 2; “Thus, if an analyst desires to choose the number of PCs required to explain at least 75% of the variability in our original data then they would choose the first 27 components… What amount of variability is reasonable? This varies by application and the data being used.” Examiner notes that the threshold amount of information (amount of variability) is determined based on the first user input (based on analyst desires, application, and data being used))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, and Bradley. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. One of ordinary skill would have motivation to combine Cao, Polichroniadis, and Bradley to find low-dimensional representations of a data set for more efficient data processing “Principal components analysis (PCA) is a method for finding low-dimensional representations of a data set that retain as much of the original variation as possible.” (Bradley Paragraph 1).
Regarding claim 18, Cao teaches A non-transitory, computer readable medium storing computer instructions which, when executed by one or more computer processors, cause the one or more computer processors to: (Cao Column 15 Line 13; “The BIOS component can include non-transitory executable code, often referred to as firmware, which can be executed by one or more processors”)
Claim 18 is a non-transitory, computer readable medium of method claim 2 and is accordingly rejected using substantially similar rationale as to that which is set for with respect to claim 2.
Regarding claim 20, claim 20 has similar limitations as of claim 13, except it is a non-transitory, computer readable media claim, therefore it is rejected under the same rationale as claim 13.
Claim(s) 5 is rejected under 35 U.S.C. 103 as being unpatentable over Kun Cao et al; US 12566973 B1 filed on Sep 27, 2021 (hereinafter “Cao”) in view of Alexander S Polichroniadis et al; US 12112519 B1 filed on Feb 23, 2022 (hereinafter “Polichroniadis”) in further view of bradleyboehmke; “Chapter 17 Principal Component Analysis” available online on Dec 23, 2022 (hereinafter “Bradley”) in further view of Stanislav Olegovich Ponomarev; US 20210042471 A1 filed on Jul 27, 2020 (hereinafter “Ponomarev”) in further view of Ravish Raj; “K-Nearest Neighbors (KNN) Algorithm in Machine Learning” available online Dec 27, 2022 (hereinafter “Raj”) in further view of Bassant Gamal; “Naïve Bayes Algorithm” available online on Aug 23, 2022 (hereinafter “Gamal”)
Regarding claim 5, Cao does not teach The method of claim 2, wherein computing the second likelihood that the first projection is to be assigned the second label comprises: determining a third likelihood that the second projection is to be assigned the second preassigned label and
determining the second likelihood based on the third likelihood.
However, Raj does teach The method of claim 2, wherein computing the second likelihood that the first projection is to be assigned the second label comprises: determining a third likelihood that the second projection is to be assigned the second preassigned label and (Raj Section Working of KNN; “Among the selected K neighbors, we need to count how many neighbors are from the different classes… Now, we have to assign the test data sample to the class for which the count of neighbors was maximum” Examiner notes that a third likelihood that the second projection is to be assigned the second preassigned label is computed(using the distance metrics, closest neighbors are selected and counted, then out of those neighbors, how many are associated to class 2 are counted; EX 2/10 are class 2))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, Ponomarev, and Raj. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. Raj teaches KNN algorithm. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, Ponomarev, and Raj to leverage KNN strengths for label classification “KNN is a very famous algorithm because of its simplicity, so let’s understand the key strengths… zero training time…sample efficiency…explainable…easy to add and remove data… less sensitive to class imbalance” (Raj Section “Strengths of the KNN algorithm”).
Cao in view of Raj does not teach determining the second likelihood based on the third likelihood.
However, Gamal does teach determining the second likelihood based on the third likelihood. (Gamal Figure attached shows determining the second likelihood (P(H|E)) based on the third likelihood (P(H)))
PNG
media_image2.png
432
661
media_image2.png
Greyscale
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, and Gamal. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. Raj teaches KNN algorithm. Gamal teaches Naïve Bayes Algorithm. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, and Gamal to leverage the pros of Naïve Bayes Algorithm for label propagation “Requires a small amount of training data. So the training takes less time. Handles continuous and discrete data, and it is not sensitive to irrelevant features. Very simple, fast, and easy to implement. Can be used for both binary and multi-class classification problems. Highly scalable as it scales linearly with the number of predictor features and data points. When the Naive Bayes conditional independence assumption holds true, it will converge quicker than discriminative models like logistic regression.” (Raj Section “Strengths of the KNN algorithm”).
