Prosecution Insights
Last updated: April 19, 2026
Application No. 18/434,291

SYSTEMS AND METHODS FOR TRAINING MACHINE LEARNING MODELS USING EMULATED DATASETS

Non-Final OA §102§103
Filed
Feb 06, 2024
Examiner
MEHMOOD, JENNIFER
Art Unit
2664
Tech Center
2600 — Communications
Assignee
Amazon Technologies, Inc.
OA Round
1 (Non-Final)
65%
Grant Probability
Moderate
1-2
OA Rounds
3y 1m
To Grant
95%
With Interview

Examiner Intelligence

Grants 65% of resolved cases
65%
Career Allow Rate
160 granted / 247 resolved
+2.8% vs TC avg
Strong +31% interview lift
Without
With
+30.6%
Interview Lift
resolved cases with interview
Typical timeline
3y 1m
Avg Prosecution
21 currently pending
Career history
268
Total Applications
across all art units

Statute-Specific Performance

§101
2.8%
-37.2% vs TC avg
§103
45.0%
+5.0% vs TC avg
§102
31.9%
-8.1% vs TC avg
§112
17.6%
-22.4% vs TC avg
Black line = Tech Center average estimate • Based on career data from 247 resolved cases

Office Action

§102 §103
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 . 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)(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) 1-11 and 13-19 is/are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Horesh (U.S.P. 11,972,333). With respect to claim 1, Horesh teaches over a long period of time, (weeks, months, years) data used for training the original AI model will generate stale and incorrect output information regarding certain standards. Such stale or misinformative information may skew the accuracy of the results. To some users, stale information is not desired and the information should be purged from the old AI model. Yet to other users, older information may provide a useful result. In the case that the stale or outdated information is to be disgorged, the old AI model generates a first output data. The old AI model is then trained with a second data training set to produce a second output data. The first and second outputs are sent to a classifier to determine their similarity of results. Based on the similarity results, the computer system may elect to adopt different policies to optimize the performance of the AI model. Horesh teaches (see para. 4, lines 22-30), using a second generative AI model trained on a second input training data set. The output from the first generative model is supplied as an input to the second generative AI model. The respective outputs of the first generative AI model 170 and the second generative AI model 140 are sent to a classification model 150 (see para. 4, lines 30-35), see also col. 8, lines 40-45. The classification model makes a comparison between the first output (initial training dataset) and second output (emulated training set). See col. 4, lines 32-34. Horsh teaches a computer system (col. 4. lines 35-36) for measuring the distance of similarity (similarity indication), see also col. 9, line 65- col. 10, line 6 where the reference teaches determining if the similarity metric is less than a defined threshold, therefore indicating that the result of the first output is not relevant or otherwise desired as the output. Horesh teaches training the second generative AI model 140, see col. 7, line 65, col. 8, line 6. At col. 10, Horesh teaches a policy engine 160 that looks at the threshold distance and makes a determination that the output from the second generative model 140 provides the output that the user indicates yields acceptable results (see col. 10, lines 25-30 and 35-40). See also col. 5, lines 5-11 regarding retraining or updating a training set based on the results of the similarity check. With respect to claim 2, Horesh teaches the outputs from both the first and second AI models, 170 and 140 respectively, may be text in in a natural language description (format) – (See col. 8, lines 55-60). Therefore, the output of the first model, which represents the initial training dataset, may be converted by a natural language processor. The BERT transformer model has been identified as one possible means for generating the natural language description. Moreover, Horesh teaches that output of the emulated training set by means of the second generative AI model may also convert text into a natural language format by a natural language processor, see col. 8, lines 55-60. With respect to claim 3, Horesh teaches the system 100 will provide a prompt to the user to approve results generated by the policy engine 160 between the first and second AI engines. See col. 10, lines 27-40. With respect to claim 4, Horesh teaches part of the first data could be data that is older, and in some regards could be considered to be outdated. However, the output result of the first AI model 170 is sent as in input to the second AI model 140. The output results of both models 140 and 170 are fed to a classifier where a determination is made as to how similar the outputs are. Horesh teaches a computer system (col. 4. Lines 35-36) for measuring the distance of similarity (similarity indication), see also col. 9, line 65- col. 10, line 6. If the similarity result is greater than a threshold value, then the results are said to be very similar and data from the first AI model has been input to the second AI model 140 and is a part of the data set for the second model which generates the second output information. With respect to claim 5, Horesh teaches a method comprising: receiving an initial training dataset by a first machine learning model (170), see col. 4, lines 20-25). Horesh teaches (see para. 4, lines 22-30), using a second generative AI model trained on a second input training data set. The output from the first generative model is supplied as an input to the second generative AI model. Horesh teaches training the second generative AI model 140, see col. 7, line 65, col. 8, line 6. At col. 10, lines 1-6 and 15-17, Horesh teaches a policy engine 160 that looks at the threshold distance and makes a determination that the output from the second generative model 140 provides the output that the user indicates yields acceptable results (see col. 10, lines 25-30 and 35-40). See also col. 5, lines 5-11 regarding retraining or updating a training set based on the results of the similarity check. With respect to claim 6, Horesh teaches wherein the first and second machine models are generative machine models, see col. 8, lines 33-36. With respect to claim 7, Horesh teaches the outputs from both the first and second AI models, 170 and 140 respectively, may be text in a natural language description (format) – (See col. 8, lines 55-60). Therefore, the output of the first model, which represents the initial training dataset, may be converted by a natural language processor. The BERT transformer model has been identified as one possible means for generating the natural language description. Moreover, Horesh teaches that output of the emulated training set by means of the second generative AI model may also convert text into a natural language format by a natural language processor, see col. 8, lines 55-60. With respect to claim 8, Horesh teaches a computer system (col. 4., lines 35-36) for measuring the distance of similarity (similarity indication). At col. 9, line 65 - col. 10, line 6. Horesh teaches determining whether the similarity metric is less than a defined threshold, therefore indicating that the result of the first output is not relevant or otherwise desired for output. At col. 5, lines 5-11, Horesh teaches making a system determination of retraining or training with a different dataset using the emulated training dataset which is fed to generative AI model 140. With respect to claim 9, Horesh teaches determining whether the similarity satisfies a predetermined distance between generative AI models. At col. 9, line 65 - col. 10, line 6, Horesh teaches determining whether the similarity metric is less than a defined threshold. The similarity metric being less than a “defined threshold”, corresponds with the claimed threshold value. With respect to claim 10, Horesh teaches the system 100 will provide a prompt to the user to approve results generated by the policy engine 160 between the first and