DETAILED ACTION
This action is responsive to communications filed on March 28, 2024. This action is made Non-Final.
Claims 1-20 are pending in the case.
Claims 1, 11, and 20 are independent claims.
Claims 1-20 are rejected.
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 Interpretation
The Specification recites “In the context of this document, a computer readable storage medium may be any tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. Para. 0009. Further, “computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.” Para. 0010. Based on the recitations, the Examiner is interpreting “computer readable medium” in claim 1 and “non-transient computer readable medium” in claim 20 as non-transitory computer-readable media. Accordingly, claims 1-10 and 20 are statutory under a 35 USC 101 computer-readable medium analysis.
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.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more.
Step 1: Independent claims 1, 11, and 20 are directed towards a system, method, and medium(interpreted as non-transitory), respectively. Therefore, these claims, as well as their dependent claims, are directed towards one of the four statutory categories (process, machine (i.e. apparatus), manufacture, or composition of matter.
With respect to claim 1:
2A Prong 1:
Claim 1 recites the following judicial exceptions:
determine a training scope of a first instance of a first Artificial Intelligence (AI) algorithm based on a training corpus used to train the first instance of the first AI algorithm (mental process –can be performed in the human mind, or by a human using a pen and paper (e.g. a person may analyze a set of data and identify the data scope).
determine if some or all of the first input AI input data is not within the training scope of the first instance of the first AI algorithm (mental process –can be performed in the human mind, or by a human using a pen and paper (e.g. a person may analyze a set of data and identify the data scope and determine if an input is not within the scope).
in response to determining that the some or all of the first AI input data is not within the training scope of the first instance of the first AI algorithm, filter out the some or all of the first AI input data (mental process –can be performed in the human mind, or by a human using a pen and paper (e.g. a person may analyze a set of data and identify the data scope and determine if an input is not within the scope and then have the data removed from the input).
2A Prong 2: The additional elements recited in the claim do not integrate the judicial exception into a practical application.
Additional elements:
a system comprising: a microprocessor; and a computer readable medium, coupled with the microprocessor and comprising microprocessor readable and executable instructions that, when executed by the microprocessor, cause the microprocessor to: (mere instructions to apply the exception or implement the exception on a computer (e.g. using a computer analyze data; see MPEP §2106.05(f).). The additional elements do not effectively integrate the abstract idea into a practical application. 2B: revisiting the additional elements, the additional elements do not amount to significantly more than the judicial exception – recited high level of generality and corresponds to storing and retrieving information in memory and performing calculations).
receive, first AI input data, wherein the first AI input data is input data for the first instance of the first AI algorithm (mere instructions to apply the exception or implement the exception on a computer (e.g. using a computer analyze data; see MPEP §2106.05(f).). The additional elements do not effectively integrate the abstract idea into a practical application. 2B: revisiting the additional elements, the additional elements do not amount to significantly more than the judicial exception – recited high level of generality and corresponds to storing and retrieving information in memory and performing calculations).
With respect to claim 2:
2A Prong 1:
Claim 2 recites the following judicial exceptions:
wherein the training scope of the first instance of the first AI algorithm is determined … but in a negative manner to identify some or all of the first AI input data (mental process –can be performed in the human mind, or by a human using a pen and paper (e.g. a person may analyze a set of data and identify the data scope and use negative examples to verify data scope boundaries).
2A Prong 2: The additional elements recited in the claim do not integrate the judicial exception into a practical application.
Additional elements:
by a second AI algorithm and wherein the second AI algorithm is also trained using the training corpus used to train the first instance of the first AI algorithm (generally linking the use of a judicial exception to a particular technological environment or field of use (e.g. AI/ML models to make determinations or predictions and training the models with appropriate data; see MPEP §2106.05(h).). The additional elements do not effectively integrate the abstract idea into a practical application. 2B: revisiting the additional elements, the additional elements do not amount to significantly more than the judicial exception – recited high level of generality and corresponds to storing and retrieving information and performing calculations.).