Claim(s) 6 is rejected under 35 U.S.C. 103 as being unpatentable over Kun Cao et al; US 12566973 B1 filed on Sep 27, 2021 (hereinafter “Cao”) in view of Alexander S Polichroniadis et al; US 12112519 B1 filed on Feb 23, 2022 (hereinafter “Polichroniadis”) in further view of bradleyboehmke; “Chapter 17 Principal Component Analysis” available online on Dec 23, 2022 (hereinafter “Bradley”) in further view of Stanislav Olegovich Ponomarev; US 20210042471 A1 filed on Jul 27, 2020 (hereinafter “Ponomarev”) in further view of Ravish Raj; “K-Nearest Neighbors (KNN) Algorithm in Machine Learning” available online Dec 27, 2022 (hereinafter “Raj”) in further view of Zhao et al; “Cyclic label propagation for graph semi-supervised learning” published on Jun 24, 2021 (hereinafter “Zhao”).
Regarding claim 6, Cao does not teach and in response to comparing the first entropy score to the first threshold entropy score, generating for display, on a user interface,
However, Polichroniadis does teach and in response to comparing the first entropy score to the first threshold entropy score, generating for display, on a user interface, (Polichroniadis Column 18 Line 39; “generate narrative display data 832 to present content within the 3D environment. In some implementations, the set of views is displayed on the device display 850 of a device (e.g., device 110 of FIG. 1).” Examiner notes that in response to the first likelihood being greater than the second likelihood, generating for display, on a user interface (display data is generated to be displayed on device display 850))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao and Polichroniadis. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. One of ordinary skill would have motivation to combine Cao and Polichroniadis to improve the systems accuracy or efficiency “the machine learning method to further improve its accuracy or efficiency” (Polichroniadis Column 7 Line 30).
Cao in view of Polichroniadis does not teach The method of claim 2, further comprising: determining a composite likelihood from the first likelihood and the second likelihood;
determining a first entropy score from the composite likelihood, wherein the first entropy score indicates randomness of label components in the composite likelihood;
comparing the first entropy score to a first threshold entropy score;
[and in response to comparing the first entropy score to the first threshold entropy score, generating for display, on the user interface,] a recommendation to review the first preassigned label being assigned to the first projection.
However, Zhao does teach The method of claim 2, further comprising: determining a composite likelihood from the first likelihood and the second likelihood; (Zhao Page 710 Paragraph 3 and Equation 6; "where fik denotes the probability of node vi belonging to class k." Examiner notes that determining a composite likelihood from the first likelihood and the second likelihood is performing a summation of fik probabilities)
PNG
media_image3.png
58
322
media_image3.png
Greyscale
determining a first entropy score from the composite likelihood, wherein the first entropy score indicates randomness of label components in the composite likelihood; (Zhao Page 710 Paragraph 3; "The regularizer is composed of a Shannon entropy function H (·)" Examiner notes that a first entropy score (output of Shannon entropy function) is determined from the composite likelihood (summation of probabilities as referenced in previous mapping), wherein the first entropy score indicates randomness of label components in the composite likelihood (Shannon entropy function is a measure of randomness))
comparing the first entropy score to a first threshold entropy score; (Zhao Page 710 Paragraph 3; "If the Shannon entropy of fi is smaller than the threshold, we set ϕi as 1 to indicate that node vi can be utilized as a label context." Examiner notes that comparing the first entropy score (Shannon entropy fi) to a first threshold entropy score (threshold))
[and in response to comparing the first entropy score to the first threshold entropy score, generating for display, on a user interface,] a recommendation to review the first preassigned label being assigned to the first projection. (Zhao Page 710 Paragraph 3; "The binary value of ϕi indicates whether node vi’s learned label is reliable or not and λ acts as a threshold to distinguish the informative labels from the uninformative labels." Examiner notes that binary value determined from Shannon entropy is a third recommendation to review the first label being assigned to the first projection (because it is not reliable))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, and Zhao. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. Raj teaches KNN algorithm. Zhao teaches using Shannon entropy function in label learning task. One of ordinary skill would have motivation to Cao, Polichroniadis, Bradley, Ponomarev, Raj, and Zhao to leverage Shannon entropy to select some highly reliable node labels in each training iteration “If the Shannon entropy of fi is smaller than the threshold, we set ϕi as 1 to indicate that node vi can be utilized as a label context. As the training process goes on, λ is gradually increased such that more learned highly reliable labels can be included in graph embedding procedure to adaptively update node embeddings.” (Zhao Page 710 Paragraph 3).
