DETAILED ACTION
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA . This action is responsive to pending claims 1-20 filed 3/8/2024.
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.
Claim(s) 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The 35 U.S.C. 101 subject matter eligibility analysis first asks whether the claim is directed to one of the four statutory categories (Step 1). It next asks whether the claim is directed to an abstract idea (Step 2A), via Prong 1, whether an abstract idea (e.g., mathematical concept, mental process, certain methods of organizing human activity) is recited, and Prong 2, whether it is integrated into a practical application. It finally asks whether the claim as a whole includes additional elements that amount to significantly more than the judicial exception (Step 2B). See MPEP 2106.
STEP 1: The claims falls within one of the four statutory categories:
All claims are directed to methods and hardware systems and hence fall within one of the four statutory categories.
STEP 2A PRONG 1: The claims recite a judicial exception:
Claim 1 is directed to a method for managing a training corpus and models. It involves segmenting or dividing a training corpus according to a model’s permission, and executing the training. However, these steps may be performed mentally.
Additional elements are underlined in the analysis below. In particular:
For claim 1: A method for managing training corpus and one or more machine learning models, the method comprising:
receiving a model permission requirement (Receiving permissions may be performed mentally);
receiving a training corpus (Receiving for management a volume of data may be performed mentally);
segmenting the training corpus into a segmented training corpus based at least in part on the model permission requirement (Dividing data based on permission may be performed mentally);
training a machine learning model using the segmented training corpus (Drawing conclusions based on different data segments is observation and judgment and may be performed mentally); and
associating the trained machine learning model with the model permission requirement (Associating data may be performed mentally);
wherein the method is performed by one or more processors.
For claim 2: The method of claim 1, wherein the model permission requirement includes a first permission level that is higher than a second permission level (Considering and managing multiple permission levels may be performed mentally);
wherein the segmenting the training corpus to a segmented training corpus includes:
segmenting the training corpus to a first set of training corpus based at least in part on the first permission level (Dividing data based on permission may be performed mentally); and
segmenting the training corpus to a second set of training corpus based at least in part on the second permission level (Dividing data based on permission may be performed mentally); and
wherein the segmented training corpus includes the first set of training corpus and the second set of training corpus (Dividing data based on permission may be performed mentally).
For claim 3: The method of claim 2, wherein the training a machine learning model includes:
training the machine learning model using the second set of training corpus to generate a first trained machine learning model (Training models based on different data sets may be performed mentally); and
training the first trained machine learning model using the first set of training corpus to generate the trained machine learning model (Training models based on different data sets may be performed mentally).
For claim 4: The method of claim 2,
wherein the training a machine learning model includes:
training the machine learning model using the second set of training corpus to generate a first trained machine learning model (Training models based on different data sets may be performed mentally); and
training the machine learning model using the first set of training corpus to generate a second trained machine learning model (Training models based on different data sets may be performed mentally);
wherein the method further comprises:
applying the first trained machine learning model to user data to generate a first model result (Generating results via a model, e.g., a heuristic may be performed mentally);
applying the second trained machine learning model to user data to generate a second model result (Generating results via a model, e.g., a heuristic may be performed mentally);
and applying an ensemble model to the first model result and the second model result to generate an ensemble result (Combining results may be performed mentally).
For claim 5: The method of claim 2, further comprising:
receiving a process request including a usage permission level and a request to use the trained machine learning model (Receiving requests for inference based on a mental model may be performed mentally);
determining whether the usage permission level satisfies the model permission requirement (Determining based on comparison may be performed mentally);
if the usage permission level satisfies the model permission requirement,
allowing the process request to access the trained machine learning model; and
if the usage permission level does not satisfy the model permission requirement,
not allowing the process request to access the trained machine learning model (Allowing or disallowing action based on comparisons may be performed mentally).
For claim 6: The method of claim 2, further comprising:
assigning a first weight to the first set of training corpus; and
assigning a second weight to the second set of training corpus;
wherein the first weight is different from the second weight (Assigning weights may be performed mentally);
wherein the training a machine learning model includes training the machine learning model using the first set of training corpus and the second set of training corpus based at least in part on the first weight and the second weight (Forming a mental model via consideration of different values or importances of sectors of data may be performed mentally).
For claim 7: The method of claim 1, wherein the machine learning model includes a large language model.
