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 . Claims 1-20 are presented in the case.
Information Disclosure Statement
The information disclosure statement submitted on 01/04/2024 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statements is being considered by the examiner.
Claim Objections
Claims 1, 7, 18-19 are objected to because of the following informalities:
Claim 1, line 1 recites the phrase “improved data efficiency” which should be “”. This appears to be part of the intended use, is a relative term and has no threshold or utility recited.
Claim 7, line 5-6 recites the phrase “one or more embedded natural language generation exemplars” which may be referring to claim 1 exemplars or are new exemplars?
Claim 18, line 1 recites the phrase “computer-implemented method of claim 1” which should be “computing system of claim 12”
Claim 19, line 1 recites the phrase “computer-implemented method of claim 18” which should be “computing system of claim 18”
For the informalities above and wherever else they may occur appropriate correction is required.
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 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 of this title, 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 1-4, 6-7, 12, 16-17 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Chen et al. (US 20240249077 A1) hereinafter Chen in view of Potts et al. (US 20200293712 A1) hereinafter Potts.
As to independent claim 1, Chen teaches a computer-implemented method for training machine-learned models for domain alignment with improved data efficiency, the method comprising: [trains models for different tasks (contexts) ¶18, ¶20]
receiving, by a computing system, an input query associated with a particular task domain; [receives input prompt for an NLP task ¶19-20 "a sequence of soft tokens; and (ii) generates, via the small LM, a task output based on an input concatenating the soft prompts as augmented context and the input query"], [computing device ¶38]
obtaining, by the computing system, a machine-learned prompt component and a curated prompt component, [machine-learned prompt (tunable global prompt Fig. 2 107) and curated prompt (soft prompt Fig. 2 103) ¶70 ¶19 "concatenating a tunable global prompt (e.g., 107 in FIG. 2), the one or more soft prompts (e.g., 103a-n in FIG. 2)"]
wherein the machine-learned prompt component comprises a plurality of machine-learned prompt values, and [global soft prompt derived from prompt values and demonstrations (Fig. 2 102-103) via translation model ¶19, ¶30 " encodes, by a prompt translation model, demonstration examples to fixed-length soft prompts—a sequence of soft tokens"]
generating, by the computing system, an output responsive to the input query by processing a combined prompt and the input query using a pre-trained machine-learned model, wherein the combined prompt comprises the machine-learned prompt component and the curated prompt component; and [concatenates (combines prompts) for LM (model) and output ¶26-27 "each augmented context 112, 114, 116 may then be concatenated with an actual query input of the respective new task (e.g., corresponding to 102, 104, 106, respectively) and fed as input to the backbone LM 120 for the final prediction 122, 124, 126"]
updating, by the computing system and based on the generated output, the machine-learned prompt component. [output is compared and used for backpropagation (updates) ¶20 "The predicted output is then compared with the demonstration output to compute a training loss. The in-context learning framework is then trained based on the training loss via backpropagation: only the prompt translation model is updated while the small LM is kept frozen during backpropagation."]
Chen does not specifically teach wherein the curated prompt component comprises a plurality of exemplar prompt values corresponding to one or more embedded natural language exemplars for the particular task domain from domain experts.
However, Potts teaches wherein the curated prompt component comprises a plurality of exemplar prompt values corresponding to one or more embedded natural language exemplars for the particular task domain from domain experts; [domain expert annotations (exemplar prompt values) for a task like healthcare ¶7, 11 "a group of human “manual” annotators explore a significant sample of unannotated documents (relevant to the domain of interest/use-case) in free-form ways and collaborate to make numerous complex decisions toward providing an annotated dataset for training NLP models"]
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filling date of the claimed invention to modify the model training disclosed by Chan by incorporating the wherein the curated prompt component comprises a plurality of exemplar prompt values corresponding to one or more embedded natural language exemplars for the particular task domain from domain experts disclosed by Potts because both techniques address the same field of machine learning and by incorporating Potts into Chen better ensures models are trained based on accurate data and therefore perform more effectively and efficiently [Potts ¶5].
