Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Arguments
Applicant’s argument filed 02/17/2026 have been fully considered but they are not persuasive. The remarks contain inconsistency with what has been filed. Claims 1-20 remains pending, no claims have been cancelled, and claims 1-20 are amended. The interview was conducted on 02/13/2026 with Examiner Gary Mac. The previous Office Action (dated 11/17/2025) did not have any claim objections to claim 8.
Applicant’s Argument: On page 12-14 of Applicant’s response to rejections under 35 U.S.C. 101, applicant states that the amended claims recite a technological improvement in ML operations. The technological improvement in ML operations is defined by the process of selecting a feature from an answer space, forming an enhanced input, and generating an output in the amended claims.
Examiner’s Response: Applicant’s argument is not persuasive. During examination, the examiner should analyze the "improvements" consideration by evaluating the specification and the claims to ensure that a technical explanation of the asserted improvement is present in the specification, and that the claim reflects the asserted improvement (see MPEP §2106.05(a)). The MPEP (§2106.05(a)(II)) also warns, “it is important to keep in mind that an improvement in the abstract idea itself (e.g. a recited fundamental economic concept) is not an improvement in technology.” Here, the alleged improvement in the form of “selecting a feature from an answer space and forming an enhanced input to generate an output” is an improvement to the abstract idea of a mental process that can be performed in the human mind.
An important consideration in determining whether a claim improves technology is the extent to which the claim covers a particular solution to a problem or a particular way to achieve a desired outcome, as opposed to merely claiming the idea of a solution or outcome (see MPEP 2106.05(a)). The amended claims do not provide sufficient details to describe any technological improvement. If the specifications explicitly set forth an improvement but in a conclusory manner (see MPEP 2106.04(d)(1): a bare assertion of an improvement without the detail necessary to be apparent to a person of ordinary skill in the art), the examiner should not determine the claim improves technology.
Applicant’s Argument: On page 15-17 of Applicant’s response to rejections under 35 U.S.C. 102 and 103, applicant states Zhang and Li fails to disclose the amended limitations. Applicant further discloses details of the interview in regards to Eiden and Saito.
Examiner’s Response: Applicant’s argument is not persuasive. Applicant’s arguments with respect to claim 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.
During the interview, the Eiden and Saito references were not discussed and the amended claims recited by the Applicant is different from the claims filed for this application. The Examiner Interview Summary Record (dated 02/13/2026) discloses details of the interview conducted.
Claim Objections
Claims 7, and 15 objected to because of the following informalities:
The claim limitation “wherein selecting the training data comprises
retrieving, by the computing system executing the second ML model, in response to the modified prompt template, the training data from the training dataset of the first ML model that is semantically similar data to the input” is missing a colon to separate the independent clauses. The colon should be added after “comprises”.
Appropriate correction is required.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-20 rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claims 1, 9, and 17 recites “selecting, by the computing system executing the second ML model, a feature from an answer space” and “generating, by the computing system executing the first ML model, an output for the enhanced input”. In the first recited claim limitation, it is not clear which is performing the selecting operation. It is not clear whether the selecting is performed by the computing system or the selecting is performed during the execution of the second ML model, where the process of executing the second ML model causes the second ML model to perform the selecting operation. Examiner interprets the claim limitation to mean that the second ML model performs selecting a feature when the second ML model is executed by the computing system.
Similarly, it is not clear which is performing the generating operation in the second recited example. It is not clear whether the generating is performed by the computing system or the generating is performed during the execution of the first ML model, where the process of executing the first ML model causes the first ML model to perform the generating operation. Examiner interprets the claim limitation to mean that the first ML model performs generating an output when the first ML model is executed by the computing system.
Dependent claims 2, 5, 6, 7, 10, 13, 14, 15, 18, and 19 contains claim limitations that have the same indefiniteness as the examples above. These claims recite “[operation], by the computing system executing the second/first ML model, ...”. It is not clear which is performing the operation in these claims. Examiner interprets the recited ML model to perform the recited operation.
