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 .
Election/Restrictions
During a telephone conversation with Sarah Newbury on 8/24/2026 a provisional election was made without traverse to prosecute the invention of A method of training a large language model (LLM), claim 1-13, and 20. Affirmation of this election must be made by applicant in replying to this Office action. Claims 14-19 are withdrawn from further consideration by the examiner, 37 CFR 1.142(b), as being drawn to a non-elected invention.
Priority
Application is a continuation of Provisional Application No. 63/770,201, filed
on March 03, 2025.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 05/18/2026 and 06/11/2026 are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Regarding claim 1, in Step 1 of the 101 analyses set forth in MPEP 2106, the claim recites A method of training a large language model (LLM). A method is one of the four statutory categories.
In Step 2a Prong 1 of the 101 analyses set forth in the MPEP 2106, the examiner has determined that the following limitations recite a process that, under the broadest reasonable interpretation, covers a [ mental process/mathematical concept] but for recitation of generic computer components:
processing the training corpus to generate one or more knowledge graphs; (A person can mentally generate one or more knowledge graphs from a training corpus by a process of simply evaluating the training corpus and making a judgement on what data is connected and how they are connected. (MPEP 2106))
extracting, from the one or more knowledge graphs, causal relationships between various nodes of the one or more knowledge graphs; (A person can mentally extract causal relationships between nodes of a knowledge graph by a process of simply evaluating the knowledge graph and making a judgement on what the causal relationships are.)
augmenting the training corpus with the causal relationships extracted from the one or more knowledge graphs; (A person can mentally augment a training corpus by a process of simply evaluating the causal relationships and making a judgement on how the training corpus should be augmented.)
If claim limitations, under their broadest reasonable interpretation, covers performance of the limitations as a [ mental process/mathematical concept] but for the recitation of generic computer components, then it falls within the mental process grouping of abstract ideas. According, the claim “recites” an abstract idea.
In Step 2a Prong 2 of the 101 analyses set forth in MPEP 2106, the examiner has determined that the following additional elements do not integrate this judicial exception into a practical application:
receiving a training corpus for the LLM; (Adding insignificant extra-solution activity (mere data gathering) to the judicial exception (MPEP 2106.05(g))).
and executing a training of the LLM using the augmented training corpus. (Merely training a generic machine learning model constitutes “applying” the machine learning model (MPEP 2106.05(f))).
Since the claim does not contain any other additional elements that are indicative of integration into a practical application, the claim is “directed” to an abstract idea.
In Step 2b of the 101 analyses set forth in the 2019 PEG, the examiner has determined that the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
receiving a training corpus for the LLM; (Adding insignificant extra-solution activity (mere data gathering) to the judicial exception (MPEP 2106.05(g)) Furthermore, the additional element is directed to receiving or transmitting data over a network / performing repetitive calculations / electronic recordkeeping / storing and retrieving information in memory / electronically scanning or extracting data from a physical document, which the courts have recognized as well‐understood, routine, and conventional when they are claimed in a generic manner. See MPEP § 2106.05(d)(II).).
and executing a training of the LLM using the augmented training corpus. (Merely training a generic machine learning model constitutes “applying” the machine learning model (MPEP 2106.05(f)) Furthermore, the additional element is directed to application of a computer tool (machine learning model), which is not indicative of significantly more (MPEP 2106.05(f)).)
Claim 20 is rejected on the same grounds as Claim 1.
Regarding claim 2 it is dependent upon claim 1, and thereby incorporates the limitations of, and corresponding analysis applied to claim 1. Further, claim 2 recites wherein the training corpus comprises multi-modal data. (In step 2A, prong 2, this recites generally linking the use of the judicial exception to a particular technological environment or field of use (MPEP 2106.05(h))). In step 2B, generally linking the use of the judicial exception to a particular technological environment is not indicative of significantly more.)
Regarding claim 3 it is dependent upon claim 2, and thereby incorporates the limitations of, and corresponding analysis applied to claim 2. Further, claim 3 recites wherein one or more knowledge graphs comprise causal graphs indicative of cause-effect relationships among different modalities of the multi-modal data. (In step 2A, prong 2, this recites generally linking the use of the judicial exception to a particular technological environment or field of use (MPEP 2106.05(h))). In step 2B, generally linking the use of the judicial exception to a particular technological environment is not indicative of significantly more.)
