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 communications: Application filed on 9/27/2024.
Claims 1-20 are pending. Claims 1, 16, and 19 are independent.
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 1-20 rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step2A, Prong 1
Claim 1 recites:
identifying a portion of training data that: was used to train an inference model of the inference models, and is undesirable;
performing a similarity analysis on the portion of the training data and other portions of the training data to identify whether any of the other portions of the training data are similar to the portion of the training data;
in a first instance of the performing where the any of the other portions of the training data are similar to the portion of the training data:
performing an untraining of the inference model using the portion of the training data to obtain an updated inference model, and
performing a compliance process for the updated inference model using the any of the other portions of the training data to obtain a compliant inference model that reflects relationships defined by the any of the other portions of the training data; and
in a second instance of the performing where the any of the other portions of the training data do not have similar embeddings to the portion of the training data:
performing the untraining of the inference model using the portion of the training data to obtain the compliant inference model.
The broadest reasonable interpretation of the bolded limitations above are directed to a mental process able to be performed in the human mind or by a human using pen and paper. A human can identify undesirable training data, compare it to similar training data.
Step 2A Prong 1 (yes).
Step2A, Prong 2
The additional elements in this claim are “performing an untraining”, and “performing a compliance process”. These elements are recited at a high level of generality and thus are generic computer components performing computer functions. Thus these are mere instructions to apply the exception using a generic computer component. See MPEP 2106.05(f).
Step2A, Prong 2 (No).
Step 2B
As explained with respect to Step 2A, the only additional element is “performing an untraining”, and “performing a compliance process” which at best is mere instructions to apply the abstract ideas and cannot provide an inventive concept, even when considered in combination. See MPEP 2106.05(f).
Step 2B (No).
Claim 1 is ineligible.
With respect to claims 16 and 19,
These claims are similar in scope to claim 1 and are rejected under a similar rationale. The processors and memory recited in these claims are also generic computing components.
Claims 16 and 19 are ineligible.
Dependent claims:
Claims 2-15, 17-18 and 20 theses claims recite further abstract ideas (mental process) and thus are ineligible.
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) 1-8, 10, and 12-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Cao et al. (“Efficient Repair of Polluted Machine Learning Systems via Causal Unlearning”) in view of Gupta et al. (US2025/0307539).
In regards to claim 1, Cao et al. discloses a method for providing computer-implemented services using inference models, the method comprising:
identifying a portion of training data that: was used to train an inference model of the inference models, and is undesirable (Cao et al. pg737 section 4.1 para1, Soracle is a small data set mostly coming from misclassification reported by users of the model);
performing a similarity analysis on the portion of the training data and other portions of the training data to identify whether any of the other portions of the training data are similar to the portion of the training data (Cao et al. pg738 section 4.2.1 para2, Karma prioritizes the search of similar data samples in the training set based on the extracted centers of misclassified data clusters);
performing a compliance process for the updated inference model using the any of the other portions of the training data to obtain a compliant inference model that reflects relationships defined by the any of the other portions of the training data (Cao et al. pg740 section 4.3 para2, the algorithm calculates the current accuracy delta, ensures that each cluster does cause the accuracy to increase, and then unlearns the cluster); and
in a second instance of the performing where the any of the other portions of the training data do not have similar embeddings to the portion of the training data:
performing the untraining of the inference model using the portion of the training data to obtain the compliant inference model (Cao et al. pg740 section 4.3 para2, the algorithm calculates the current accuracy delta, ensures that each cluster does cause the accuracy to increase, and then unlearns the cluster).
Cao et al. does not explicitly disclose in a first instance of the performing where the any of the other portions of the training data are similar to the portion of the training data:
performing an untraining of the inference model using the portion of the training data to obtain an updated inference model.
However Gupta et al. substantially discloses in a first instance of the performing where the any of the other portions of the training data are similar to the portion of the training data:
performing an untraining of the inference model using the portion of the training data to obtain an updated inference model (Gupta et al. para[0090], performs unlearning on natural language inference models).
It would have been obvious to one of ordinary skill in the art before the filing date of the invention to have combined the unlearning method of Cao et al. with the unlearning method of Gupta et al. in order to remove specific information without sacrificing performance (Gupta et al. para[0013]).
In regards to claim 2, Gupta et al. as modifed by Cau et al. disclose the method of claim 1, further comprising:
providing computer-implemented services using the compliant inference model (Gupta et al. para[0055], the system maintaining the LM (LLM-based natural language processing system 200 in the example of FIG. 2A) may incorporate the system for unlearning data from language models 100).