Claim(s) 7 is rejected under 35 U.S.C. 103 as being unpatentable over Kun Cao et al; US 12566973 B1 filed on Sep 27, 2021 (hereinafter “Cao”) in view of Alexander S Polichroniadis et al; US 12112519 B1 filed on Feb 23, 2022 (hereinafter “Polichroniadis”) in further view of bradleyboehmke; “Chapter 17 Principal Component Analysis” available online on Dec 23, 2022 (hereinafter “Bradley”) in further view of Stanislav Olegovich Ponomarev; US 20210042471 A1 filed on Jul 27, 2020 (hereinafter “Ponomarev”) in further view of Ravish Raj; “K-Nearest Neighbors (KNN) Algorithm in Machine Learning” available online Dec 27, 2022 (hereinafter “Raj”) in further view of Jason; “Semi-Supervised Learning With Label Propagation” available on Dec 06, 2022 (hereinafter “Jason”).
Regarding claim 7, Cao does not teach and generating for display, on a user interface
However, Polichroniadis does teach and generating for display, on a user interface (Polichroniadis Column 18 Line 39; “generate narrative display data 832 to present content within the 3D environment. In some implementations, the set of views is displayed on the device display 850 of a device (e.g., device 110 of FIG. 1).” Examiner notes that in response to the first likelihood being greater than the second likelihood, generating for display, on a user interface (display data is generated to be displayed on device display 850))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao and Polichroniadis. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. One of ordinary skill would have motivation to combine Cao and Polichroniadis to improve the systems accuracy or efficiency “the machine learning method to further improve its accuracy or efficiency” (Polichroniadis Column 7 Line 30).
Cao in view of Polichroniadis does not teach The method of claim 2, further comprising: determining a first predicted label for the first projection;
in response to comparing the first consistency score to the first threshold consistency score, filtering the first projection to a first group;
However, Raj does teach The method of claim 2, further comprising: determining a first predicted label for the first projection; (Raj Section Working of KNN; “Among the selected K neighbors, we need to count how many neighbors are from the different classes… Now, we have to assign the test data sample to the class for which the count of neighbors was maximum” Examiner notes that a first predicted label (class) is determined for the first projection (test data sample))
in response to comparing the first consistency score to the first threshold consistency score, filtering the first projection to a first group; (Examiner refers to previous mapping to show that a first predicted label is determined for the test sample data; assigning label to test sample data is grouping/filtering the data with the group/label)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, Ponomarev, and Raj. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. Raj teaches KNN algorithm. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, Ponomarev, and Raj to leverage KNN strengths for label classification “KNN is a very famous algorithm because of its simplicity, so let’s understand the key strengths… zero training time…sample efficiency…explainable…easy to add and remove data… less sensitive to class imbalance” (Raj Section “Strengths of the KNN algorithm”).
Cao in view of Polichroniadis in further view of Raj does not teach determining a first consistency score for the first projection based on a comparison of the first predicted label and the first preassigned label, wherein the first consistency score indicates a degree of consensus between the first predicted label and the first preassigned label;
comparing the first consistency score to a first threshold consistency score;
[and generating for display, on the user interface,] a fourth recommendation to use the first group as a training sample for a supervised learning task.
