DETAILED ACTION
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 .
Status of Claims
This action is in reply to the election/restriction response filed on 07/09/2026.
Claims 1-15 are currently pending and have been examined.
Priority
Applicant’s claim to the benefit of and priority to US Provisional Application 63/535,327 is acknowledged. Accordingly, a priority date of 08/30/2023 has been given to this application.
IDS
The information disclosure statement (IDS) submitted on 03/21/24 have been considered by the examiner. The submission is in compliance with the provisions of 37 CFR 1.97.
Subject Matter Eligibility
The claims are understood to be directed to an improvement in the functioning of large language models (LLMs) to reduce the problem of hallucinations of fact in LLMs. Therefore, the claims are subject-matter eligible.
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim(s) 1, 14, 15, is/are rejected under 35 U.S.C. 103 as being unpatentable over Amini et. al. (US Publication 20230386207A1) in view of Roh et. al. (US Publication 20230215560A1).
Regarding Claim 1, Amini discloses generating a first portion of a trainable prompt using negative influential features and positive influential features of a predicted condition ([0200] teaching on forming a training set of features and training labels by identifying a positive training set of features which have a characteristic and a negative training set of features that lack the desired characteristic); and training a second portion of the trainable prompt to steer a pre-trained large language model (PLLM) to generate the safe text using at least the first portion of the trainable prompt ([0202] teaches on training a ML model with feature vectors of the positive training set and the negative training set); the ML system applies the trained ML model to the features of the validation set to quantify the accuracy of the ML model; the model is iteratively re-trained to until the occurrence of a stopping condition, such as an accuracy measurement indicating the model is sufficiently accurate is reached – accuracy is interpreted as reading on “safe text”; [0203] teaches on the ML system being a generative AI system such as a large language model (LLM) to produce data based on the training data set).
Amini does not teach using the trainable prompt in the context of a predicted condition. However, Roh, which is directed to a system for a virtual triage assistant, teaches:
a predicted condition ([0007] teaches on a method including training a machine learning model using training data, where the training data includes predicting a diagnosis based on patient data).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to modify Amini with these teachings of Roh, to use Hong’s specific training data for a predicted condition, with the motivation of using Amini’s system to diagnose a patient with a particular condition (Hong [0104]).
Regarding Claim 14, Amini/Roh teach the limitations of Claim 1. Claim 14 recites limitations that are the same or substantially similar to Claim 1, and the discussion above with respect to Claim 1 is equally applicable to Claim 14. Claim 14 additionally recites the following limitations. Amini additionally discloses the following limitations of Claim 14: a computer system comprising one or more hardware processors ([0018], [0093]).
Regarding Claim 15, Amini/Roh teach the limitations of Claim 1. Claim 15 recites limitations that are the same or substantially similar to Claim 1, and the discussion above with respect to Claim 1 is equally applicable to Claim 15. Claim 15 additionally recites the following limitations. Amini additionally discloses the following limitations of Claim 15: a tangible, non-transitory computer-readable medium having instructions thereon which, upon being executed by one or more processors ([0025]).
Claim(s) 9 is/are rejected under 35 U.S.C. 103 as being unpatentable over Amini et. al. (US Publication 20230386207A1) in view of Roh et. al. (US Publication 20230215560A1) as applied to Claim 1 above, and further in view of Peng et. al. (US Publication 20240070394A1).
Regarding Claim 9, Amini/Roh teach the limitations of Claim 1, but do not teach the following. Peng, which is directed to ensembling soft prompts for tuning language models, teaches: wherein generating the first portion of the trainable prompt using the negative influential features and the positive influential features includes mapping the negative influential features and the positive influential features to a text with templates and a verbalizer ([0022] teaching on PLM (pretrained language model) mapping class labels to a corresponding verbalizer or verbalizer-template sequence).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to modify Amini/Roh with these teachings of Peng, to map the negative influential feature sand positive influential features of Amini to a text with template and verbalizer as taught by Peng, with the motivation of an efficient and effective solution for adapting pre-trained language models to downstream tasks (Peng [0017]).
Claim(s) 13 is/are rejected under 35 U.S.C. 103 as being unpatentable over Amini et. al. (US Publication 20230386207A1) in view of Roh et. al. (US Publication 20230215560A1) as applied to Claim 1 above, and further in view of Clement et. al. (US Publication 20230161567 A1).