It would have been obvious to one of ordinary skill in the art before the filing date of the invention to have combined the unlearning method of Cao et al. with the unlearning method of Gupta et al. in order to remove specific information without sacrificing performance (Gupta et al. para[0013]).
In regards to Claim 3 Cao et al. as modified by Gupta substantially discloses the method of claim 1, wherein the portion of the training data is undesirable due to the portion of the training data comprising at least one type of training data selected from a list of types of training data consisting of: poisoned training data comprising malicious relationships established by a malicious entity (Cao et al. pg738 section 4.2.1 para1, Karma will conduct a guided process that prioritize exploration of training samples that have a higher probability of being polluted); and proprietary training data comprising confidential relationships ascribed to an owner of the proprietary training data (Gupta et al. para[0012], copyright owner of a work may want their work removed).
It would have been obvious to one of ordinary skill in the art before the filing date of the invention to have combined the unlearning method of Cao et al. with the unlearning method of Gupta et al. in order to remove specific information without sacrificing performance (Gupta et al. para[0013]).
In regards to Claim 4, Cao et al. as modified by Gupta substantially discloses the method of claim 1, wherein performing the similarity analysis comprises:
obtaining, using at least the portion of the training data, embeddings for the portion of the training data (Gupta et al. para[0025], the attention unit produces embeddings for every token in context that contain information not about the token itself, but also weighted);
obtaining, using the inference model and the other portions of the training data, embeddings for the other portions of the training data (Gupta et al. para[0030], he first encoder takes position information and embeddings of the input data as its input);
performing, using the embeddings for the portion of the training data and the embeddings for the other portions of the training data, a comparison process to obtain similarity measures between the embeddings for the portion of the training data and the embeddings for the other portions of the training data (Cao et al. pg378 section 4.2.1 para3, similarity in Karma is measured by a definition called divergence score, i.e., one divided by the number of common features between two samples.);
making a first determination regarding whether any of the similarity measures exceed a similarity measure threshold (Cao et al. pg378 section 4.2.1 para9, performs a depth first, priority searching algorithm, and maintains a minimum divergence score dmin between ci and the searched samples); and
in an instance of the first determination in which at least one similarity measure of the similarity measures exceeds the similarity measure threshold:
concluding that the any of the other portions of the training data are similar to the portion of the training data (Cao et al. pg739 section4.2.3 para1, determines whether a causality cluster is polluted as well as the size of the cluster.).
It would have been obvious to one of ordinary skill in the art before the filing date of the invention to have combined the unlearning method of Cao et al. with the unlearning method of Gupta et al. in order to remove specific information without sacrificing performance (Gupta et al. para[0013]).
In regards to Claim 5, Cao et al. as modified by Gupta substantially discloses the method of claim 4, wherein performing the compliance process comprises:
obtaining, based on the any of the other portions of the training data that are similar to the portion of the training data, a set of prompts intended to elicit responses that have a same information content as an information content of the any of the other portions of the training data that have the similar embeddings (Gupta et al. para[0062], The preliminary action plan generation component 240 may be used to determine relevant context 118 (e.g., context data 48) relevant to the user input query by calling one or more APIs as described below. In addition, the preliminary action plan generation component 240 may retrieve and/or generate the irrelevant context 114 (i.e., adversarial data) that may be included in a separate prompt relative to the prompt including the relevant context and the prompt used by the LLM 110 to respond to the input query);
performing, using the set of prompts, a first testing process to determine whether the updated inference model provides consistent and accurate responses to the set of prompts (Gupta et al. para[0080], Processing may continue at action 312, at which an updated first LM may be generated by minimizing a first prediction difference between a prediction of the first LM and a prediction of the second auxiliary LM for the text and by maximizing a second prediction difference between the prediction of the first LM and a prediction of the first auxiliary LM for the text); and
in a first instance of the performing the first testing process in which the updated inference model does not provide the consistent and accurate responses:
performing, using the any of the other portions of the training data, a re-training process for the updated inference model to obtain a re-trained updated inference model (Gupta et al. para[0080], The training process may be iterative and may iterate over the subset F until model convergence. FIG. 3 reflects the iterative nature of the training process by showing an arrow from action 312 to action 310);
performing, using the set of prompts, a second testing process to determine whether the re-trained updated inference model provides the consistent and accurate responses (Gupta et al. fig. 3para[0080], The training process may be iterative and may iterate over the subset F until model convergence. FIG. 3 reflects the iterative nature of the training process by showing an arrow from action 312 to action 310); and
in a first instance of the performing the second testing process in which the re-trained updated inference model provides the consistent and accurate responses:
concluding that the re-trained updated inference model is the compliant inference model (Cao et al. pg740 section 4.3 para2, the algorithm calculates the current accuracy delta, ensures that each cluster does cause the accuracy to increase, and then unlearns the cluster).