However, Jason does teach determining a first consistency score for the first projection based on a comparison of the first predicted label and the first preassigned label, wherein the first consistency score indicates a degree of consensus between the first predicted label and the first preassigned label; (Jason Section "Label Propagation for Semi-Supervised Learning" Paragraph 13; "model = LabelPropagation() # fit model on training dataset model.fit(X_train_mixed, y_train_mixed) # make predictions on hold out test set yhat = model.predict(X_test) # calculate score for test set score = accuracy_score(y_test, yhat)" Examiner notes that a first consistency score (accuracy score) is determined for the first projection based on comparison of the first predicted label (yhat) and the first label (y_test), wherein the first consistency score indicates a degree of consensus between the first predicted label and the first label (as shown from accuracy score))
comparing the first consistency score to a first threshold consistency score; (Jason Section "Label Propagation for Semi-Supervised Learning" Paragraph 16; "we can see that the label propagation model achieves a classification accuracy of about 85.6 percent, which is slightly higher than a logistic regression fit only on the labeled training dataset that achieved an accuracy of about 84.8 percent." Examiner notes that the first consistency score (85.6 percent) is compared to a first threshold consistency score (84.8 percent))
[and generating for display, on a user interface,] a recommendation to use the first group as a training sample for a supervised learning task. (Jason Section "Label Propagation for Semi-Supervised Learning" Paragraph 16; "So far, so good. Another approach we can use with the semi-supervised model is to take the estimated labels for the training dataset and fit a supervised learning model." Examiner notes that using the first group as a training sample for a supervised learning task (take the estimated labels for the training dataset and fit a supervised learning model.))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, and Jason. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. Raj teaches KNN algorithm. Jason teaches semi-supervised learning with label propagation. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, and Jason to use the approach to improve the classification accuracy of the model “In this case, we can see that this hierarchical approach of the semi-supervised model followed by supervised model achieves a classification accuracy of about 86.2 percent on the holdout dataset, even better than the semi-supervised learning used alone that achieved an accuracy of about 85.6 percent.” (Jason Section “Label Propagation for Semi-Supervised Learning” Paragraph 25).
Claim(s) 8 is rejected under 35 U.S.C. 103 as being unpatentable over Kun Cao et al; US 12566973 B1 filed on Sep 27, 2021 (hereinafter “Cao”) in view of Alexander S Polichroniadis et al; US 12112519 B1 filed on Feb 23, 2022 (hereinafter “Polichroniadis”) in further view of bradleyboehmke; “Chapter 17 Principal Component Analysis” available online on Dec 23, 2022 (hereinafter “Bradley”) in further view of Stanislav Olegovich Ponomarev; US 20210042471 A1 filed on Jul 27, 2020 (hereinafter “Ponomarev”) in further view of Ravish Raj; “K-Nearest Neighbors (KNN) Algorithm in Machine Learning” available online Dec 27, 2022 (hereinafter “Raj”) in further view of Sewak; US 20220414137 A1 filed on Apr 1, 2022 (hereinafter “Sewak”)
Regarding claim 8, Cao does not teach The method of claim 2, further comprising: determining a first outlier score based on the first distance;
comparing the first outlier score to a first threshold outlier score;
and selecting a recommendation from a plurality of recommendations based on comparing the first outlier score to a first threshold outlier score.
However, Sewak does teaches The method of claim 2, further comprising: determining a first outlier score based on the first distance; (Sewak Paragraph 0106; "Now assume that the search score of the search engine is cosine similarity between the documents in a semantic space yielding associated scores (GR-EX-L1=0.5, GR-EX-AL2=0.3, GR-EX-AL1=0.21, GR-EX-L2=0.05)." Examiner notes that a first outlier score (search score) is determined based on the first distance (cosine similarity))
comparing the first outlier score to a first threshold outlier score; (Sewak Paragraph 0106; "if a search score threshold of 0.08 is used." Examiner notes that first outlier score (search score) is compared to a first threshold outlier score (search score threshold))
and selecting a recommendation from a plurality of recommendations based on comparing the first outlier score to a first threshold outlier score. (Sewak Paragraph 0106; "reconciliation rule 3 would choose anti-label if a search score threshold of 0.08 is used." Examiner notes that based on comparing the first outlier score to a first threshold outlier score, select a recommendation from a plurality of recommendations (choose or not choose anti-label))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, and Sewak. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. Raj teaches KNN algorithm. Sewak teaches a method for automatic labeling of text data. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, and Sewak to improve the success rate of classification, while maintaining improved efficiency “The success rate of the classification can be improved, while maintaining this improved efficiency, by obtaining a second generative result from a generative model and estimating label probability using the second generative result.” (Sewak Paragraph 0008).