Claims 2-8, 10-16, and 18-20 are dependent claims of claims 1, 9, and 17. Therefore, the dependent claims are rejected under the same basis as the parent claims.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding Claim 1:
Subject Matter Eligibility Analysis Step 1:
Claim 1 recites “A computer-implemented method, comprising” and is thus a process, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
“selecting, ” (a mental process that can be performed in the human mind, i.e. judgement)
“forming, adding the feature to the input” (a mental process that can be performed in the human mind, i.e. judgement; Modifying data to include additional known information)
“generating, enhanced input based on the feature in the enhanced input, without modifying the first ML model between receipt of the enhanced input , wherein the output for the enhanced input has an improved accuracy relative to another output for the input without the feature” (a mental process that can be performed in the human mind with the aid of pen and paper, i.e. judgement; generate a prediction from data and ensuring the provided data improves the algorithm)
Claim 1 therefore recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
"receiving, by a computing system that includes a first machine learning (ML) model and a second ML model that is separate from the first ML model, an input, the first ML model being deployed” (This step is directed to data gathering, which is understood to be insignificant extra solution activity - see MPEP 2106.05(g))
"providing, by the computing system, the enhanced input to the first ML model” (This step is directed to data gathering, which is understood to be insignificant extra solution activity - see MPEP 2106.05(g))
“... by the computing system executing the second ML model, ... (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
“... by the computing system, ... (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
“generating, by the computing system executing the first ML model, by the first ML model and generation of the output by the first ML model, wherein the output generated by the first ML model generated by the first ML model (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
The additional elements as disclosed above alone or in combination do not integrate the judicial exception into practical application as they are mere insignificant extra solution activity in combination of generic computer functions being implemented with generic computer elements in a high level of generality to perform the disclosed abstract idea above. Therefore, Claim 1 is directed to the abstract idea.
Subject Matter Eligibility Analysis Step 2B:
"receiving, by a computing system that includes a first machine learning (ML) model and a second ML model that is separate from the first ML model, an input, the first ML model being deployed” (This step is directed to transmitting or receiving information, which is understood to be insignificant extra solution activity and well understood, routine and conventional activity of transmitting and receiving data as identified by the court - see MPEP 2106.05(d))
"providing, by the computing system, the enhanced input to the first ML model” (This step is directed to transmitting or receiving information, which is understood to be insignificant extra solution activity and well understood, routine and conventional activity of transmitting and receiving data as identified by the court - see MPEP 2106.05(d))
“... by the computing system executing the second ML model, ... (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
“... by the computing system, ... (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
“generating, by the computing system executing the first ML model, by the first ML model and generation of the output by the first ML model, wherein the output generated by the first ML model generated by the first ML model (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
The additional elements as disclosed above alone or in combination do not recite significantly more than the abstract idea itself as they are mere insignificant extra solution activity in combination of generic computer functions being implemented with generic computer elements in a high level of generality to perform the disclosed abstract idea above. Therefore, Claim 1 is subject-matter ineligible.
Regarding Claim 9:
The claim recites a system (“A computer system for enhancing deployed machine learning (ML) models using prompt learning, comprising”) that performs the method as described in claim 1. Therefore, claim 9 is rejected for the same reasons as disclosed for claim 1. The limitations for additional elements of claim 9 are analyzed below.
Subject Matter Eligibility Analysis Step 2A Prong 1:
Please see Step 2A Prong 1 analysis of claim 1
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“one or more processors, one or more computer-readable memories, one or more computer-readable tangible storage media, and program instructions stored on at least one of the one or more computer-readable tangible storage media for execution by at least one of the one or more processors via at least one of the one or more memories, wherein the computer system is capable of performing a method comprising” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Regarding Claim 17:
The claim recites an article of manufacture (“A computer program product for enhancing deployed machine learning (ML) models using prompt learning, comprising”) that performs the method as described in claim 1. Therefore, claim 17 is rejected for the same reasons as disclosed for claim 1. The limitations for additional elements of claim 17 are analyzed below.