Regarding claim 4 it is dependent upon claim 1, and thereby incorporates the limitations of, and corresponding analysis applied to claim 1. Further, claim 4 recites further comprising dynamically updating the one or more knowledge graphs during the training of the LLM. (In step 2A, prong 2, this recites generally linking the use of the judicial exception to a particular technological environment or field of use (MPEP 2106.05(h))). In step 2B, generally linking the use of the judicial exception to a particular technological environment is not indicative of significantly more.)
Regarding claim 5 it is dependent upon claim 4, and thereby incorporates the limitations of, and corresponding analysis applied to claim 4. Further, claim 5 recites wherein the one or more knowledge graphs are dynamically updated using a counterfactual regret minimization (CFRM) process configured to simulate multiple alternative scenarios to predict corresponding outcomes. (In step 2A, prong 2, this recites generally linking the use of the judicial exception to a particular technological environment or field of use (MPEP 2106.05(h))). In step 2B, generally linking the use of the judicial exception to a particular technological environment is not indicative of significantly more.)
Regarding claim 6 it is dependent upon claim 5, and thereby incorporates the limitations of, and corresponding analysis applied to claim 5. Further, claim 6 recites wherein the one or more knowledge graphs are dynamically updated by preserving topological features that appear in multiple of the various alternative scenarios. (In step 2A, prong 2, this recites generally linking the use of the judicial exception to a particular technological environment or field of use (MPEP 2106.05(h))). In step 2B, generally linking the use of the judicial exception to a particular technological environment is not indicative of significantly more.)
Regarding claim 7 it is dependent upon claim 6, and thereby incorporates the limitations of, and corresponding analysis applied to claim 6. Further, claim 7 recites wherein the topological features are identified using persistent homology (PH) analysis. (In step 2A, prong 2, this recites generally linking the use of the judicial exception to a particular technological environment or field of use (MPEP 2106.05(h))). In step 2B, generally linking the use of the judicial exception to a particular technological environment is not indicative of significantly more.)
Regarding claim 8 it is dependent upon claim 4, and thereby incorporates the limitations of, and corresponding analysis applied to claim 8. Further, claim 4 recites wherein at least one of the updated knowledge graphs represents a structural prior used during executing an inference process using the trained LLM. (In step 2A, prong 2, this recites generally linking the use of the judicial exception to a particular technological environment or field of use (MPEP 2106.05(h))). In step 2B, generally linking the use of the judicial exception to a particular technological environment is not indicative of significantly more.)
Regarding claim 9 it is dependent upon claim 2, and thereby incorporates the limitations of, and corresponding analysis applied to claim 2. Further, claim 9 recites wherein the one or more knowledge graphs representing the multi-modal data comprises multi-dimensional tensors where different modalities of the multi-modal data are represented as different axes. (In step 2A, prong 2, this recites generally linking the use of the judicial exception to a particular technological environment or field of use (MPEP 2106.05(h))). In step 2B, generally linking the use of the judicial exception to a particular technological environment is not indicative of significantly more.)
Regarding claim 10 it is dependent upon claim 1, and thereby incorporates the limitations of, and corresponding analysis applied to claim 1. Further, claim 10 recites wherein training of the LLM using the augmented corpus comprises: processing data from the augmented corpus by the LLM to generate a model-response; (In step 2A prong 2, merely training a generic machine learning operation constitutes “applying” the machine learning operation (MPEP 2106.05(f)). In step 2B, merely applying a generic machine learning operation is not indicative of significantly more.) processing the model-response by at least one auxiliary machine-learning model (In step 2A prong 2, merely training a generic machine learning operation constitutes “applying” the machine learning operation (MPEP 2106.05(f)). In step 2B, merely applying a generic machine learning operation is not indicative of significantly more.) to evaluate a quality of the model response; (In step 2A, prong 1, this recites a mental process without significantly more. A person can mentally evaluate the quality of a model response by a process of simply evaluating the model response and making a judgement on how good the response was (MPEP 2106).) and updating training of the LLM in accordance with the quality of the model response. (In step 2A, prong 2, this recites generally linking the use of the judicial exception to a particular technological environment or field of use (MPEP 2106.05(h))). In step 2B, generally linking the use of the judicial exception to a particular technological environment is not indicative of significantly more.)