With respect to claim 3:
2A Prong 1:
Claim 3 recites the following judicial exceptions:
wherein the training scope of the first instance of the first AI algorithm is determined … and wherein feedback filter data is generated for use as a negative input to the first AI algorithm (mental process –can be performed in the human mind, or by a human using a pen and paper (e.g. a person may analyze a set of data and identify the data scope and put forth negative input feedback for processing).
2A Prong 2: The additional elements recited in the claim do not integrate the judicial exception into a practical application.
Additional elements:
by a second AI algorithm (generally linking the use of a judicial exception to a particular technological environment or field of use (e.g. AI/ML models to make determinations or predictions and training the models with appropriate data; see MPEP §2106.05(h).). The additional elements do not effectively integrate the abstract idea into a practical application. 2B: revisiting the additional elements, the additional elements do not amount to significantly more than the judicial exception – recited high level of generality and corresponds to storing and retrieving information and performing calculations.).
With respect to claim 4:
2A Prong 1:
Claim 4 recites the following judicial exceptions:
wherein the training scope of the first instance of the first AI algorithm is determined … scans AI output data from the first AI algorithm to identify additional AI input data that is outside the training scope of the first AI algorithm (mental process –can be performed in the human mind, or by a human using a pen and paper (e.g. a person may analyze a set of data and identify the data scope and analyze output in relation to scope boundary.).
2A Prong 2: The additional elements recited in the claim do not integrate the judicial exception into a practical application.
Additional elements:
by a second AI algorithm and wherein the second AI algorithm (generally linking the use of a judicial exception to a particular technological environment or field of use (e.g. AI/ML models to make determinations or predictions and training the models with appropriate data; see MPEP §2106.05(h).). The additional elements do not effectively integrate the abstract idea into a practical application. 2B: revisiting the additional elements, the additional elements do not amount to significantly more than the judicial exception – recited high level of generality and corresponds to storing and retrieving information and performing calculations.).
With respect to claim 5:
2A Prong 2: The additional elements recited in the claim do not integrate the judicial exception into a practical application.
Additional elements:
wherein the microprocessor readable and executable instructions further cause the microprocessor to: receive training input data, wherein the training input data is used to define the training scope of the first instance of the first AI algorithm, wherein the training input data is used to create the training corpus … based on a search of a corpus database (mere instructions to apply the exception or implement the exception on a computer (e.g. using a computer receive and analyze data and generate more data by searching and retrieving the data from a database; see MPEP §2106.05(f).). The additional elements do not effectively integrate the abstract idea into a practical application. 2B: revisiting the additional elements, the additional elements do not amount to significantly more than the judicial exception – recited high level of generality and corresponds to storing and retrieving information in memory and performing calculations).
used to train the first instance of the first AI algorithm, and wherein the training corpus used to train the first instance of the first AI algorithm is created (generally linking the use of a judicial exception to a particular technological environment or field of use (e.g. AI/ML models to make determinations or predictions and training the models with appropriate data; see MPEP §2106.05(h).). The additional elements do not effectively integrate the abstract idea into a practical application. 2B: revisiting the additional elements, the additional elements do not amount to significantly more than the judicial exception – recited high level of generality and corresponds to storing and retrieving information and performing calculations.).
With respect to claim 6:
2A Prong 1:
Claim 6 recites the following judicial exceptions:
for training data that matches the some or all of the first AI input data that is not within the training scope to identify new training data; add the new training data to the training corpus (mental process –can be performed in the human mind, or by a human using a pen and paper (e.g. a person may analyze a set of data and identify the data scope and identify matching data outside of the scope to use as new data.).
2A Prong 2: The additional elements recited in the claim do not integrate the judicial exception into a practical application.
Additional elements:
wherein the microprocessor readable and executable instructions further cause the microprocessor to: provide the some or all of the first AI input data that is not within the training scope to a search engine; search, using the search engine (mere instructions to apply the exception or implement the exception on a computer (e.g. using a computer receive and analyze data and generate more data by searching and retrieving the data from a database; see MPEP §2106.05(f).). The additional elements do not effectively integrate the abstract idea into a practical application. 2B: revisiting the additional elements, the additional elements do not amount to significantly more than the judicial exception – recited high level of generality and corresponds to storing and retrieving information in memory and performing calculations).
used to train the first instance of the first AI algorithm to produce a new training corpus; and retrain a second instance of the first AI algorithm using the new training corpus (generally linking the use of a judicial exception to a particular technological environment or field of use (e.g. AI/ML models to make determinations or predictions and training and retraining the models with appropriate data; see MPEP §2106.05(h).). The additional elements do not effectively integrate the abstract idea into a practical application. 2B: revisiting the additional elements, the additional elements do not amount to significantly more than the judicial exception – recited high level of generality and corresponds to storing and retrieving information and performing calculations.).