second AI engines. See col. 10, lines 27-40. With respect to claim 11, Horesh teaches part of the first data could be data that is older and in some regards could be considered to be outdated. However, the output result of the first AI model 170 is sent as in input to the second AI model 140. The output results of both models 140 and 170 are fed to a classifier where a determination is made as to how similar the outputs are. Horesh teaches a computer system (col. 4. lines 35-36) for measuring the distance of similarity (similarity indication), see also col. 9, line 65- col. 10, line 6. If the similarity result is greater than a threshold value, then the results are said to be very similar and data from the first AI model has been input to the second AI model 140 and is a part of the data set for the second model which generates the second output information. With respect to claim 13, Horesh teaches a system having a memory (computer readable medium, which may include a RAM, ROM, EEPROM or other disk or optical storage media), see col. 25, lines 12-21. Horesh teaches wherein the computer readable medium stores instructions (col. 25, lines 14, 15 and lines 20-26) Horesh teaches one or more processors, see para. 24, lines 55-64, for executing the instructions stored in the memory, as set forth above. Horesh teaches receiving an initial training dataset by a first machine learning model (170); col. 4, lines 20-25). Horesh teaches (see para. 4, lines 22-30), using a second generative AI model trained on a second input training data set. The output from the first generative model is supplied as an input to the second generative AI model. Horesh teaches training the second generative AI model 140, see col. 7, line 65, and col. 8, line 6. At col. 10, lines 1-6 and 15-17,Horesh teaches a policy engine 160 that looks at the threshold distance and makes a determination that the output from the second generative model 140 provides the output that the user indicates yields acceptable results (see col. 10, lines 25-30 and 35-40). See also col. 5, lines 5-11 regarding retraining or updating a training set based on the results of the similarity check. With respect to claim 14, Horesh teaches the first and second machine models are generative machine models, see col. 8, lines 33-36. With respect to claim 15, Horesh teaches one or more processors for executing the instructions stored in the memory, see para. 24, lines 55-64. Horesh teaches causing the instructions to output from both the first and second AI models, 170 and 140 respectively, a text in the natural language description (format) – (See col. 8, lines 55-60). Therefore, the output of the first model, which represents the initial training dataset, may be converted by a natural language processor. The BERT transformer model has been identified as one possible means for generating the natural language description. Moreover, Horesh teaches that output of the emulated training set by means of the second generative AI model may also convert text into a natural language format by a natural language processor, see col. 8, lines 55-60. With respect to claim 16, Horesh teaches a computer system (see col. 4., lines 35-36) for measuring the distance of similarity (similarity indication). At col. 9, line 65 - col. 10, line 6, Horesh teaches determining whether the similarity metric is less than a defined threshold, therefore indicating that the result of the first output is not relevant or otherwise desired for output. At col. 5, lines 5-11, Horesh teaches making a system determination of retraining or training with a different dataset using the emulated training dataset which is fed to generative AI model 140. With respect to claim 17, Horesh teaches determining whether the similarity satisfies a predetermined distance between generative AI models. At col. 9, line 65 - col. 10, line 6 Horesh, teaches determining whether the similarity metric is less than a defined threshold. The similarity metric being less than a “defined threshold”, corresponds with the claimed threshold value. With respect to claim 18, Horesh teaches the system 100 will provide a prompt to the user to approve results generated by the policy engine 160 between the first and second AI engines. See col. 10, lines 27-40. With respect to claim 19, Horesh teaches part of the first data could be data that is older and in some regards could be considered to be outdated. However, the output result of the first AI model 170 is sent as in input to the second AI model 140. The output results of both models 140 and 170 are fed to a classifier where a determination is made as to how similar the outputs are. Horesh teaches a computer system (col. 4. lines 35-36) for measuring the distance of similarity (similarity indication), see also col. 9, line 65- col. 10, line 6. If the similarity result is greater than a threshold value, then the results are said to be very similar and data from the first AI model has been input to the second AI model 140 and is a part of the data set for the second model which generates the second output information. 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) 12 and 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Horesh (US Pat. 11, 972,333) in view of Sargent (2021/0082092). With respect to claim 12, Horesh teaches all of the subject matter upon which the claim depends except for a first data including biometric information and the second image not including biometric information. Horesh does not teach biometric information. Therefore, biometric information is not included in the second image. Sargent teaches a computer system that uses a processor for removing medical artifacts in an image. Sargent teaches biometric information 114, such as prior images of the patient, area of imaging and possible the voiceprints of the patient, see col. 26, lines 1-4, col. 27, lines 5-9 and para. 28, last four lines. Since Horesh and Sargent are directed to removing artifacts that are either stale or personal, the purpose of removing biometric information would have been recognized by Horesh as set forth by Sargent. It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to use the teaching of Horesh and to use any combination of voice or text as the data that is subject to Generative models 140 and 170 for the purpose of purging a first generative model which could contain biometric information or any other information that is stale or private. With respect to claim 20, Horesh teaches one or more processors for executing the instructions stored in the memory, see para. 24, lines 55-64 to perform the limitations of the claim. Horesh teaches causing the instructions to control the first and second AI models, 170 and 140 respectively. Horesh does not teach biometric information. Therefore, biometric information is not included in the second image. Sargent teaches a computer system that uses a processor for removing medical artifacts in an image. Sargent teaches biometric information 114, such as prior images of the patient, area of imaging and possible the voiceprints of the patient, see col. 26, lines 1-4, col. 27, lines 5-9 and para. 28, last four lines. Since Horesh and Sargent are directed to removing artifacts that are either stale or personal, the purpose of removing biometric information would have been recognized by Horesh as set forth by Sargent. It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to use the teaching of Horesh and to use any combination of voice or text as the data that is subject to Generative models 140 and 170 for the purpose of purging a first generative model which could contain biometric information or any other information that is stale or private. Any inquiry concerning this communication or earlier communications from the examiner should be directed to JEROME GRANT II whose telephone number is (571) 272-7463. The examiner can normally be reached M-F 9:00 a.m. - 5:00 p.m.. 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 Mehmood can be reached at 571-272- 2976 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. /JEROME GRANT II/Primary Examiner, Art Unit 2664
Read full office action