Regarding claim 11 it is dependent upon claim 10, and thereby incorporates the limitations of, and corresponding analysis applied to claim 10. Further, claim 11 recites wherein the at least one auxiliary machine-learning model is a generative artificial intelligence model configured to evaluate the model-output based on a prompt. (In step 2A, prong 2, this recites generally linking the use of the judicial exception to a particular technological environment or field of use (MPEP 2106.05(h))). In step 2B, generally linking the use of the judicial exception to a particular technological environment is not indicative of significantly more.)
Regarding claim 12 it is dependent upon claim 1, and thereby incorporates the limitations of, and corresponding analysis applied to claim 1. Further, claim 12 recites wherein executing the training of the LLM comprises updating weights of a self-attention mechanism associated with the LLM based on the causal relationships extracted from the one or more knowledge graphs. (In step 2A, prong 2, this recites generally linking the use of the judicial exception to a particular technological environment or field of use (MPEP 2106.05(h))). In step 2B, generally linking the use of the judicial exception to a particular technological environment is not indicative of significantly more.)
Regarding claim 13 it is dependent upon claim 12, and thereby incorporates the limitations of, and corresponding analysis applied to claim 12. Further, claim 13 recites wherein similarity measures associated with the self-attention mechanism are dynamically modified as a function of the updated weights. (In step 2A, prong 2, this recites generally linking the use of the judicial exception to a particular technological environment or field of use (MPEP 2106.05(h))). In step 2B, generally linking the use of the judicial exception to a particular technological environment is not indicative of significantly more.)
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 1-3, 8-9 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Unnikrishnan et al. Pub. No.: US 20240420205 A1 in view of Yamaura et al. Pub No.: US 20250307663 A1.
Regarding claim 1 Unnikrishnan teaches A method of training a large language model (LLM), the method comprising: receiving a training corpus for the LLM; (Unnikrishnan, Paragraph 0023, 0053, 0083, 0086, 0110, teaches the receiving of a training set of data that will be used to train a large language model.)
processing the training corpus to generate one or more knowledge graphs; 1… (Unnikrishnan, Paragraph 0021, teaches the use of an AI model to generate knowledge graphs based on the relationships present in the training data.)
… augmenting the training corpus with the causal relationships extracted from the one or more knowledge graphs; (Unnikrishnan, Paragraphs 0021, 0027, 0057, 0097, teaches the use of the relationships that are stored in the knowledge graphs to augment the training data that is used to train the LLM.)
and executing a training of the LLM using the augmented training corpus. (Unnikrishnan, Paragraphs 0027, teaches the training of a language model using the knowledge graph relationships in the training data.)
Unnikrishnan does not teach 1… extracting, from the one or more knowledge graphs, causal relationships between various nodes of the one or more knowledge graphs; … However, Yamaura in analogous art teaches this limitation (Yamaura, paragraph 0053 – 0054, teaches the creation of knowledge graphs that contain causal relationships through diverse types of data.)
It would have been obvious to one skilled in the art before the effective filing date of the claimed invention to combine Yamaura’s teaching of causal relationships in knowledge graphs with Unnikrishnan’s teaching of Language models that utilize knowledge graphs to modify the training data. The motivation to do so would be to allow the language model to better understand causal reasoning in relationships.
Claim 20 is rejected on the same grounds as Claim 1.
Regarding claim 2 the combination of Unnikrishnan and Yamura teaches The method of claim 1, wherein the training corpus comprises multi-modal data. (Unnikrishnan, Paragraphs 0023, teaches the training data used in the knowledge graphs and training of the language models, being multi-modal data.)
Regarding claim 3 the combination of Unnikrishnan and Yamura teaches teaches The method of claim 2, wherein the one or more knowledge graphs comprise causal graphs indicative of cause-effect relationships among different modalities of the multi-modal data. (Yamaura, paragraph 0053 – 0054, teaches the creation of knowledge graphs that contain causal relationships through diverse types of data.)
Regarding claim 8 the combination of Unnikrishnan and Yamura teaches teaches The method of claim 4, wherein at least one of the updated knowledge graphs represents a structural prior used during executing an inference process using the trained LLM. (Unnikrishnan, Paragraph 0027, 0042, 0051, teaches the use of the knowledge graph as an integral part of the inference process of the language model meaning that the knowledge graph represents a structural prior.)