With respect to claim 7:
2A Prong 2: The additional elements recited in the claim do not integrate the judicial exception into a practical application.
Additional elements:
wherein, upon the second instance of the first AI algorithm being trained using the new training corpus, the second instance of the first AI algorithm becomes active, and the first instance of the first AI algorithm becomes inactive (generally linking the use of a judicial exception to a particular technological environment or field of use (e.g. AI/ML models to make determinations or predictions and training and retraining the models with appropriate data and using updated models; see MPEP §2106.05(h).). The additional elements do not effectively integrate the abstract idea into a practical application. 2B: revisiting the additional elements, the additional elements do not amount to significantly more than the judicial exception – recited high level of generality and corresponds to storing and retrieving information and performing calculations.).
With respect to claim 8:
2A Prong 1:
Claim 8 recites the following judicial exceptions:
wherein the first instance of the first AI algorithm is selected based on an AI algorithm type as part of an AI as a Service (AIaaS) and wherein the training scope of the first instance of the first AI algorithm is defined when selecting the first instance of the first AI algorithm based on the AI algorithm type (mental process –can be performed in the human mind, or by a human using a pen and paper (e.g. a person may analyze and choose an algorithm type and corresponding data.).
With respect to claim 9:
2A Prong 2: The additional elements recited in the claim do not integrate the judicial exception into a practical application.
Additional elements:
wherein the training scope of the first instance of the first AI algorithm is generated (generally linking the use of a judicial exception to a particular technological environment or field of use (e.g. AI/ML models to make determinations or predictions and training and retraining the models with appropriate data and using updated models; see MPEP §2106.05(h).). The additional elements do not effectively integrate the abstract idea into a practical application. 2B: revisiting the additional elements, the additional elements do not amount to significantly more than the judicial exception – recited high level of generality and corresponds to storing and retrieving information and performing calculations.).
based on a search of a corpus database (mere instructions to apply the exception or implement the exception on a computer (e.g. using a computer to make determinations based on database searches; see MPEP §2106.05(f).). The additional elements do not effectively integrate the abstract idea into a practical application. 2B: revisiting the additional elements, the additional elements do not amount to significantly more than the judicial exception – recited high level of generality and corresponds to storing and retrieving information in memory and performing calculations).
With respect to claim 10:
2A Prong 1:
Claim 10 recites the following judicial exceptions:
determine that the training corpus used to train the first instance of the first AI algorithm does not have enough training data for the training scope (mental process –can be performed in the human mind, or by a human using a pen and paper (e.g. a person may analyze a set of data and identify the data scope and determine more data is needed.).
2A Prong 2: The additional elements recited in the claim do not integrate the judicial exception into a practical application.
Additional elements:
wherein the microprocessor readable and executable instructions further cause the microprocessor to: retrieve training data that is within the training scope of the first instance of the first Artificial Intelligence (Al) algorithm; generate new synthetic training data that is within the training scope of the first instance of the first AI algorithm; and add the new synthetic training data to the training corpus used to train the first instance of the first AI algorithm. (mere instructions to apply the exception or implement the exception on a computer (e.g. using a computer retrieve and analyze data and generate more data by searching and retrieving the data from a database and add the data to the dataset; see MPEP §2106.05(f).). The additional elements do not effectively integrate the abstract idea into a practical application. 2B: revisiting the additional elements, the additional elements do not amount to significantly more than the judicial exception – recited high level of generality and corresponds to storing and retrieving information in memory and performing calculations).
used to train the first instance of the first AI algorithm to produce a new training corpus; and retrain a second instance of the first AI algorithm using the new training corpus (generally linking the use of a judicial exception to a particular technological environment or field of use (e.g. AI/ML models to make determinations or predictions and training and retraining the models with appropriate data; see MPEP §2106.05(h).). The additional elements do not effectively integrate the abstract idea into a practical application. 2B: revisiting the additional elements, the additional elements do not amount to significantly more than the judicial exception – recited high level of generality and corresponds to storing and retrieving information and performing calculations.).