For claim 8: The method of claim 1, wherein the machine learning model includes a part of the training corpus embedded in the machine learning model (Forming a model containing particular portions of the training data may be performed mentally).
For claim 9: The method of claim 1, wherein the model permission requirement includes a security level or an access level (Considering access or security levels may be performed mentally).
For claim 10: A method for managing generative Al models, the method comprising:
receiving a plurality of generative Al models associated with a plurality of model permission requirements (Receiving models for consideration, particular to a permission domain, may be performed mentally);
receiving a process request including a usage permission level (Receiving permission requests including level may be performed mentally);
comparing the usage permission level with each model permission requirement of the plurality of model permission requirements (Judging each model as containing or not containing adequate permission may be performed mentally);
selecting a requested generative Al model from the plurality of generative Al models based at least in part on the comparison (Selection based on criteria may be performed mentally); and
allowing the process request to access the requested generative Al model (Allowing for access may be performed mentally);
wherein the method is performed by one or more processors.
For claim 11. The method of claim 10, wherein the selecting a requested generative Al model from the plurality of generative Al models comprises:
selecting one or more generative Al models from the plurality of generative Al models, each selected generative Al model of the one or more selected generative Al models corresponding to a model permission requirement that has a same or lower permission requirement than the usage permission level (Selection based on criteria may be performed mentally).
For claim 12. The method of claim 11, wherein the selecting a requested generative AI model from the plurality of generative AI models further comprises:
selecting the requested generative AI model from the one or more selected generative AI models, the requested generative AI model having a highest permission requirement among one or more model permission requirements corresponding to the one or more selected generative AI models (selection based on permission criteria or ranking may be performed mentally).
For claim 13. The method of claim 11, wherein the selecting a requested generative AI model from the plurality of generative AI models further comprises:
selecting the requested generative AI model from the one or more selected generative AI models, the requested generative AI model having a lowest permission requirement among one or more model permission requirements corresponding to the one or more selected generative AI models (selection based on permission criteria or ranking may be performed mentally).
Claim(s) 14-20 recite systems analogous to the above methods and are hence rejected for the same reasons.
STEP 2A PRONG 2: The claims do not integrate the exception into a practical application:
For claim 1, 10, 14 the additional elements including performing via a processor, including storing instructions on computer-readable media for performance on said processor. However, this is mere instructions to implement the abstract idea on a computer and hence does not constitute an integration into a practical application.
For claim 3-4, 14, the additional elements including performing via a machine learning model. However, this is mere instructions to implement the abstract idea via a machine learning model and hence does not constitute an integration into a practical application.
For claim 7, 10-13, the additional elements including using a language model or a generative AI model. However, the use of language models and generative AI models in data processing to implement an abstract idea is well-understood, routine, and conventional in the field of data processing and hence does not constitute significantly more.
STEP 2B: The claim as a whole do not include additional elements that amount to significantly more than the abstract idea:
For claim 1, 10, 14 the additional elements including performing via a processor, including storing instructions on computer-readable media for performance on said processor. However, the use of processors in a general purpose computing environment to implement an abstract idea is well-understood, routine, and conventional in the field of data processing and hence does not constitute significantly more.
For claims 3-4, 14, the additional elements including performing via a machine learning model. However, the use of machine learning models for in a data analysis is well-understood, routine, and conventional in the field of data processing and hence does not constitute significantly more.
For claim 7, 10-13, the additional elements including performing via a language model or a generative AI model. However, the use of language models and generative AI models is well-understood, routine, and conventional in the field of data processing and hence does not constitute significantly more.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale or otherwise available to the public before the effective filing date of the claimed invention.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claim(s) 1-3, 6, 9, 14-16, 19 are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Cooper (US 20240386119 A1).