Subject Matter Eligibility Analysis Step 2A Prong 1:
Please see Step 2A Prong 1 analysis of claim 1
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“one or more computer-readable storage media and program instructions collectively stored on the one or more computer-readable storage media, the program instructions executable by a processor to cause the processor to perform a method comprising” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Regarding Claims 2, 10, and 18:
Subject Matter Eligibility Analysis Step 2A Prong 1:
“mapping, from the training dataset of the first ML model, wherein the training data includes the feature” (a mental process, i.e. judgement)
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“by the computing system executing the second ML model, ” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Regarding Claims 3, and 11:
Subject Matter Eligibility Analysis Step 2A Prong 1: None
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“wherein the training data further comprises a label associated with the feature, and wherein the output generated by the first ML model includes the label of the training data associated with the feature” (merely specifies a particular technological environment in which the abstract idea is to take place, ie. a field of use, and thus does not integrate the abstract idea into a practical application nor cannot provide significantly more than the abstract idea itself - see MPEP 2106.05(h))
Regarding Claims 4, and 12:
Subject Matter Eligibility Analysis Step 2A Prong 1: None
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“wherein the second ML model is trained to map the input to a training data from the training dataset of the first ML model that is semantically similar to the input” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Regarding Claims 5, and 13:
Subject Matter Eligibility Analysis Step 2A Prong 1:
“recognizing, the first ML model” (a mental process, i.e. judgement)
“predicting, the output as a label corresponding to the training data” (a mental process, i.e. judgement)
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“..., by the computing system executing the first ML model, ...” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Regarding Claims 6, 14, and 19:
Subject Matter Eligibility Analysis Step 2A Prong 1:
“processing, semantically similar to the input” (a mental process, i.e. judgement)
“selecting, a training data, from the training dataset of the first ML model, that is semantically similar to the input” (a mental process, i.e. judgement)
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“..., by the computing system executing the second ML model, ...” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Regarding Claims 7, and 15:
Subject Matter Eligibility Analysis Step 2A Prong 1:
“modifying, a prompt template to include the input and ” (a mental process, i.e. judgement)
“wherein selecting the training data comprises retrieving, the training data from the training dataset of the first ML model that is semantically similar data to the input” (a mental process, i.e. judgement)
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“, by the computing system, ” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
“providing the modified prompt template to the second model” (This step is directed to data gathering, which is understood to be insignificant extra solution activity (2106.05(g) in step 2A prong 2) and well understood, routine and conventional activity of transmitting and receiving data as identified by the court (2106.05(d) in step 2B))
“, by the computing system executing the second ML model, ” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Regarding Claims 8, 16, and 20:
Subject Matter Eligibility Analysis Step 2A Prong 1:
“generating, a first set of embeddings associated with the training dataset of the first ML model” (a mental process that can be perform in the human mind with the aid of pen and paper, i.e. judgement; converting data into a vector representation)
“generating, a second set of embeddings by applying a random dropout to the generated first set of embeddings, wherein the generated first set of embeddings and the generated second set of embeddings are similar in a vector space” (a mental process that can be perform in the human mind with the aid of pen and paper, i.e. judgement; selectively removing a subset of values to create a new vector)
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“..., by the computing system, ...” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
“training, by the computing system, the second ML model using the generated second set of embeddings as training input and the training dataset of the first ML model as training output, wherein, for each embedding of the generated second set of embeddings, the trained second ML model returns a corresponding training data from the training dataset of the first ML model” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-7, 9-15, and 17-19 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang (US20250094833A1) in view of Sun “Recitation-Augmented Language Models”.
Regarding claim 1, Zhang teaches:
“A computer-implemented method, comprising” (abstract, A method of training a classification model using instances retrieve from a knowledge base.)
“receiving, by a computing system that includes a first machine learning (ML) model and a second ML model an input, ” ([0039; 0073-0074; Figure 2], A classification model (first machine learning model) receives input instance text to be processed. The classification model consists of the prediction and classification module and a pre-trained language model (second machine learning model). The classification model can perform relationship or emotion classification tasks after parameter optimization.)
“selecting, by the computing system executing the second ML model, a feature from an answer space, the answer space being a training dataset of the first ML model” ([0036; 0039-0041; 0049-0050; 0062-0064], The language model processes the input instance text by transforming the data using a prompt template and generate a query vector. Using k-nearest neighbor method, a plurality of neighboring instance phrases are determined from a knowledge base that are similar to the input. The knowledge base contains data that are used to fine-tune the classification model and thus, the classification model may be trained on the data in the knowledge base. After parameter optimization of the classification model, a classification task can be performed, which causes the language model to query and select neighboring instance text.)