Claim 11:
Claim 11 substantially corresponds to claim 1 and is rejected under the same rationale.
Claim 12:
Claim 12 substantially corresponds to claim 2 and is rejected under the same rationale.
Claim 13:
Claim 13 substantially corresponds to claim 3 and is rejected under the same rationale.
Claim 14:
Claim 14 substantially corresponds to claim 4 and is rejected under the same rationale.
Claim 15:
Claim 15 substantially corresponds to claim 5 and is rejected under the same rationale.
Claim 16:
Claim 16 substantially corresponds to claim 6 and is rejected under the same rationale.
Claim 17:
Claim 17 substantially corresponds to claim 7 and is rejected under the same rationale.
Claim 18:
Claim 18 substantially corresponds to claim 8 and is rejected under the same rationale.
Claim 19:
Claim 19 substantially corresponds to claim 10 and is rejected under the same rationale.
Claim 20:
Claim 20 substantially corresponds to claim 1 and is rejected under the same rationale.
2B continued: After considering all claim elements individually and as an ordered combination, it is determined that the claims do not include any additional elements that are sufficient to amount to significantly more than the judicial exception.
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 20 is/are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Lang et al., US Publication 2024/0202405 (“Lang”).
Claim 1:
Lang teaches or suggests a system comprising:
a microprocessor; and a computer-readable medium, coupled with the microprocessor and comprising microprocessor readable and executable instructions that, when executed by the microprocessor, cause the microprocessor to:
determine a training scope of a first instance of a first Artificial Intelligence (AI) algorithm based on a training corpus used to train the first instance of the first (AI) algorithm (see para. 0105 - Out of Distribution (OOD)-in ML, OOD is an uncertainty that arises when an AI model sees an input that differs (potentially substantially) from its training data, leading to potentially incorrect predictions; para. 0120 – detecting vulnerabilities and/or risks, evaluating the models against OOD data; para. 0330 - detecting OOD data, etc. This module may quarantine adversarial data and/or prevent further downstream actions.);
receive, first AI input data, wherein the first AI input data is input data for the first instance of the first AI algorithm (see para. 0105 - Out of Distribution (OOD)-in ML, OOD is an uncertainty that arises when an AI model sees an input that differs (potentially substantially) from its training data, leading to potentially incorrect predictions; para. 0274 - vulnerabilities are discovered, it may automatically and/or semi-automatically (using a human-in-the-loop mechanism) mitigate and/or defend the AI system(s). para. 0122 – analysis system may for example monitor for specific boundary violations in inputs and/or outputs of the model, detect model drift, etc.; This may include … blocking specific inputs (e.g., adversarial attacks, OOD data, etc.); para. 0330 - detecting OOD data, etc. This module may quarantine adversarial data and/or prevent further downstream actions.);
determine if some or all of the first AI input data is not within the training scope of the first instance of the first AI algorithm (see para. 0105 - Out of Distribution (OOD)-in ML, OOD is an uncertainty that arises when an AI model sees an input that differs (potentially substantially) from its training data, leading to potentially incorrect predictions; para. 0274 - vulnerabilities are discovered, it may automatically and/or semi-automatically (using a human-in-the-loop mechanism) mitigate and/or defend the AI system(s). This may include … blocking specific inputs (e.g., adversarial attacks, OOD data, etc.); para. 0122 – analysis system may for example monitor for specific boundary violations in inputs and/or outputs of the model, detect model drift, etc.; para. 0330 - detecting OOD data, etc. This module may quarantine adversarial data and/or prevent further downstream actions.);
in response to determining that the some or all of the first AI input data is not within the training scope of the first instance of the first AI algorithm, filter out the some or all of the first AI input data (see para. 0105 - Out of Distribution (OOD)-in ML, OOD is an uncertainty that arises when an AI model sees an input that differs (potentially substantially) from its training data, leading to potentially incorrect predictions; para. 0122 - Mitigations may include, but are not limited to, blocking access to specific components of the systems, shutting down entire systems, blocking access by a specific user, blocking specific types of inputs. analysis system may for example monitor for specific boundary violations in inputs and/or outputs of the model, detect model drift, etc.; para. 0274 - vulnerabilities are discovered, it may automatically and/or semi-automatically (using a human-in-the-loop mechanism) mitigate and/or defend the AI system(s). This may include … blocking specific inputs (e.g., adversarial attacks, OOD data, etc.); para. 0330 - detecting OOD data, etc. This module may quarantine adversarial data and/or prevent further downstream actions.).