Claim(s) 9 and 11 are rejected under 35 U.S.C. 103 as being unpatentable over Kun Cao et al; US 12566973 B1 filed on Sep 27, 2021 (hereinafter “Cao”) in view of Alexander S Polichroniadis et al; US 12112519 B1 filed on Feb 23, 2022 (hereinafter “Polichroniadis”) in further view of bradleyboehmke; “Chapter 17 Principal Component Analysis” available online on Dec 23, 2022 (hereinafter “Bradley”) in further view of Stanislav Olegovich Ponomarev; US 20210042471 A1 filed on Jul 27, 2020 (hereinafter “Ponomarev”) in further view of Ravish Raj; “K-Nearest Neighbors (KNN) Algorithm in Machine Learning” available online Dec 27, 2022 (hereinafter “Raj”) in further view of Ahmet et al; “Label Propagation for Deep Semi-Supervised learning” published on 2019 (hereinafter “Ahmet”)
Regarding claim 9, Cao does not teach The method of claim 2, wherein the plurality of messages comprises a first subset having ground truth labels, and a second subset, wherein the second subset comprises between 90 and 99.99 percent of the plurality of messages
However, Ahmet does teach The method of claim 2, wherein the plurality of messages comprises a first subset having ground truth labels, and a second subset, wherein the second subset comprises between 90 and 99.99 percent of the plurality of messages (Ahmet Page 5075 Paragraph 3; "The training set consists of 50k images coming from 10 classes, while the test set consists of 10k images from the same 10 classes. All images have resolution 32 × 32. Evaluation is performed with 50, 100, 200, and 400 labeled images per classes, corresponding to l = 500, 1k, 2k, and 4k label images in total." Examiner notes that first subset (ground truth labeled images) contains 500 labeled images and second subset (unlabeled images) contains 50000 - 500 = 49500 unlabeled images; second subset makes up 90 percent of the dataset 49500/50000 = 0.99; Examiner interprets images as messages because people can message each other with pictures)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, and Ahmet. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. Raj teaches KNN algorithm. Ahmet teaches a method for label propagation for deep semi-supervised learning. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, and Ahmet to perform the method on more sparse labeled data for a larger benefit “The proposed approach performs the best out of the pseudo-label based approaches on CIFAR-10. Results in Figure 6 show that our benefit is larger when the number of labels is reduced” (Ahmet Page 5077 Paragraph 4).
Regarding claim 11, Cao does not teach The method of claim 9, wherein the first subset comprises the first message having the first preassigned label,
and wherein the method further comprises: comparing the first preassigned label to a corresponding ground truth label from the ground truth labels of the first subset;
and determining an evaluation of the artificial intelligence model in response to comparing the first preassigned label to the corresponding ground truth label.
However, Ahmet does teach The method of claim 10, wherein the first subset comprises the first message having the first preassigned label, (Ahmet Page 5075 Paragraph 3; "Evaluation is performed with 50, 100, 200, and 400 labeled images per classes" Examiner notes that first message (labeled images) comprises the first input (image) having the first preassigned label (label per class); Examiner interprets images a messages)
and wherein the method further comprises: comparing the first preassigned label to a corresponding ground truth label from the ground truth labels of the first subset; (Ahmet Fig 4 and Page 5077 Paragraph 1; "In Figure 4, we report the progress of the pseudo-label accuracy on unlabeled images XU throughout the training." Examiner notes that comparing the first preassigned label (ground truth) to a corresponding ground truth label (predicted pseudo label) from the ground truth labels of the first subset is represented as prediction accuracy)
and determining an evaluation of the artificial intelligence model in response to comparing the first preassigned label to the corresponding ground truth label.
(Ahmet Page 5077 Paragraph 1; "Diffusion predictions are consistently better than network predictions." Examiner notes that evaluation is determined of the artificial intelligence model (diffusion predication outperforms network predictions) in response to comparing (prediction accuracy))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, and Ahmet. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. Raj teaches KNN algorithm. Ahmet teaches a method for label propagation for deep semi-supervised learning. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, and Ahmet to perform the method on more sparse labeled data for a larger benefit “The proposed approach performs the best out of the pseudo-label based approaches on CIFAR-10. Results in Figure 6 show that our benefit is larger when the number of labels is reduced” (Ahmet Page 5077 Paragraph 4).
Claim(s) 12 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Kun Cao et al; US 12566973 B1 filed on Sep 27, 2021 (hereinafter “Cao”) in view of Alexander S Polichroniadis et al; US 12112519 B1 filed on Feb 23, 2022 (hereinafter “Polichroniadis”) in further view of bradleyboehmke; “Chapter 17 Principal Component Analysis” available online on Dec 23, 2022 (hereinafter “Bradley”) in further view of Stanislav Olegovich Ponomarev; US 20210042471 A1 filed on Jul 27, 2020 (hereinafter “Ponomarev”) in further view of Ravish Raj; “K-Nearest Neighbors (KNN) Algorithm in Machine Learning” available online Dec 27, 2022 (hereinafter “Raj”) in further view of Dane et al; US 20250022615 A1 filed on Nov 14, 2022 (hereinafter “Dane”)
Regarding claim 12, Cao does not teach The method of claim 2, wherein the plurality of messages comprises a number of unique tokens, wherein the first number of dimensions of the semantic graph has original dimensions, and wherein the original dimensions have a number less than or equal to the number of unique tokens.