It would have been obvious to one of ordinary skill in the art before the filing date of the invention to have combined the unlearning method of Cao et al. with the unlearning method of Gupta et al. in order to remove specific information without sacrificing performance (Gupta et al. para[0013]).
In regards to Claim 6, Cao et al. as modified by Gupta substantially discloses the method of claim 5, further comprising: in a second instance of the performing the first testing process in which the updated inference model provides the consistent and accurate responses: concluding that the updated inference model is the compliant inference model (Cao et al. pg740 section 4.3 para2, the algorithm calculates the current accuracy delta, ensures that each cluster does cause the accuracy to increase, and then unlearns the cluster).
In regards to Claim 7, Cao et al. as modified by Gupta substantially discloses the method of claim 5, further comprising: in a second instance of the performing the second testing process in which the re-trained updated inference model does not provide the consistent and accurate responses: performing additional training and/or untraining for the re-trained updated inference model to increase a likelihood that a further updated inference model provides the consistent and accurate responses to the set of prompts (Gupta et al. para[0080], Processing may continue at action 308, at which a second auxiliary LM may be trained using a second training corpus D/F. As described herein, the corpus D/F may be the training corpus D (or a distillation/partition thereof) less the forget set F.).
It would have been obvious to one of ordinary skill in the art before the filing date of the invention to have combined the unlearning method of Cao et al. with the unlearning method of Gupta et al. in order to remove specific information without sacrificing performance (Gupta et al. para[0013]).
In regards to Claim 8, Cao et al. as modified by Gupta substantially discloses the method of claim 5, wherein performing the first testing process comprises:
obtaining, using the set of prompts, a first set of responses from the updated inference model, the first set of responses comprising:
a first response of the first set of responses to a first prompt of the set of prompts (Gupta et al. para[0080], a prediction of the first LM ); and
a second response of the first set of responses to a second prompt of the set of prompts (Gupta et al. para[0080], a prediction of the second auxiliary LM);
performing a first response agreement testing process to obtain a first level of agreement between at least the first response of the first set of responses and the second response of the first set of responses (Gupta et al. para[0080], minimizing prediction difference);
making a second determination regarding whether the first level of agreement meets agreement criteria (Gupta et al. para[0080], the training process may be iterative);
in a first instance of the second determination in which the first level of agreement meets the agreement criteria:
concluding that the updated inference model provides consistent responses to the set of prompts (Gupta et al. para[0080], iterate over the subset F until model convergence); and
in a second instance of the second determination in which the first level of agreement does not meet the agreement criteria:
concluding that the updated inference model does not provide the consistent and accurate responses to the set of prompts (Gupta et al. para[0080], the gradient may be calculated from loss defined by Equation (1) and may be back-propagated to the first LM to update the parameters).
In regards to Claim 10, Cao et al. as modified by Gupta substantially discloses the method of claim 5, wherein performing the second testing process comprises:
obtaining, using the set of prompts, a second set of responses from the re-trained updated inference model, the second set of responses comprising:
a first response of the second set of responses to a first prompt of the set of prompts (Gupta et al. para[0080], a prediction of the first LM ); and
a second response of the second set of responses to a second prompt of the set of prompts (Gupta et al. para[0080], a prediction of the second auxiliary LM);
performing a second response agreement testing process to obtain a second level of agreement between at least the first response of the second set of responses and the second response of the second set of responses (Gupta et al. para[0080], minimizing prediction difference);
making a fourth determination regarding whether the second level of agreement meets agreement criteria (Gupta et al. para[0080], the training process may be iterative);
in a first instance of the fourth determination in which the second level of agreement meets the agreement criteria:
concluding that the re-trained updated inference model provides consistent responses to the set of prompts (Gupta et al. para[0080], iterate over the subset F until model convergence); and
in a second instance of the fourth determination in which the second level of agreement does not meet the agreement criteria:
concluding that the re-trained updated inference model does not provide the consistent and accurate responses to the set of prompts (Gupta et al. para[0080], the gradient may be calculated from loss defined by Equation (1) and may be back-propagated to the first LM to update the parameters).