Claim(s) 11 and 20:
Claim(s) 11 and 20 correspond to claim 1, and thus, Lang discloses the limitations of claim(s) 11 and 20 as well.
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) 2 and 12 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lang, in view of Pezzotti et al., US Publication 2023/0377314 (“Pezzotti”), and further in view of Yuan et al., US Publication 2022/0036890 (“Yuan”).
Claim 2:
Lang does not explicitly disclose wherein the training scope of the first instance of the first AI algorithm is determined by a second AI algorithm and wherein the second AI algorithm is also trained using the training corpus used to train the first instance of the first AI algorithm, but in a negative manner to identify the some or all of the first input data.
Pezzotti teaches wherein the training scope of the first instance of the first AI algorithm is determined by a second AI algorithm and wherein the second AI algorithm is also trained using the training corpus used to train the first instance of the first AI algorithm … (see para. 0007 - perform the OOD detection, multiple secondary models may be used. A secondary model may be trained on the same training dataset on which the main model is trained; 0008 - provide OOD detection, various aspects use multiple secondary models, trained on the same training dataset as the main model; para. 0010 – by using secondary models with fewer trainable parameters and/or smaller inputs and/or smaller outputs, the overhead of computing the OOD score with respect to applying the main model may be limited; para. 0071 - secondary models SMi may be models that are trained on the same training dataset as the main model MM; para. 0073 - enable to use smaller secondary models, e.g., having fewer trainable parameters than the main model, leading to reduced storage and computational requirements.).
Accordingly, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Pezzotti with those of Lang. One would have been motivated to do so for the purpose of using less computing resources by training the secondary model(s) to perform secondary tasks including OOD detection tasks, as taught by Pezzotti (0010, 0071, and 0073).
Yuan further teaches or suggests to but in a negative manner to identify the some or all of the first input data (see para. 0105 - determining a corresponding OOD detector network parameter and a corresponding domain classification network parameter in the semantic understanding model; para. 0136 - negative sample set is configured to adjust an OOD detector network parameter and a domain classification network parameter of the semantic understanding model; para. 0180 - negative sample distribution optimization strategy is provided through analysis of bad cases and experiments. In some embodiments of this application, the strategy includes: grouping negative samples according to importance; para. 0181 - negative samples are grouped and weights thereof are finely adjusted, which can effectively reduce a misrecognition rate of the model; para. 0182 - Train the semantic understanding model by using the optimized training samples to determine parameters of the semantic understanding model; para. 0183 - used to recognize and process speech instructions in a noisy environment; para. 0187 – because OOD and domain classifiers are two very related tasks, if the corpus is OOD, the corpus is definitely a negative sample of binary classifiers in all domains; and if the corpus is IND, the corpus is definitely one of the domain classifiers or a positive sample in a plurality of fields; para. 0190 - the negative samples are grouped and different weights are assigned to different groups to adjust the internal sample distribution, so that the misrecognition rate is further reduced.).
Accordingly, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Yuan with those of Lang. One would have been motivated to do so for the purpose efficiently reducing misrecognition rate by using appropriately weighted negative samples in model training, as taught by Yuan (0180, 0181, and 0190).
Claim(s) 12:
Claim(s) 12 correspond to claim 2, and thus, Lang, Pezzotti, and Yuan teach or suggest discloses the limitations of claim(s) 12 as well.