“forming, by the computing system, an enhanced input by adding the feature to the input” ([0039-0042], The pre-trained language model generates a first query vector and finds additional example input from a knowledge base. The neighboring instance phrases from the knowledge base are combined with the input instance text to generate contextual enhancement information. The aggregated results become a new input data for the system.)
“providing, by the computing system, the enhanced input to the first ML model” ([0041-0043], The new input data is provided to the prediction and classification module by the pre-trained language model.)
“generating, by the computing system executing the first ML model, an output for the enhanced input based on the feature in the enhanced input, , wherein the output generated by the first ML model for the enhanced input has an improved accuracy relative to another output generated by the first ML model for the input without the feature” ([0042-0050], The classification model classify and predict the input data provided from the language model. The fine-tuning model of the classification model improves the classification model in scenarios with few and zero samples. Thus, the fine-tuning method of aggregating neighboring instance phrases with input instance text improves the classification model better than simply training the model with a small training dataset with no additional example input.)
Zhang does not explicitly disclose an implementation of “... a second ML model that is separate from the first ML model ...”, “... the first ML model being deployed”, and “generating, by the computing system executing the first ML model, an output for the enhanced input based on the feature in the enhanced input, without modifying the first ML model between receipt of the enhanced input by the first ML model and generation of the output by the first ML model”. However, Sun discloses in the same field of endeavor:
“receiving, by a computing system that includes a first machine learning (ML) model and a second ML model that is separate from the first ML model, an input, the first ML model being deployed” ([pg. 4. Section 3, par. 1; pg. 6, Section 4.1.1, par. 1; pg. 7, Section 4.1.2, par. 1-3; pg. 5, Figure 3], A Natural Questions dataset is used in the experiments to provide the language model with a single-hop question answering task. A trained model, such as PaLM, is deployed to conduct the experiments for the question answering tasks. It is implied that the experiments are conducted on a computer system. The system includes a finetuned reciter language model and a frozen language model as shown in Figure 3.)
“generating, by the computing system executing the first ML model, an output for the enhanced input based on the feature in the enhanced input, without modifying the first ML model between receipt of the enhanced input by the first ML model and generation of the output by the first ML model, wherein the output generated by the first ML model for the enhanced input has an improved accuracy relative to another output generated by the first ML model for the input without the feature” ([abstract; pg. 4. Section 3, par. 1; pg. 4; Section 3.1, par. 2; pg. 9, Section 4.3.2., par. 1; pg. 5, Figure 3], The proposed framework consists of an evidence-recitation module for reciting relevant passages (second ML model) and a separate question-answering module (first ML model) for generating answers given the recited evidence. The recited passages are appended to the beginning of the original question-answer input as a single prompt. Figure 3 shows the finetuned reciter language model and the frozen language model, which is not modified when receiving the aggregated recitations and outputting the answer. The performance of the recite-and-answer method is better compared to standard prompting, which is providing the input prompt directly to the model without any additional evidence.)
It would be obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of “... a second ML model that is separate from the first ML model ...”, “... the first ML model being deployed”, and “generating, by the computing system executing the first ML model, an output for the enhanced input based on the feature in the enhanced input, without modifying the first ML model between receipt of the enhanced input by the first ML model and generation of the output by the first ML model” from Sun into the teaching of Zhang. Doing so improves the prediction performance of a model by utilizing recitation as the intermediate step to provide relevant information before generating the predicted output (Sun, abstract).
Regarding claim 9:
Claim 9 recites a system that performs the same process as described in Claim 1. Therefore claim 9 is rejected under the same reasons mention for claim 1. The additional elements of claim 9 is addressed below by Zhang:
“A computer system for enhancing deployed machine learning (ML) models using prompt learning, comprising” ([abstract, 0036], The system includes a language model to generate improved input data for classification.)