For claim 1, Cooper discloses: a method for managing training corpus and one or more machine learning models (Overview 0081-129 gives an overview of a hierarchical tenant-based technique for siloing ML training data, with exemplary figures in figs.6-19, 0203 and onward), the method comprising:
receiving a model permission requirement (fig.12:1244, 0254-261: model permission requirement is received by the system for instantiating an active machine learning model 1244 indicating permissions to data 1246, 724, 726);
receiving a training corpus (fig.12:1246, 724, 726, 728: training corpuses are received);
segmenting the training corpus into a segmented training corpus based at least in part on the model permission requirement (ibid: training corpuses are segmented based on permission and hierarchy structure into 1246, 724, 726);
training a machine learning model using the segmented training corpus (fig.12, 0254-261: ML model is allowed access to said segment, such as for training, see 0171); and
associating the trained machine learning model with the model permission requirement (fig.12:1244: the ML trained ML model is associated with the permission data access requirement);
wherein the method is performed by one or more processors (fig.22, 0330, 0383).
For claim 2, Cooper discloses the method of claim 1, as described above. Cooper further discloses: wherein the model permission requirement includes a first permission level that is higher than a second permission level (figs.6-20 contemplate various access levels for machine learning models, in particular, consider fig. fig.12, 0259 contemplating access to data 1240, 620, 606 (see checked boxes) and fig.7, 0215 disclosing access to subgroup 620, 610, hence, the first permission level being higher permission, being able to access the additional Admin model data);
wherein the segmenting the training corpus to a segmented training corpus includes:
segmenting the training corpus to a first set of training corpus based at least in part on the first permission level (fig.12, 0259); and
segmenting the training corpus to a second set of training corpus based at least in part on the second permission level (fig.7, 0215); and
wherein the segmented training corpus includes the first set of training corpus and the second set of training corpus (fig.12, fig.7).
For claim 3, Cooper discloses the method of claim 2, as described above. Cooper further discloses: wherein the training a machine learning model includes:
training the machine learning model using the second set of training corpus to generate a first trained machine learning model (fig.7:722, 0171: data training corpus data is used, such as group data, to train the machine learning model, such as via gradient descent, (0079), hence, model is turned during training to generated a trained model); and
training the first trained machine learning model using the first set of training corpus to generate the trained machine learning model (fig.12, 0259: the model is trained additionally on first corpus including group data, such as via additional data in the group data, to generate a final model, such as via gradient descent, see 0079).
For claim 6, Cooper discloses the method of claim 2, as described above. Cooper further discloses: assigning a first weight to the first set of training corpus (fig.7:620, 0217-220: assigning various weights to the group portion, e.g., those associated with the contribution of various tenants); and
assigning a second weight to the second set of training corpus (ibid: additional weights are assigned to the second set of the training corpus, such as weights indicating amount of contribution to the tenant data, see 0223-226);
wherein the first weight is different from the second weight (ibid: weights would be different based on history);
wherein the training a machine learning model includes training the machine learning model using the first set of training corpus and the second set of training corpus based at least in part on the first weight and the second weight (0220-224: using weighted dataset for various operations, including training, see 0171).
For claim 9, Cooper discloses the method of claim 1, as described above. Cooper further discloses: wherein the model permission requirement includes a security level or an access level (fig.12:1244, fig.7:722 shows security level or access level list).
Claim 14 recites a system corresponding to the method of claim 1 and is hence rejected under the same rationale. Cooper further discloses: a system for managing training corpus and one or more machine learning models (fig.22), the system comprising:
one or more memories having instructions stored thereon (fig.22, 0330); and
one or more processors configured to execute the instructions and perform operations (fig.22:0330, 0383).
Claims 15-16, 19 recite systems corresponding to the above claims and are hence likewise rejected.
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.
Claim(s) 4, 17 are rejected under 35 U.S.C. 103 as being unpatentable over Cooper (US 20240386119 A1) in view of Naibo ("Data-Free Diversity-Based Ensemble Selection For One-Shot Federated Learning in Machine Learning Model Market", published 2/23/2023).
For claim 4, Cooper discloses the method of claim 2, as described above. Cooper further discloses: wherein the training a machine learning model includes:
training the machine learning model using the second set of training corpus to generate a first trained machine learning model (fig.7, 0215); and
training the machine learning model using the first set of training corpus to generate a second trained machine learning model (fig.12, 0259);
wherein the method further comprises:
applying the first trained machine learning model to user data to generate a first model result (fig.4, 0171: generating inference);
applying the second trained machine learning model to user data to generate a second model result (fig.4, 0171: generating inference);
and applying an ensemble model to the first model result model result to generate an ensemble result (0171: ensemble models).
Cooper does not disclose: wherein the applying includes to the second model result.