Claim(s) 3 and 13 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lang, in view of Pezzotti, and further in view of Skarphedinsson et al., US Patent 12,719,896 (“Skarphedinsson”).
Claim 3:
Pezzotti further teaches or suggests wherein the training scope of the first instance of the AI algorithm is determined by a second AI algorithm (see para. 0007 - perform the OOD detection, multiple secondary models may be used. A secondary model may be trained on the same training dataset on which the main model is trained; 0008 - provide OOD detection, various aspects use multiple secondary models, trained on the same training dataset as the main model; para. 0010 – by using secondary models with fewer trainable parameters and/or smaller inputs and/or smaller outputs, the overhead of computing the OOD score with respect to applying the main model may be limited; para. 0071 - secondary models SMi may be models that are trained on the same training dataset as the main model MM; para. 0073 - enable to use smaller secondary models, e.g., having fewer trainable parameters than the main model, leading to reduced storage and computational requirements.).
Accordingly, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Pezzotti with those of Lang. One would have been motivated to do so for the purpose of using less computing resources by training the secondary model(s) to perform secondary tasks including OOD detection tasks, as taught by Pezzotti (0010, 0071, and 0073).
Skarphedinsson further teaches or suggests and wherein feedback filter data is generated for use as a negative input to the first AI algorithm (see col. 123, lines 41-54 - selection of the other selectable element may cause a command or message to be provided to the generative AI model indicating that a user declined performance of the one or more actions. For example, such a command or message may be included in a prompt to the 45 generative AI model to reprocess the previously received natural language input (e.g., to determine different actions to be performed via another UI widget). In some embodiments, data indicating selection of the other selectable element may be stored as feedback (e.g., negative feedback) for retraining the generative AI model. In some embodiments, a user may be solicited for additional information as to why they declined performance of the one or more actions by selecting the other selectable element.).
Accordingly, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Skarphedinsson with those of Lang. One would have been motivated to do so for the purpose of tailoring a generative AI model by providing negative feedback to the model directing the model not to perform one or more actions, improving model performance, as taught by Skarphedinsson (col. 123).
Claim(s) 13:
Claim(s) 13 correspond to claim 3, and thus, Lang, Pezzotti, and Skarphedinsson teach or suggest discloses the limitations of claim(s) 13 as well.
Claim(s) 4 and 14 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lang, in view of Pezzotti.
Claim 4:
Lang further teaches or suggests wherein the second AI algorithm scans AI output data from the first AI algorithm to identify additional AI input data (see para. 0110 - Surrogate model-A mechanism, program, function, or other representation of an AI model that is capable of representing and/or simulating the behavior and/or the characteristics of a given AI model. In some instances, surrogate models may be less complex than the AI model they are representing/simulating; less complex may include being smaller in size (e.g., fewer layers, inputs, outputs of a neural net, less complex formulas), more performant (produces same/similar result with less processing resources required), etc.; para. 0120 - surrogate model analysis may be used to assess the model; para. 0122 - Monitoring may include continuously assessing logs, inputs, and/or outputs, etc., of the model. Analysis system may for example monitor for specific boundary violations in inputs and/or outputs of the model, detect model drift, etc.; para. 0125 - the inputs and/or outputs may for example be analyzed using surrogate model analysis for explainability and/or to detect deviations from a baseline; para. 0131 – verification that a model is acting within expected bounds.).
Pezzotti further teaches or suggests wherein the training scope of the first instance of the first AI algorithm is determined by a second AI algorithm … that is outside the training scope of the first instance of the first AI algorithm (see para. 0007 - perform the OOD detection, multiple secondary models may be used. A secondary model may be trained on the same training dataset on which the main model is trained; 0008 - provide OOD detection, various aspects use multiple secondary models, trained on the same training dataset as the main model; para. 0010 – by using secondary models with fewer trainable parameters and/or smaller inputs and/or smaller outputs, the overhead of computing the OOD score with respect to applying the main model may be limited; para. 0071 - secondary models SMi may be models that are trained on the same training dataset as the main model MM; para. 0073 - enable to use smaller secondary models, e.g., having fewer trainable parameters than the main model, leading to reduced storage and computational requirements.).