Regarding claim 9 the combination of Unnikrishnan and Yamura teaches teaches The method of claim 2, wherein the one or more knowledge graphs representing the multi-modal data comprises multi-dimensional tensors where different modalities of the multi-modal data are represented as different axes. (Unnikrishnan, Paragraph 0027, 0031, 0050, 0053, 0095, teaches use of vector (first order tensor) representations of multi-modal data in the knowledge graphs. Further paragraph 0031 teaches the calculation of the Euclidian distance which means that the different multi-modal data are represented as the different axes.)
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Unnikrishnan et al. Pub. No.: US 20240420205 A1 in view of Yamaura et al. Pub No.: US 20250307663 A1 in further view of Khuti Pub. No.: US 20250272771 A1.
Regarding claim 4 the combination of Unnikrishnan and Yamura teaches The method of claim 1,
The combination of Unnikrishnan and Yamaura does not teach further comprising dynamically updating the one or more knowledge graphs during the training of the LLM. However, Khuti in analogous art teaches this limitation (Khuti, paragraph 0024, teaches the dynamic updating of the LLM and its knowledge graphs at the same time.)
It would have been obvious to one skilled in the art before the effective filing date of the claimed invention to combine Khuti’s teaching of dynamic updating with The combination of Unnikrishnan and Yamaura’s teaching of Language models that utilize knowledge graphs to modify the training data. The motivation to do so would be to allow the language model and knowledge graph to constantly learn new information as they are trained and put to use.
Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over Unnikrishnan et al. Pub. No.: US 20240420205 A1 in view of Yamaura et al. Pub No.: US 20250307663 A1 and Khuti Pub. No.: US 20250272771 A1 in further view of Xu et al. Pub. No.: US 20250053776 A1.
Regarding claim 5 The combination of Unnikrishnan, Yamaura, and Khuti teaches The method of claim 4,
The combination of Unnikrishnan, Yamaura, and Khuti does not teach wherein the one or more knowledge graphs are dynamically updated using a counterfactual regret minimization (CFRM) process configured to simulate multiple alternative scenarios to predict corresponding outcomes. However, Xu in analogous art teaches this limitation (Xu, paragraphs 0044-0048, teaches the use of counterfactual causal relationships to expand and update knowledge graph to help learn different counterfactual scenarios.)
It would have been obvious to one skilled in the art before the effective filing date of the claimed invention to combine Xu’s teaching of counter factual learning in language models with The combination of Unnikrishnan, Yamaura, and Khuti’s teaching of Language models that utilize knowledge graphs to modify the training data. The motivation to do so would be to allow the language model and knowledge graph to learn opposite relationships and predict unknown information to increase the accuracy of the model.
Claims 6-7 are rejected under 35 U.S.C. 103 as being unpatentable over Unnikrishnan et al. Pub. No.: US 20240420205 A1 in view of Yamaura et al. Pub No.: US 20250307663 A1, Khuti Pub. No.: US 20250272771 A1 and Xu et al. Pub. No.: US 20250053776 A1 in further view of Iu et al. Pub. No.: US 20240273286 A1.
Regarding claim 6 The combination of Unnikrishnan, Yamaura, Khuti, and Xu teaches The method of claim 5,
The combination of Unnikrishnan, Yamaura, Khuti, and Xu does not teach wherein the one or more knowledge graphs are dynamically updated by preserving topological features that appear in multiple of the various alternative scenarios. However, Iu in analogous art teaches this limitation (Iu, paragraph 0086, teaches the preservation of the knowledge graphs mapping (i.e. topological features) during the knowledge graph updating process.)
It would have been obvious to one skilled in the art before the effective filing date of the claimed invention to combine Iu’s teaching of preserving topological features in updating with The combination of Unnikrishnan, Yamaura, Khuti, and Xu’s teaching of Language models that utilize knowledge graphs to modify the training data. The motivation to do so would be to allow the language model and knowledge graph to maintain previously discovered relationships in knowledge graphs to prevent the forgetting of learned information.
Regarding claim 7 The combination of Unnikrishnan, Yamaura, Khuti, Xu, and Iu teaches The method of claim 6, wherein the topological features are identified using persistent homology (PH) analysis. (Iu, paragraph 0098, teaches the identifying of topological features by using keys to create relationships and join data in different tables. The relationships between different data are the topological features and the keys that represent these relationships represent identifying the features using persistent homology.)
Claims 10 and 12-13 are rejected under 35 U.S.C. 103 as being unpatentable over Unnikrishnan et al. Pub. No.: US 20240420205 A1 in view of Yamaura et al. Pub No.: US 20250307663 A1 A1 in further view of Iu et al. Pub. No.: US 20240273286 A1.