Prosecution Timeline

Feb 06, 2024
Application Filed
Jan 22, 2026
Non-Final Rejection — §102, §103
Feb 23, 2026
Applicant Interview (Telephonic)
Feb 24, 2026
Examiner Interview Summary

Precedent Cases

Applications granted by this same examiner with similar technology

Patent 12572774
NEURAL NETWORK PROCESSOR AND METHOD OF NEURAL NETWORK PROCESSING
2y 5m to grant Granted Mar 10, 2026
Patent 10269295
ORGANIC LIGHT EMITTING DISPLAY DEVICE AND DRIVING METHOD THEREOF
2y 5m to grant Granted Apr 23, 2019
Patent 9245189
OBJECT APPEARANCE FREQUENCY ESTIMATING APPARATUS
2y 5m to grant Granted Jan 26, 2016
Patent 8344909
METHOD AND SYSTEM FOR COLLECTING TRAFFIC DATA, MONITORING TRAFFIC, AND AUTOMATED ENFORCEMENT AT A CENTRALIZED STATION
2y 5m to grant Granted Jan 01, 2013
Patent 8294567
METHOD AND SYSTEM FOR FIRE DETECTION
2y 5m to grant Granted Oct 23, 2012
Study what changed to get past this examiner. Based on 5 most recent grants.

AI Strategy Recommendation

Get an AI-powered prosecution strategy using examiner precedents, rejection analysis, and claim mapping.
Powered by AI — typically takes 5-10 seconds

Prosecution Projections

1-2
Expected OA Rounds
65%
Grant Probability
95%
With Interview (+30.6%)
3y 1m
Median Time to Grant
Low
PTA Risk
Based on 247 resolved cases by this examiner. Grant probability derived from career allow rate.

Sign in with your work email

Enter your email to receive a magic link. No password needed.

Personal email addresses (Gmail, Yahoo, etc.) are not accepted.

Free tier: 3 strategy analyses per month