Accordingly, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Pezzotti with those of Lang. One would have been motivated to do so for the purpose of using less computing resources by training the secondary model(s) to perform secondary tasks including OOD detection tasks, as taught by Pezzotti (0010, 0071, and 0073).
Claim(s) 14:
Claim(s) 14 correspond to claim 4, and thus, Lang and Pezzotti teach or suggest discloses the limitations of claim(s) 14 as well.
Claim(s) 5 and 15 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lang, and further in view of Chen et al., US Publication 2022/0179691 (“Chen”).
Claim 5:
Lang further teaches or suggests receive training data input (see para. 0279 - potential items that may be selected as targets or configurations by a user, by an automated training system, and/or through a human-in-the-loop mechanism. A human-in-the-loop mechanism may include feedback from a user on the automated training selections, and/or may provide feedback to the system prior to automated training. In the depicted example in FIG. 23, these items may include framework 2305, model type 2310, optimization function 2315, loss function 2320, epochs 2325, and/or learning rate 2330 selections. There may be other items presented on this screen, including but not limited to, batch size, optimizations, and/or model architecture evaluation metrics, etc.).
Chen further teaches or suggests wherein the training input data is used to define the training scope of the first instance of the first AI algorithm, wherein the training input data is used to create the training corpus used to train the first instance of the first AI algorithm, and wherein the training corpus used to train the first instance of the first AI algorithm is created based on a search of a corpus database (see para. 0023 - receive the experiment request associated with a target dataset, wherein the experiment request includes the information of the target dataset for training a machine learning model. More particularly, a user may upload files of multiple datasets to a file system which can also be called Dataset Store in advance, and the description of the target dataset may be stored in the database 16. When a user would like to train a machine learning model using the target dataset, the user may input a selection instruction of the target dataset through input interface 11. The experiment generator 12 asks the file system for the file of the target dataset according to the selection instruction, and the file system accordingly searches for the description of the target dataset in the database 16. Or, the user may directly provide the file of the target dataset to the machine learning system 1 through the target dataset.).
Accordingly, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Chen with those of Lang. One would have been motivated to do so for the purpose efficiently enabling a user to configure model training including configuring the training data used, improving model tailoring, as taught by Chen (0023).
Claim(s) 15:
Claim(s) 15 correspond to claim 5, and thus, Lang and Chen teach or suggest discloses the limitations of claim(s) 15 as well.
Claim(s) 6, 7, 16, and 17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lang, and further in view of Park et al., US Publication 2025/01567452 (“Park”).
Claim 6:
As indicated above, Lang discloses the first AI input data.
Park further teaches or suggests provide the some or all of the … data that is not within the training scope to a search engine; search, using the search engine, for training data that matches some or all of the … data that is not within the training scope to identify new training data; add the new training data to the training corpus used to train the first instance of the first AI algorithm to produce a new training corpus; and retrain a second instance of the first AI algorithm using the new training corpus (see para. 0037 - selects some sentences from the out-of-distribution dataset that are most similar to at least one in-distribution sentence of the same class. In one embodiment, this can further be summarized as selecting some sentences from the auxiliary dataset that are most similar to at least one primary sentence of the same class; para. 0040 - final fine-tuned model can then be deployed and further adapted accordingly; para. 0044 - some of out-of-distribution data that are especially in the similar distribution as the in-distribution data is selected and used for training. In other words, in one embodiment, the primary data and auxiliary data into a shared semantic space, and then select some of auxiliary data that are especially in the similar distribution as the primary data and train on them; para. 0045 - then selects some sentences from the out-of-distribution data that are more similar to in-distribution data and append those to the training set; para. 0058 – foundation models is adjusted by using he augmented training dataset to train the AI engine. In one embodiment, the adjusting comprises tuning and fine tuning the foundation models, reiteratively as appropriate; Claim 1 - adjusting said foundation model by using said augmented training dataset to train said AI engine.).
Accordingly, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Park with those of Lang. One would have been motivated to do so for the purpose efficiently enhancing a training set by identifying matching or substantially matching out-of-distribution samples and adding those samples to the training set, improving model performance, as taught by Park (0037, 0044, 0045, 0058, and Claim 1).