As to dependent claim 2, the rejection of claim 1 is incorporated, Chen and Potts further teach wherein parameters of one or more layers of the machine-learned model are not updated based on the generated output. [Chen backbone small model is frozen (has layers not updated) ¶20, ¶36 "The in-context learning framework may then be updated based on the training loss 420 via the backpropagation path 430, while tθ′{θ, p.sub.g} is updated but the LM 120 is kept frozen along the backpropagation path 430"]
As to dependent claim 3, the rejection of claim 2 is incorporated, Chen and Potts further teach wherein the parameters were updated by fine-tuning over a set of natural language fine-tuning instructions. [Chen instruction tuned model ¶91 " instruction-tuned models T0++(11B) (Sanh et al., Multitask prompted training enables zero-shot task generalization, in proceedings of International Conference on Learning and Representation (ICLR), 2022) and FLAN-T5-xxl (11B) "]
As to dependent claim 4, the rejection of claim 3 is incorporated, Chen and Potts further teach wherein the particular task domain is a clinical task domain, and wherein the set of natural-language fine-tuning instructions was not specific to the clinical task domain. [Potts health care domain ¶11]
As to dependent claim 6, the rejection of claim 1 is incorporated, Chen and Potts further teach wherein the curated prompt component is task domain specific, and wherein the machine-learned prompt component is shared across multiple task domains. [Chen demonstrations corresponding to tasks with prompts for particular tasks and global across variety of tasks ¶19-20]
As to dependent claim 7, the rejection of claim 6 is incorporated, Chen and Potts further teach receiving, by a computing system, a different input query associated with a different task domain; [Chen unseen task input ¶37]
obtaining, by the computing system, a different curated prompt component, wherein the different curated prompt component comprises one or more embedded natural language generation exemplars for the different task domain; and [Chen outputs result context accordingly ¶37]
generating a different output responsive to the different input query by processing the machine-learned prompt component, the different curated prompt component, and the different input query using the machine-learned model. [Chen outputs final prediction accordingly ¶37]
As to independent claim 12, Chen teaches a computing system comprising: [computing device ¶38]
one or more processors; and [processor ¶38]
one or more non-transitory computer-readable media storing instructions that are executable by the one or more processors to cause the computing system to perform operations, the operations comprising: [media with code ran on device ¶41]
receiving an input query associated with a particular task domain [receives input prompt for an NLP task ¶19-20 "a sequence of soft tokens; and (ii) generates, via the small LM, a task output based on an input concatenating the soft prompts as augmented context and the input query"]
obtaining a machine-learned prompt component and a curated prompt component, [machine-learned prompt (tunable global prompt Fig. 2 107) and curated prompt (soft prompt Fig. 2 103) ¶70 ¶19 "concatenating a tunable global prompt (e.g., 107 in FIG. 2), the one or more soft prompts (e.g., 103a-n in FIG. 2)"]
wherein the machine-learned prompt component comprises a plurality of machine-learned prompt values, and [global soft prompt derived from prompt values and demonstrations (Fig. 2 102-103) via translation model ¶19, ¶30 " encodes, by a prompt translation model, demonstration examples to fixed-length soft prompts—a sequence of soft tokens"]
generating an output responsive to the input query by processing a combined prompt and the input query using a pre-trained machine-learned model, wherein the combined prompt comprises the machine-learned prompt component and the curated prompt component; and [concatenates (combines prompts) for LM (model) and output ¶26-27 "each augmented context 112, 114, 116 may then be concatenated with an actual query input of the respective new task (e.g., corresponding to 102, 104, 106, respectively) and fed as input to the backbone LM 120 for the final prediction 122, 124, 126"]
updating, by the computing system and based on the generated output, the machine-learned prompt component. [output is compared and used for backpropagation (updates) ¶20 "The predicted output is then compared with the demonstration output to compute a training loss. The in-context learning framework is then trained based on the training loss via backpropagation: only the prompt translation model is updated while the small LM is kept frozen during backpropagation."]