Naibo discloses: wherein the applying includes to the second model result (§I ¶4: using ensemble learning, hence, ensembling results such as via voting).
It would have been obvious before the effective filing date to one of ordinary skill in the art to modify the method of Cooper by incorporating the ensembling technique of Naibo. Both concern the art of machine learning with privacy constrained environments, and the incorporation would have, according to Naibo, implement a straightforward, cost effective technique to boost machine intelligence.
Claim(s) 17 recite systems corresponding to the above claims and are hence likewise rejected.
Claim(s) 5, 18 are rejected under 35 U.S.C. 103 as being unpatentable over Cooper (US 20240386119 A1) in view of Cohen (US 20210099453 A1).
For claim 5, Cooper discloses the method of claim 2, as described above. Cooper does not disclose the limitations of claim 5: receiving a process request including a usage permission level and a request to use the trained machine learning model;
determining whether the usage permission level satisfies the model permission requirement;
if the usage permission level satisfies the model permission requirement,
allowing the process request to access the trained machine learning model; and
if the usage permission level does not satisfy the model permission requirement,
not allowing the process request to access the trained machine learning model.
Cohen discloses: receiving a process request (0064 contemplates a computing environment including segmented data (fig.1:110) as well as resources that access, process, and serve data, e.g., as software services, hence, combination with Cooper yielding application to inference models such as in Cooper fig.4) including a usage permission level and a request to use the trained machine learning model (0075-80: a permission level is associated with the user, hence, permission level is received at the Data-based Access Control System for approving access, see fig.1:120. Fig.2, 0062; hence, a user permission level or user data association is received at a request processing and determination function for approving o rejecting user requests);
determining whether the usage permission level satisfies the model permission requirement (0085: associating user level associations with resource (e.g., software processing resource such as ML models of cooper, see 0064) associations with the data segments, in order to determine access, hence, determining whether user permission level satisfies model permission requirements );
if the usage permission level satisfies the model permission requirement,
allowing the process request to access the trained machine learning model (0087-88, fig.3, 0090 verifying association with the data segment, hence, verifying correspondence and satisfaction of user level with model permission requirement level, and providing access); and
if the usage permission level does not satisfy the model permission requirement,
not allowing the process request to access the trained machine learning model (ibid: rejecting access based on non-satisfaction of correspondence).
It would have been obvious before the effective filing date to one of ordinary skill in the art to modify the method of Cooper by incorporating the data segment based user and service access control of Cohen. Both concern the art of data access control and the incorporation would have, according to Cohen, allow higher security, better auditing, and prevent errors in access of data access (0003, 0051)
Claim(s) 18 recite systems corresponding to the above claims and are hence likewise rejected.
Claim(s) 7-8, 10-13, 20 are rejected under 35 U.S.C. 103 as being unpatentable over Cooper (US 20240386119 A1) in view of rpm ("Introduction to Large Language Models and the Transformer Architecture", published 3/9/2023).
For claim 7, Cooper discloses the method of claim 1, as described above. Cooper does not disclose: wherein the machine learning model includes a large language model.
RPM discloses: wherein the machine learning model includes a large language model (p.1: Large Language Models (LLM)).
It would have been obvious before the effective filing date to one of ordinary skill in the art to modify the method of Cooper by incorporating the large language model technique of RPM. Both concern the art of machine learning, and the incorporation would have, according to RPM, increase capacity to process natural language, text summarization, fine tuning for specific tasks (p.1 § “Large Language Models (LLM)” ¶1).
For claim 8, Cooper discloses the method of claim 1, as described above. Cooper does not disclose: wherein the machine learning model includes a part of the training corpus embedded in the machine learning model.
RPM discloses: wherein the machine learning model includes a part of the training corpus embedded in the machine learning model (p.3 shows input, such as during training, being transformed into input embeddings, the embeddings being transformed via attention to an output, hence, LLM model including part of the training corpus being embedded in the machine learning model during training as inputs and embedded via backpropagation during training as adjustments to the weights of the transformer matrices).
It would have been obvious before the effective filing date to one of ordinary skill in the art to modify the method of Cooper by incorporating the large language model technique of RPM. Both concern the art of machine learning, and the incorporation would have, according to RPM, increase capacity to process natural language, text summarization, fine tuning for specific tasks (p.1 § “Large Language Models (LLM)” ¶1).