It would have been obvious to one of ordinary skill in the art before the filing date of the invention to have combined the unlearning method of Cao et al. with the unlearning method of Gupta et al. in order to remove specific information without sacrificing performance (Gupta et al. para[0013]).
In regards to Claim 12, Cao et al. as modified by Gupta substantially discloses the method of claim 5, wherein providing the consistent and accurate responses to the set of prompts indicates that a knowledge base of the re-trained updated inference model has an information content of the any of the other portions of the training data (Gupta et al. fig.3 para[0080], The training process may be iterative and may iterate over the subset F until model convergence. FIG. 3 reflects the iterative nature of the training process by showing an arrow from action 312 to action 310.).
It would have been obvious to one of ordinary skill in the art before the filing date of the invention to have combined the unlearning method of Cao et al. with the unlearning method of Gupta et al. in order to remove specific information without sacrificing performance (Gupta et al. para[0013]).
In regards to Claim 13, Cao et al. as modified by Gupta substantially discloses the method of claim 1, wherein performing the untraining of the inference model comprises:
modifying weights of an architecture of the inference model until responses generated by the inference model are not based on an information content of the portion of the training data (Gupta et al. para[0025], The attention unit produces embeddings for every token in context that contain information not only about the token itself, but also a weighted combination of other relevant tokens weighted by the attention weights.).
It would have been obvious to one of ordinary skill in the art before the filing date of the invention to have combined the unlearning method of Cao et al. with the unlearning method of Gupta et al. in order to remove specific information without sacrificing performance (Gupta et al. para[0013]).
In regards to Claim 14, Cao et al. as modified by Gupta substantially discloses the method of claim 1, wherein the inference model is a generative artificial intelligence (AI) model (Gupta et al. para[0010], An LM is an artificial intelligence model that may be capable of processing and generating human-like text based on latent information).
It would have been obvious to one of ordinary skill in the art before the filing date of the invention to have combined the unlearning method of Cao et al. with the unlearning method of Gupta et al. in order to remove specific information without sacrificing performance (Gupta et al. para[0013]).
In regards to Claim 15, Cao et al. as modified by Gupta substantially discloses the method of claim 1, wherein the similarity analysis is an embeddings based similarity analysis or an information content based similarity analysis (Gupta et al. para[0026], For each token i, the input embedding x.sub.i is multiplied with each of the three weight matrices to produce a query vector q.sub.i=x.sub.i W.sub.Q, a key vector k.sub.i=x.sub.i W.sub.K, and a value vector v.sub.i=x.sub.i W.sub.V. ).
It would have been obvious to one of ordinary skill in the art before the filing date of the invention to have combined the unlearning method of Cao et al. with the unlearning method of Gupta et al. in order to remove specific information without sacrificing performance (Gupta et al. para[0013]).
Claims 16-18 recite substantially similar limitations to claims 1-3. Thus claims 16-18 are rejected along the same rationale as claims 1-3.
Claims 19-20 recite substantially similar limitations to claims 1-2. Thus claims 19-20 are rejected along the same rationale as claims 1-2.
Claim(s) 9 and 11 is/are rejected under 35 U.S.C. 103 as being unpatentable over Cao et al. in view of Gupta et al. and Yatabolu et al. (US2024/0273184).
In regards to Claim 9, Cao et al. as modified by Gupta substantially discloses the method of claim 8. Cao et al. does not explicitly disclose wherein performing the first testing process further comprises:
in the first instance of the second determination in which the first level of agreement meets the agreement criteria:
comparing a first information content of the consistent responses to a second information content of the any of the other portions of the training data to obtain a first level of similarity between the first information content and the second information content;
making a third determination regarding whether the first level of similarity meets a level of similarity threshold;
in a first instance of the third determination in which the first level of similarity meets the level of similarity threshold:
concluding that the updated inference model provides the consistent and accurate responses to the set of prompts; and
in a second instance of the third determination in which the first level of similarity does not meet the level of similarity threshold:
concluding that the updated inference model does not provide the consistent and accurate responses to the set of prompts.