“one or more processors, one or more computer-readable memories, one or more computer-readable tangible storage media, and program instructions stored on at least one of the one or more computer-readable tangible storage media for execution by at least one of the one or more processors via at least one of the one or more memories, wherein the computer system is capable of performing a method comprising” ([0051-0058], The fine-tuning device for classification model consists of multiple units and the functional units may be part of a server. It is inherent that the server is a system with a process and memory that may execute the functional units.)
Regarding claim 17:
Claim 17 recites an article of manufacture that performs the same process as described in Claim 1. Therefore claim 17 is rejected under the same reasons mention for claim 1. The additional elements of claim 17 is addressed below by Zhang:
“A computer program product for enhancing deployed machine learning (ML) models using prompt learning, comprising” ([abstract, 0036], The system includes a language model to generate improved input data for classification.)
“one or more computer-readable storage media and program instructions collectively stored on the one or more computer-readable storage media, the program instructions executable by a processor to cause the processor to perform a method comprising” ([0051-0058], The fine-tuning device for classification model consists of multiple units and the functional units may be part of a server. It is inherent that the server is a system with a process and memory that may execute the functional units.)
Regarding Claims 2, 10, and 18, Zhang teaches:
“mapping, by the computing system executing the second ML model, the input to a training data from the training dataset of the first ML model, wherein the training data includes the feature” ([0036, 0039-0041, 0050], The language model processes the input instance text by transforming the data using a prompt template and generate a query vector. Using k-nearest neighbor method, a plurality of instance phrases are determined from a knowledge base that are similar to the input. The knowledge base contains data that are used to fine-tune the classification model and thus, the classification model may be trained on the data in the knowledge base.)
Regarding Claims 3, and 11, Zhang teaches:
“wherein the training data further comprises a label associated with the feature, and wherein the output generated by the first ML model includes the label of the training data associated with the feature” ([0036-0037, 0042-0047], The instance phrase in the knowledge base contains a value that stores the label truth value. The true value of the label is included as input data for classification. The system computes a classification loss based on the classification and prediction using the true value label. The classifier is finetuned to learn to correctly map the feature of the input data to the correct category. Thus, a finetuned classification model can generate a prediction that is the same as the ground truth label.)
Regarding Claims 4, and 12, Zhang teaches:
“wherein the second ML model is trained to map the input to a training data from the training dataset of the first ML model that is semantically similar to the input” ([0039-0041, 0075], The language model using k-nearest neighbor method, retrieves a plurality of instance phrases from a knowledge base that are similar to the input. The knowledge base contains data that are used to fine-tune the classification model and thus, the classification model may be trained on the data in the knowledge base.)
Regarding Claims 5, and 13, Zhang teaches:
“recognizing, by the computing system executing the first ML model, the feature in the enhanced input as the training data from the training dataset of the first ML model” ([0041-0043; 0049], The input data consists of the input instance text and neighboring instance phrase. The input data is provided to the prediction and classification module. During each training round, the instance phrases are updated to the knowledge base. The instance phrases are used to finetune the classification model and the classification model recognizes the data because the instance phrase may have served as additional example input in a previous iteration.)
“predicting, by the computing system executing the first ML model, the output as a label corresponding to the training data” ([0043-0044], The prediction and classification module of the classification model classify and predict the output from the input data. The output relates to the class that the input data belongs to.)
Regarding Claims 6, 14, and 19, Zhang teaches:
“processing, by the computing system executing the second ML model, a prompt to return data that is semantically similar to the input” ([0039-0041], The pre-trained language model uses a prompt template to extract embedding vectors at masked positions. The pre-trained language model generates a query vector. The query vector is used to find the similar instance phrases using KNN.)
“selecting, by the computing system executing the second ML model, a training data, from the training dataset of the first ML model, that is semantically similar to the input” ([0036, 0040-0041], The language model finds m instance phrases that are closest to the query vector. The instance phrases are retrieved from a knowledge base, which stores decoupled knowledge representation of the classification model.)
Regarding Claims 7, and 15, Zhang teaches:
“modifying, by the computing system, a prompt template to include the input and providing the modified prompt template to the second model” ([0039, 0041], A prompt template is modified with the input instance text. The language model creates a query vector based on the embedding vector of the masked words in the input instance text using the prompt template.)