Regarding claim 10 The combination of Unnikrishnan, and Yamaura teaches The method of claim 1, wherein training of the LLM using the augmented corpus comprises: processing data from the augmented corpus by the LLM to generate a model-response; (Unnikrishnan, Paragraph 0040, 0076, 0089, teaches the use of the training data that has been augmented by the knowledge graphs to produce model responses by processing the training data.)
The combination of Unnikrishnan, and Yamaura does not teach processing the model-response by at least one auxiliary machine-learning model to evaluate a quality of the model response; However, Iu in analogous art teaches this limitation (Iu, paragraph 0057, teaches an auxiliary pre-publication or post-publication feedback model that is able to evaluate the quality of the model response to improve the quality of the generative model.)
Further the The combination of Unnikrishnan, and Yamaura does not teach and updating training of the LLM in accordance with the quality of the model response. However, Iu in analogous art teaches this limitation (Iu, paragraph 0057-0060, teaches the refinement of the language model based on the quality feedback and making modifications to the language models based on the feedback.)
It would have been obvious to one skilled in the art before the effective filing date of the claimed invention to combine Iu’s teaching of model quality evaluation with The combination of Unnikrishnan, and Yamaura’s teaching of Language models that utilize knowledge graphs to modify the training data. The motivation to do so would be to allow the language model and knowledge graph to be effectively evaluated and scored in order to know how well the models are functioning and determine if they need to be tuned or retrained.
Regarding claim 12 The combination of Unnikrishnan, and Yamaura teaches The method of claim 1,
The combination of Unnikrishnan, and Yamaura does not teach wherein executing the training of the LLM comprises updating weights of a self-attention mechanism associated with the LLM based on the causal relationships extracted from the one or more knowledge graphs. However, Iu in analogous art teaches this limitation (Iu, paragraph 0053, teaches the use of self-attention layers that allow the model to assign different weights to different words or phrases based on the complex relationships between the words and phrases in different contexts.) It would have been obvious to one skilled in the art before the effective filing date of the claimed invention to combine Iu’s teaching of model quality evaluation with The combination of Unnikrishnan, and Yamaura’s teaching of Language models that utilize knowledge graphs to modify the training data. The motivation to do so would be to allow the language model and knowledge graph to be effectively evaluated and scored in order to know how well the models are functioning and determine if they need to be tuned or retrained.
Regarding claim 13 The combination of Unnikrishnan, Yamaura and Iu teaches The method of claim 12, wherein similarity measures associated with the self-attention mechanism are dynamically modified as a function of the updated weights. (Iu, paragraph 0053 – 0058, teaches the updating of similarity metrics based on the attention layers that are dynamically updated updated.)
Claim 11 rejected under 35 U.S.C. 103 as being unpatentable over Unnikrishnan et al. Pub. No.: US 20240420205 A1 in view of Yamaura et al. Pub No.: US 20250307663 A1 A1 and Iu et al. Pub. No.: US 20240273286 A1 in further view of Mukherjee.
Regarding claim 11 The combination of Unnikrishnan, Yamaura and Iu teaches The method of claim 10,
Further the The combination of Unnikrishnan, Yamaura, and Iu does not teach wherein the at least one auxiliary machine-learning model is a generative artificial intelligence model configured to evaluate the model-output based on a prompt. However, Mukherjee in analogous art teaches this limitation (Mukherjee, paragraph 0029, 0035, teaches the use of generative artificial intelligence models configured to evaluate other model outputs based on the prompts given to the agents that use the generative models.)
It would have been obvious to one skilled in the art before the effective filing date of the claimed invention to combine Mukherjee’s teaching of generative model that evaluates another model with The combination of Unnikrishnan, Yamaura and Iu’s teaching of Language models that utilize knowledge graphs to modify the training data. The motivation to do so would be to allow the language model and knowledge graph evaluation model to be prompted by the user to evaluate specific aspects of the models for targeted improvements.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to THOMAS B LANE whose telephone number is (571)272-1872. The examiner can normally be reached M-Th: 7:20am-5:20pm; F: Out of Office.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, MARIELA REYES can be reached at (571) 270-1006. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/THOMAS BERNARD LANE/ Examiner, Art Unit 2142
/HAIMEI JIANG/ Primary Examiner, Art Unit 2142
Prosecution Insights