Regarding Claim 13, Amini/Roh teach the limitations of Claim 1 but do not teach the following. Clement, which is directed to custom models for source code generation via pre-dix tuning, teaches: wherein the second portion of the trainable prompt includes a number of continuous vectors of a same size as embedding vectors of the PLLM, wherein the embedding vectors are initialized randomly prior to training ([0031], teaching on subtoken embeddings and positional embeddings being kept frozen during prefix tuning process; the embedding contains subtoken matrix and positional embedding matrix; the subtoken embedding matrix contains a vector for each token/subtoken; the embedding dimension is the size of vector of real numbers that represents each unique token/subtoken).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to modify Amini/Roh with these teachings of Clement, include in the trainable prompt of Amini/Roh, a number of continuous vectors of same size as embedding vectors initialized prior to training, with the motivation of fine-tuning to generate a model tailored for a particular downstream task (Clement [0002]).
Subject Matter Free of the Prior Art
Claim 2 is free of the prior art. Kim et. al. (US Publication 20220151540 A1) teaches on an XAI system for diagnosis of mental diseases, which an XAI system determines a brain region corresponding to a diagnosed mental disease among cerebral regions of a patient ([0050]), in which the system determines an important of brain waves and weights them based on influence on determination of importance of a brain wave feature ([0129]-[0131]), but does not teach on the particulars of obtaining patient data and using the patient data as a basis for predicting a disease as the predicted condition using XAI, wherein the XAI provides the negative and positive influential features along with explanation values.
A search of publicly available prior art fails to yield a reference or combination of references that would make the claimed combination obvious when considered as a whole.
Claims 3-5 depend from Claim 2 and are subsequently free of the prior art.
Claim 8 is free of the prior art. Kim et. al. (US Publication 20220151540 A1) teaches on an XAI system for diagnosis of mental diseases, which an XAI system determines a brain region corresponding to a diagnosed mental disease among cerebral regions of a patient ([0050]), in which the system determines an important of brain waves and weights them based on influence on determination of importance of a brain wave feature ([0129]-[0131]), but does not teach on the particulars of obtaining patient data and using the patient data as a basis for predicting a disease as the predicted condition using XAI, wherein the XAI provides the negative and positive influential features along with explanation values.
A search of publicly available prior art fails to yield a reference or combination of references that would make the claimed combination obvious when considered as a whole.
Claim 6 is free of the prior art. Zhang et. al. (US Publication 20200303049A1) teaches on generating a medical diagnostic imaging report which may receive corrections to generate a corrected report (see at least [0115], but does not teach on any of the particulars pertaining to generating a trainable prompt to steer a PLLM to generate safe reports, using negative influential features and positive influential features for pseudo patients to generate the reports, or optimizing the loss of training data for the embedding vectors.
A search of publicly available prior art fails to yield a reference or combination of references that would make the claimed combination obvious when considered as a whole.
Claim 7 depends from Claim 6 and is subsequently free of the prior art.
Claim 10 is free of the prior art. A search of publicly available prior art fails to yield a reference or combination of references that would make the claimed combination obvious when considered as a whole.
Claim 11 is free of the prior art. Wang et. al. (WO 2023082900 A1) teaches on a pre-trained model for natural language processing and teaches on using “cloze prompts” for natural language understanding tasks, but does not teach on any of the particulars of instant Claim 11 pertaining to the cloze instruction being part of a trainable prompt that corresponds to the negative influential features and positive influential features, and responsively generating a report.
A search of publicly available prior art fails to yield a reference or combination of references that would make the claimed combination obvious when considered as a whole.
Claim 12 depends from Claim 11 and is subsequently free of the prior art.
Response to Applicant’s Remarks/Arguments
Please note: When referencing page numbers of Applicant’s response, references are to page numbers as printed.
Election/Restriction
After further consideration and in light of Applicant’s arguments, the requirement for election/restriction has been withdrawn.
Conclusion
In the interest of expediting prosecution, Examiner respectfully requests that Applicant provides citations to relevant paragraphs of specification for support for amendments in future correspondence.
The following relevant prior art not cited is made of record:
US Publication 20240095463A1, teaching on natural language processing applications using LLMS
US Publication 20240378400A1, teaching on hallucination prevention for natural language insights
US Publication 20250078471A1, teaching on generating labelled training data for creating high-performing ai systems using large language models
US Publication 12619604B2, prompt tuning using one or more machine-learned models
US Publication 20200210817A1, method and system for providing explanation of prediction generated by an artificial neural network model
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ANNE-MARIE K ALDERSON whose telephone number is (571)272-3370. The examiner can normally be reached on Mon-Fri 9:00am-5:00pm EST and generally schedules interviews in the timeframe of 2:00-5:00pm 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, Fonya Long, can be reached on 571-270-5096. 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.
/ANNE-MARIE K ALDERSON/Primary Examiner, Art Unit 3682