“wherein selecting the training data comprises retrieving, by the computing system executing the second ML model, in response to the modified prompt template, the training data from the training dataset of the first ML model that is semantically similar data to the input” ([0036, 0040-0042, 0050], The language model finds m instance phrases that are closest to the query vector. The instance phrases are retrieved from a knowledge base, which stores decoupled knowledge representation of the classification model. The information in the knowledge base is used to fine-tune the classification model and the model may have been trained with the data in the knowledge base in a previous training iteration.)
Claims 8, 16, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang (US20250094833A1) in view of Sun “Recitation-Augmented Language Models” and Li “SEDNN: Shared And Enhanced Deep Neural Network Model For Cross-Prompt Automated Essay Scoring”. Li was provided as an NPL reference in the IDS dated 11/29/2022 and thus, Li will be omitted from the Notice of References Cited because a copy of the reference is already in the Application file.
Regarding claims 8, 16, and 20 Zhang in view of Sun teaches:
“generating, by the computing system, a first set of embeddings associated with the training dataset of the first ML model” ([Zhang, 0048-0050, 0059-0060], The classification model is trained on the input instance text and the first embedding of the input instance text is used to generate instance phrases for the knowledge base. After training, the classification model extracts a third embedding vector (first set of embeddings) of the masked words in the input instance text. The input instance text is associated with the classification model because the model is optimized using the input instance text as training data. The third embedding vector is combined with neighboring instance phrase to obtain the input data.)
“generating, by the computing system, a second set of embeddings ” ([Zhang, 0064-0066], The language model extracts a fourth query vector (second set of embeddings) from the input data to find a set of neighboring instance text that is similar to the fourth query vector. An inner product is calculated on the 2 vectors, which shows that the 2 vectors are similar in vector space.)
“training, by the computing system, the second ML model using the generated second set of embeddings as training input and the training dataset of the first ML model as training output, wherein, for each embedding of the generated second set of embeddings, the trained second ML model returns a corresponding training data from the training dataset of the first ML model” ([Zhang, 0036, 0041-0049; 0065-0067], The classification model is optimized using a classification loss and by performing a number of training iterations to improve the model. The classification model consists of a pre-trained language model and the language model is also trained during each training round. The language model generates a query vector to find similar examples in a knowledge base (returns a corresponding training data from the training dataset of the first ML model) and the refinement is based on a loss calculated using the true value of label masked words (training dataset of the first ML model as a training output), which is obtained from the instance phrase in the knowledge base. KNN method is used to determine neighboring instance text from the query vector.)
Zhang in view of Sun does not explicitly disclose an implementation of “generating a second set of embeddings by applying a random dropout to the generated first set of embeddings”. However, Li discloses in the same field of endeavor:
“generating a second set of embeddings by applying a random dropout to the generated first set of embeddings, wherein the generated first set of embeddings and the generated second set of embeddings are similar in a vector space” ([pg. 7, Section 3.4.1, par. 1-2; pg. 8, Section 4.2, par. 2], In training the EModel, a new initial representation (second set of embeddings) is generated by performing operations to the pretrained word embeddings of the essay and the prompt title (first set of embeddings). The 2 set of embeddings are in the similar vector space because a concatenation and subtraction operation can be performed on the vectors. Through this operation, the initial representation of the prompt title can be dynamically changed with different input essays. In addition, the EModel incorporates a dropout parameter value of 0.2, which means that the word embedding layer may randomly drop 20% of its features during training.)
It would be obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of “generating a second set of embeddings by applying a random dropout to the generated first set of embeddings” from Li into the teaching of Zhang in view of Sun. Doing so can enhance a model’s performance by fusing data from multiple source prompts and extracting their shared knowledge into a target prompt (Li, abstract).
Conclusion
THIS ACTION IS MADE FINAL. 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 GARY MAC whose telephone number is (703)756-1517. The examiner can normally be reached Monday - Friday 8:00 AM - 5:00 PM.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Abdullah Kawsar can be reached at (571) 270-3169. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/GARY MAC/Examiner, Art Unit 2127
/ABDULLAH AL KAWSAR/Supervisory Patent Examiner, Art Unit 2127
Prosecution Insights