However Yarabolu et al. discloses wherein performing the first testing process further comprises:
in the first instance of the second determination in which the first level of agreement meets the agreement criteria:
comparing a first information content of the consistent responses to a second information content of the any of the other portions of the training data to obtain a first level of similarity between the first information content and the second information content (Yarabolu et al. para[0041], he data poisoning and model drift prevention computing system 104 identifies that clustering characteristics have changed, the data poisoning and model drift prevention computing system 104 may perform boundary analysis to identify whether the changes are due to data poisoning or have occurred due to changes to the underlying data set);
making a third determination regarding whether the first level of similarity meets a level of similarity threshold (Yarabolu et al. fig. 4 para[0044], determines if the changes between the models still meet the acceptance criteria);
in a first instance of the third determination in which the first level of similarity meets the level of similarity threshold:
concluding that the updated inference model provides the consistent and accurate responses to the set of prompts (Yarabolu et al. fig. 4 para[0042], determines the model changes are within expected or pre-configured boundary conditions, the data poisoning and model drift prevention computing system 104 may certify the data at 432); and
in a second instance of the third determination in which the first level of similarity does not meet the level of similarity threshold:
concluding that the updated inference model does not provide the consistent and accurate responses to the set of prompts (Yarabolu et al. fig. 4 para[0041], the data poisoning and model drift prevention computing system 104 may flag a data poisoning condition at 430).
It would have been obvious to one of ordinary skill in the art before the filing date of the invention to have combined the unlearning method of Cao et al. with the drift prevention system of Yarabolu et al. in order to identify and prevent malicious actors from influencing models (Yarabolu et al. para[0003]).
In regards to Claim 11, Cao et al. as modified by Gupta substantially discloses the method of claim 10. Cao et al. does not explicitly disclose wherein performing the second testing process further comprises:
in the first instance of the fourth determination in which the second level of agreement meets the agreement criteria:
comparing a third information content of the consistent responses provided by the re-trained updated inference model to a second information content of the any of the other portions of the training data to obtain a second level of similarity between the third information content and the second information content;
making a fifth determination regarding whether the second level of similarity meets the level of similarity threshold;
in a first instance of the fifth determination in which the second level of similarity meets the level of similarity threshold:
concluding that the re-trained updated inference model provides the consistent and accurate responses to the set of prompts; and
in a second instance of the fifth determination in which the second level of similarity does not meet the level of similarity threshold:
concluding that the re-trained updated inference model does not provide the consistent and accurate responses to the set of prompts.
However Yarabolu et al. discloses wherein performing the second testing process further comprises:
in the first instance of the fourth determination in which the second level of agreement meets the agreement criteria:
comparing a third information content of the consistent responses provided by the re-trained updated inference model to a second information content of the any of the other portions of the training data to obtain a second level of similarity between the third information content and the second information content (Yarabolu et al. para[0041], he data poisoning and model drift prevention computing system 104 identifies that clustering characteristics have changed, the data poisoning and model drift prevention computing system 104 may perform boundary analysis to identify whether the changes are due to data poisoning or have occurred due to changes to the underlying data set);
making a fifth determination regarding whether the second level of similarity meets the level of similarity threshold (Yarabolu et al. fig. 4 para[0044], determines if the changes between the models still meet the acceptance criteria);
in a first instance of the fifth determination in which the second level of similarity meets the level of similarity threshold:
concluding that the re-trained updated inference model provides the consistent and accurate responses to the set of prompts (Yarabolu et al. fig. 4 para[0042], determines the model changes are within expected or pre-configured boundary conditions, the data poisoning and model drift prevention computing system 104 may certify the data at 432); and
in a second instance of the fifth determination in which the second level of similarity does not meet the level of similarity threshold:
concluding that the re-trained updated inference model does not provide the consistent and accurate responses to the set of prompts (Yarabolu et al. fig. 4 para[0041], the data poisoning and model drift prevention computing system 104 may flag a data poisoning condition at 430).
It would have been obvious to one of ordinary skill in the art before the filing date of the invention to have combined the unlearning method of Cao et al. with the drift prevention system of Yarabolu et al. in order to identify and prevent malicious actors from influencing models (Yarabolu et al. para[0003]).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Mathews et al. (US2024/0195826) teaches detecting adversarial attacks on federated learning systems.
Sun et al. (US2024/0070525) teaches deleting user data from a recommendation model.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to NICHOLAS HASTY whose telephone number is (571)270-7775. The examiner can normally be reached Monday-Friday 8:30am-5:00pm.
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, Matt Ell can be reached at (571)270-3264. 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.
/N.H/Examiner, Art Unit 2141
/MATTHEW ELL/Supervisory Patent Examiner, Art Unit 2141