Chen does not specifically teach wherein the curated prompt component comprises a plurality of exemplar prompt values corresponding to one or more embedded natural language exemplars for the particular task domain from domain experts and a plurality of clinical task domains;
However, Potts teaches wherein the curated prompt component comprises a plurality of exemplar prompt values corresponding to one or more embedded natural language exemplars for the particular task domain from domain experts; [domain expert annotations (exemplar prompt values) for a task like healthcare ¶7, 11 "a group of human “manual” annotators explore a significant sample of unannotated documents (relevant to the domain of interest/use-case) in free-form ways and collaborate to make numerous complex decisions toward providing an annotated dataset for training NLP models"]
and a plurality of clinical task domains; [health care domain ¶11]
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filling date of the claimed invention to modify the model training disclosed by Chan by incorporating the wherein the curated prompt component comprises a plurality of exemplar prompt values corresponding to one or more embedded natural language exemplars for the particular task domain from domain experts and a plurality of clinical task domains disclosed by Potts because both techniques address the same field of machine learning and by incorporating Potts into Chen better ensures models are trained based on accurate data and therefore perform more effectively and efficiently [Potts ¶5].
As to dependent claim 16, the rejection of claim 12 is incorporated, Chen and Potts further teach wherein the curated prompt component is task domain specific, and wherein the machine-learned prompt component is shared across multiple task domains. [Chen demonstrations corresponding to tasks with prompts for particular tasks and global across variety of tasks ¶19-20]
As to dependent claim 17, the rejection of claim 16 is incorporated, Chen and Potts further teach wherein the operations comprise:
receiving a different input query associated with a different task domain; [Chen unseen task input ¶37]
obtaining a different curated prompt component, wherein the different curated prompt component comprises one or more exemplars for the different task domain; and [Chen outputs result context accordingly ¶37]
generating a different output responsive to the different input query by processing the machine-learned prompt component, the different curated prompt component, and the different input query using the machine-learned model. [Chen outputs final prediction accordingly ¶37]
As to independent claim 20, Chen teaches One or more non-transitory computer-readable media storing instructions that are executable by one or more processors to cause a computing system to perform operations, the operations comprising: [media with code ran on device ¶41]
receiving an input query associated with a particular task domain of a plurality of available task domains; [receives input prompt for an NLP task ¶19-20 "a sequence of soft tokens; and (ii) generates, via the small LM, a task output based on an input concatenating the soft prompts as augmented context and the input query"]
obtaining a machine-learned prompt component and a curated prompt component, [machine-learned prompt (tunable global prompt Fig. 2 107) and curated prompt (soft prompt Fig. 2 103) ¶70 ¶19 "concatenating a tunable global prompt (e.g., 107 in FIG. 2), the one or more soft prompts (e.g., 103a-n in FIG. 2)"]
wherein the machine-learned prompt component comprises a plurality of machine-learned prompt values, and [global soft prompt derived from prompt values and demonstrations (Fig. 2 102-103) via translation model ¶19, ¶30 " encodes, by a prompt translation model, demonstration examples to fixed-length soft prompts—a sequence of soft tokens"]
generating an output responsive to the input query by processing a combined prompt and the input query using a pre-trained machine-learned model, wherein the combined prompt comprises the machine-learned prompt component and the curated prompt component; and [concatenates (combines prompts) for LM (model) and output ¶26-27 "each augmented context 112, 114, 116 may then be concatenated with an actual query input of the respective new task (e.g., corresponding to 102, 104, 106, respectively) and fed as input to the backbone LM 120 for the final prediction 122, 124, 126"]
updating, by the computing system and based on the generated output, the machine-learned prompt component. [output is compared and used for backpropagation (updates) ¶20 "The predicted output is then compared with the demonstration output to compute a training loss. The in-context learning framework is then trained based on the training loss via backpropagation: only the prompt translation model is updated while the small LM is kept frozen during backpropagation."]
Chen does not specifically teach wherein the curated prompt component comprises a plurality of exemplar prompt values corresponding to one or more embedded natural language exemplars for the particular task domain from domain experts;
However, Potts teaches wherein the curated prompt component comprises a plurality of exemplar prompt values corresponding to one or more embedded natural language exemplars for the particular task domain from domain experts; [domain expert annotations (exemplar prompt values) for a task like healthcare ¶7, 11 "a group of human “manual” annotators explore a significant sample of unannotated documents (relevant to the domain of interest/use-case) in free-form ways and collaborate to make numerous complex decisions toward providing an annotated dataset for training NLP models"]
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filling date of the claimed invention to modify the model training disclosed by Chan by incorporating the wherein the curated prompt component comprises a plurality of exemplar prompt values corresponding to one or more embedded natural language exemplars for the particular task domain from domain experts disclosed by Potts because both techniques address the same field of machine learning and by incorporating Potts into Chen better ensures models are trained based on accurate data and therefore perform more effectively and efficiently [Potts ¶5].
Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over Chen in view of Potts as applied in the rejection of claim 1 above, and further in view of Barbetta et al. (US 20150161106 A1) hereinafter Barbetta.
As to dependent claim 5, Chen and Potts teach the method of claim 1 above that is incorporated,
Chen and Potts do not specifically teach obtaining, by the computing system, one or more expert responses to one or more example queries in the particular task domain; and generating, by the computing system, the curated prompt component based on the one or more expert responses to the one or more example queries.
However, Barbetta teaches obtaining, by the computing system, one or more expert responses to one or more example queries in the particular task domain; and [domain expert provided question answer sets/pairs (examples) ¶6-10]
generating, by the computing system, the curated prompt component based on the one or more expert responses to the one or more example queries. [trains Q/A system with sets and provides input recommendations (curated prompts) ¶35 "The system describes problems in the input text or shortcomings in the ability of the QA system to analyze it. Further, the system recommends actions a domain expert might take to address the identified issues"]
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filling date of the claimed invention to modify the model interface disclosed by Chen and Potts by incorporating the obtaining, by the computing system, one or more expert responses to one or more example queries in the particular task domain; and generating, by the computing system, the curated prompt component based on the one or more expert responses to the one or more example queries disclosed by Barbetta because all techniques address the same field of machine learning and by incorporating Barbetta into Chen and Potts improves accuracy in testing sets and models for better predictions [Barbetta ¶3].
Claims 8, 10-11 and 18-19 are rejected under 35 U.S.C. 103 as being unpatentable over Chen in view of Potts as applied in the rejection of claim 1 above, and further in view of Simons et al. (US 20220399086 A1) hereinafter Simons.
As to dependent claim 8, Chen and Potts teach the method of claim 1 above that is incorporated,
Chen and Potts do not specifically teach wherein the machine-learned model is configured to interact with one or more clinical software tools to obtain the output.
However, Simons teaches wherein the machine-learned model is configured to interact with one or more clinical software tools to obtain the output. [API to interact with clinical software ¶73 "FHIR instance application programming interface (“API”) 530 to FHIR instance generator logic 532 which is configured to extract one or more instance values/definitions from FHIR instance models to generate individual FHIR resource objects"]
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filling date of the claimed invention to modify the model building disclosed by Chen and Potts by incorporating the wherein the machine-learned model is configured to interact with one or more clinical software tools to obtain the output disclosed by Simons because all techniques address the same field of machine learning and by incorporating Simons into Chen and Potts takes in to account possible errors and inefficiencies in medical information [Simons ¶22].
As to dependent claim 10, the rejection of claim 8 is incorporated, Chen, Potts and Simons further teach wherein the one or more clinical software tools comprise at least one tool selected from the following list: an electronic health record database, a data acquisition interface, medical image-processing software, patient communication software, biochemical simulation software, or insurance billing software. [Simons clinical data sources FHIR, CCDA (records, data, communication) ¶21]
As to dependent claim 11, the rejection of claim 8 is incorporated, Chen, Potts and Simons further teach wherein the machine-learned model is configured to interact with the one or more clinical software tools to obtain the output using an API call to the one or more clinical software tools. [Simons API used to access records ¶73]
As to dependent claim 18, Chen and Potts teach the method of claim 1 above that is incorporated,
Chen and Potts do not specifically teach wherein the machine-learned model is configured to interact with one or more clinical software tools to obtain the output.
However, Simons teaches wherein the machine-learned model is configured to interact with one or more clinical software tools to obtain the output. [API to interact with clinical software ¶73 "FHIR instance application programming interface (“API”) 530 to FHIR instance generator logic 532 which is configured to extract one or more instance values/definitions from FHIR instance models to generate individual FHIR resource objects"]
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filling date of the claimed invention to modify the model building disclosed by Chen and Potts by incorporating the wherein the machine-learned model is configured to interact with one or more clinical software tools to obtain the output disclosed by Simons because all techniques address the same field of machine learning and by incorporating Simons into Chen and Potts takes in to account possible errors and inefficiencies in medical information [Simons ¶22].