However, Dane does teach The method of claim 2, wherein the plurality of messages comprises a number of unique tokens, wherein the first number of dimensions of the semantic graph has original dimensions, and wherein the original dimensions have a number less than or equal to the number of unique tokens. (Dane Fig 4 and Paragraph 0111; "The word piece tokens 403 and the positional embeddings 407 may simply be summed to form the input representation of the input text sequence 401… The classification layer 409 is trained to output a probability 410 for each of the possible unique biological target identifiers 411" Examiner notes that plurality of messages (input text sequence 401) has a number of unique tokens (word piece tokens), wherein the semantic graph has original dimensions (possible unique biological target identifier), and wherein the original dimensions have a number less than or equal to the number of unique tokens (10 tokens > 3 identifiers))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, and Dane. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. Raj teaches KNN algorithm. Dane teaches using knowledge graphs to predict new biological targets. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, and Dane to leverage the advantages of a knowledge graph to enhance the ability of the model “This method therefore benefits from the advantages associated with knowledge graph inference and language models to further enhance the ability of the model to determine biological entities of interest for a given user-specified biological context.” (Dane Paragraph 0026).
Regarding claim 19, claim 19 has similar limitations as of claim 12, except it is a non-transitory, computer readable media claim, therefore it is rejected under the same rationale as claim 12.
Claim(s) 17 is rejected under 35 U.S.C. 103 as being unpatentable over Kun Cao et al; US 12566973 B1 filed on Sep 27, 2021 (hereinafter “Cao”) in view of Alexander S Polichroniadis et al; US 12112519 B1 filed on Feb 23, 2022 (hereinafter “Polichroniadis”) in further view of bradleyboehmke; “Chapter 17 Principal Component Analysis” available online on Dec 23, 2022 (hereinafter “Bradley”) in further view of Stanislav Olegovich Ponomarev; US 20210042471 A1 filed on Jul 27, 2020 (hereinafter “Ponomarev”) in further view of Ravish Raj; “K-Nearest Neighbors (KNN) Algorithm in Machine Learning” available online Dec 27, 2022 (hereinafter “Raj”) in further view of Nitzan et al; US 20170204455 A1 filed on Jul 17, 2015 (hereinafter “Nitzan”).
Regarding claim 17, Cao does not teach The method of claim 13, wherein the threshold amount of information is between ninety-five and ninety-nine percent of the total amount of information in the semantic graph.
However, Nitzan does teach The method of claim 13, wherein the threshold amount of information is between ninety-five and ninety-nine percent of the total amount of information in the semantic graph. (Nitzan Paragraph 0040; "the cumulative distribution function (CDF) value of that genetic variant reaches a predefined threshold value (CDF_thresh) of 0.99, 0.995, 0.999, 0.9999, 0.99999 or greater." Examiner notes that the cut-off score (threshold value) is between ninety-five and ninety-nine percent of the amount of information in the semantic graph (0.99))
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, Dane, and Nitzan. Cao teaches a framework to generate semantic embeddings from input text content and pass the semantic embeddings to generate graph labels. Polichroniadis teaches inputting semantic embeddings into machine learning methods such as PCA. Bradley teaches principal component analysis. Ponomarev teaches a method of embedding graphs into a semantic multidimensional space by receiving a dataset. Raj teaches KNN algorithm. Nitzan teaches using a probability distribution model to determine a frequency threshold. One of ordinary skill would have motivation to combine Cao, Polichroniadis, Bradley, Ponomarev, Raj, Dane, and Nitzan to perform identification of entities with improved statistical confidence “The method also allows for the identification of the presence of genetic mutations at low frequencies with improved statistical confidence.” (Nitzan Paragraph 0032).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DANIEL DUC TRAN whose telephone number is (571)272-6870. The examiner can normally be reached Mon-Fri 8:00-5:00 EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Viker Lamardo can be reached at (571) 270-5871. 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.
/D.D.T./Examiner, Art Unit 2147
/VIKER A LAMARDO/Supervisory Patent Examiner, Art Unit 2147
Prosecution Insights