For claim 10, Cooper discloses: a method for managing Al models (Overview 0081-129 gives an overview of a hierarchical tenant-based technique for siloing ML training data, with exemplary figures in figs.6-19, 0203 and onward), the method comprising:
receiving a plurality of Al models associated with a plurality of model permission requirements (figs.7-8, etc. disclose receiving of various ML models associated with respective permission requirements);
receiving a process request (fig.4, 0171: request is initiated for instantiating ML model) including a usage permission level (fig.11, 0248-230: receiving custom permission settings, hence, a data access request, such as for model training, is received alongside usage permission level, the request leading to model initiation, such as for training (0171));
comparing the usage permission level with each model permission requirement of the plurality of model permission requirements (fig.11: based on the authorized permission level of the request parameters in 932 (606, 602), data access is granted based on set membership);
selecting a requested Al model from the plurality of generative Al models (0171: selection of model for instantiation) based at least in part on the comparison (fig.11, 0171, fig.4: the model constructed based on the comparison, is selected for training, inference, such as in fig.4); and
allowing the process request to access the requested Al model (fig.4, 0171: received model initiation, type, permission (0248-230) is used to instantiate model, hence, process request access model for processing);
wherein the method is performed by one or more processors (fig.22, 0330, 0383).
Cooper does not disclose: wherein the AI model is generative.
RPM discloses: wherein the AI model is generative. (p.1: Large Language Models (LLM)).
It would have been obvious before the effective filing date to one of ordinary skill in the art to modify the method of Cooper by incorporating the large language model technique of RPM. Both concern the art of machine learning, and the incorporation would have, according to RPM, increase capacity to process natural language, text summarization, fine tuning for specific tasks (p.1 § “Large Language Models (LLM)” ¶1).
For claim 11, Cooper discloses the method of claim 10, as described above. Cooper further discloses: wherein the selecting a requested generative Al model from the plurality of generative Al models comprises:
selecting one or more generative Al models from the plurality of generative Al models, each selected generative Al model of the one or more selected generative Al models corresponding to a model permission requirement that has a same or lower permission requirement than the usage permission level (figs.6-20 contemplate various access levels for machine learning models, in particular, consider fig. fig.12, 0259 contemplating access to data 1240, 620, 606 (see checked boxes) and fig.7, 0215 disclosing access to subgroup 620, 610, hence, the first permission level being higher permission, being able to access the additional Admin model data; alternatively).
For claim 12, Cooper discloses the method of claim 11, as described above. Cooper further discloses: wherein the selecting a requested generative AI model from the plurality of generative AI models further comprises:
selecting the requested generative AI model from the one or more selected generative AI models, the requested generative AI model having a highest permission requirement among one or more model permission requirements corresponding to the one or more selected generative AI models (fig.7, fig.12 as described above: selection of fig.12 by the system for instantiation, training, inference would constitute selecting model having highest permission request from among the models selected for instantiation, training, inference).
For claim 13, Cooper discloses the method of claim 11, as described above. Cooper further discloses: wherein the selecting a requested generative AI model from the plurality of generative AI models further comprises:
selecting the requested generative AI model from the one or more selected generative AI models, the requested generative AI model having a lowest permission requirement among one or more model permission requirements corresponding to the one or more selected generative AI models (fig.7, fig.12 as described above: selection of fig.7 by the system for instantiation, training, inference would constitute selecting model having highest permission request from among the models selected for instantiation, training, inference).
Claim(s) 20 recite systems corresponding to the above claims and are hence likewise rejected.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Tschiatschek (US 20210089819 A1) discloses a multi-tenant privacy training environment.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to LIANG LI whose telephone number is (303)297-4263. The examiner can normally be reached Mon-Fri 9-12p, 3-11p MT (11-2p, 5-1a ET).
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. The examiner is available for interviews Mon-Fri 6-11a, 2-7p MT (8-1p, 4-9p ET).
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor Jennifer Welch can be reached on (571)272-7212. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from Patent Center and the Private Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from Patent Center or Private PAIR. Status information for unpublished applications is available through Patent Center or Private PAIR to authorized users only. Should you have questions about access to Patent Center or the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
/LIANG LI/
Primary examiner AU 2143
Prosecution Insights