As to dependent claim 19, the rejection of claim 18 is incorporated, Chen, Potts and Simons further teach wherein the one or more clinical software tools comprise at least one tool selected from the following list: an electronic health record database, a data acquisition interface, medical image-processing software, patient communication software, biochemical simulation software, or insurance billing software. [Simons clinical data sources FHIR, CCDA (records, data, communication) ¶21]
Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over Chen in view of Potts and Simons, as applied in the rejection of claim 1 above, and further in view of Mishchenko et al. (US 11922144 B1) hereinafter Mishchenko.
As to dependent claim 9, Chen, Potts and Simons teach the method of claim 8 above that is incorporated,
Chen, Potts and Simons do not specifically teach wherein the combined prompt comprises tokens indicating available clinical software tools.
However, Mishchenko teaches wherein the combined prompt comprises tokens indicating available clinical software tools. [API manifest and prompts including API functions accordingly Col. 9-10 ln 65-19 "a natural language model may be provided, in the model's system prompt, with a description of functions specific to one or more particular APIs via the manifest."]
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filling date of the claimed invention to modify the model interface disclosed by Chen and Potts by incorporating the wherein the combined prompt comprises tokens indicating available clinical software tools disclosed by Mishchenko because all techniques address the same field of machine learning and by incorporating Mishchenko into Chen, Potts and Simons improve the trainability, scalability, and generation of natural language models with respect to external data and applications. [Mishchenko Col. 1 ln. 27-33].
Claims 13-15 are rejected under 35 U.S.C. 103 as being unpatentable over Chen in view of Potts as applied in the rejection of claim 12 above, and further in view of Mukhopadhyay et al. (US 20210056226 A1) hereinafter Mukhopadhyay.
As to dependent claim 13, Chen and Potts teach the method of claim 12 above that is incorporated,
Chen and Potts do not specifically teach receiving the input query from a client device; obtaining the machine-learned prompt component, the curated prompt component, or both, from a secure data store associated with the client device; and returning the output to the client device.
However, Mukhopadhyay teaches receiving the input query from a client device; [feature input for model ¶8]
obtaining the machine-learned prompt component, the curated prompt component, or both, from a secure data store associated with the client device; and [secure container on user device for machine learning models, features, data and training ¶86 , ¶8 "secured virtual container is enabled to securely store personal data corresponding to a user, where such data is inaccessible to processes running outside the secured virtual container"]
returning the output to the client device. [returns output ¶8, ¶27 " The output of the machine learning model may comprise the above mentioned hashed/normalized/feature engineered digest form of the user data that may be provided to the operating system and/or applications"]
Accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filling date of the claimed invention to modify the model building disclosed by Chen and Potts by incorporating the receiving the input query from a client device; obtaining the machine-learned prompt component, the curated prompt component, or both, from a secure data store associated with the client device; and returning the output to the client device disclosed by Mukhopadhyay because all techniques address the same field of machine learning and by incorporating Mukhopadhyay into Chen and Potts provide more personalized models with personal data use [Mukhopadhyay ¶4]
As to dependent claim 14, the rejection of claim 13 is incorporated, Chen, Potts and Mukhopadhyay further teach wherein the machine-learned prompt component was learned in a secure environment based on training data associated with the client device. [Mukhopadhyay features from secure user data store ¶8]
As to dependent claim 15, the rejection of claim 14 is incorporated, Chen, Potts and Mukhopadhyay further teach wherein the machine-learned prompt component was learned on the client device. [Mukhopadhyay trained and executed on user device ¶86, ¶45 " train machine learning modules residing in ML engine 208"]
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Applicant is required under 37 C.F.R. § 1.111(c) to consider these references fully when responding to this action.
Austin et al. (US 20240420012 A1) teaches curated prompts for an LLM (see ¶46).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to BEAU SPRATT whose telephone number is (571)272-9919. The examiner can normally be reached M-F 8:30-5 PST.
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, Jennifer Welch can be reached on 5712127212. 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.
/BEAU D SPRATT/ Primary Examiner, Art Unit 2143
Prosecution Insights