Claim(s) 16:
Claim(s) 16 correspond to claim 6, and thus, Lang and Park teach or suggest discloses the limitations of claim(s) 16 as well.
Claim 7:
Park further teaches or suggests wherein, upon the second instance of the first AI algorithm being trained using the new training corpus, the second instance of the first AI algorithm becomes active, and the first instance of the first AI algorithm becomes inactive (see para. 0037 - selects some sentences from the out-of-distribution dataset that are most similar to at least one in-distribution sentence of the same class. In one embodiment, this can further be summarized as selecting some sentences from the auxiliary dataset that are most similar to at least one primary sentence of the same class; para. 0040 - final fine-tuned model can then be deployed and further adapted accordingly; para. 0044 - some of out-of-distribution data that are especially in the similar distribution as the in-distribution data is selected and used for training. In other words, in one embodiment, the primary data and auxiliary data into a shared semantic space, and then select some of auxiliary data that are especially in the similar distribution as the primary data and train on them; para. 0045 - then selects some sentences from the out-of-distribution data that are more similar to in-distribution data and append those to the training set; para. 0058 – foundation models is adjusted by using he augmented training dataset to train the AI engine. In one embodiment, the adjusting comprises tuning and fine tuning the foundation models, reiteratively as appropriate; Claim 1 - adjusting said foundation model by using said augmented training dataset to train said AI engine.).
Accordingly, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Park with those of Lang. One would have been motivated to do so for the purpose efficiently enhancing a training set by identifying matching or substantially matching out-of-distribution samples and adding those samples to the training set, improving model performance, as taught by Park (0037, 0044, 0045, 0058, and Claim 1).
Claim(s) 17:
Claim(s) 17 correspond to claim 7, and thus, Lang and Park teach or suggest discloses the limitations of claim(s) 17 as well.
Claim(s) 8 and 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lang, and further in view of Bagga et al., US Publication 2025/0200962 (“Bagga”).
Claim 8:
Bagga further teaches or suggests wherein the first instance of the AI algorithm is selected based on an AI algorithm type as part of an AI as a Service (AIaaS) and wherein the training scope of the first instance of the first AI algorithm is defined when selecting the first instance of the first AI algorithm based on the AI algorithm type (see para. 0003 - receiving a request associated with a plurality of AI services. Each of the plurality of AI services is associated with one or more of a plurality of ML models. At least one of the plurality of ML models is trained by predetermined guidelines. The method further includes providing for presentation, via a front end, the plurality of AI services. The method further includes receiving a user input through the front end, where the user input indicates selection of one of the plurality of AI services; para. 0026 - allows an individual module/microservice to be integrated into or taken out of the framework in a plug-and-play fashion, thereby providing flexibility and scalability. For example, it is easier to update one of the AI service modules (e.g., updating a corresponding ML model) or add a new AI service module to the system without affecting existing functions of the system; para. 0032 - system 100 is trained by predetermined guidelines. use an ML model trained by these predetermined guidelines to ensure that the generated content automatically complies with the internal company policies; para. 0033 - this framework can make it easier to update one of the AI service modules (e.g. , updating a corresponding ML model) or add a new AI service module to the system 100 without affecting existing functions of the system 100; para. 0042 - input data can be specific to the particular AI service; para. 0048 - system can replace the predetermined guidelines with new guidelines and update the at least one of the plurality of ML models by training the at least one of the plurality of ML models using the new guidelines.).
Accordingly, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Bagga with those of Lang. One would have been motivated to do so for the purpose efficiently enabling a user to customize AI services using an AI services interface and selectable training guidelines, improving model tailoring, as taught by Bagga (0026, 0032, 0033, 0042, and 0048).
Claim(s) 18:
Claim(s) 18 correspond to claim 8, and thus, Lang and Bagga teach or suggest discloses the limitations of claim(s) 18 as well.
Claim(s) 9 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lang, in view of Bagga, and further in view of Chen.
Claim 9:
Chen further teaches or suggests wherein the training scope of the first instance of the first AI algorithm is generated based on a search of a corpus database (see para. 0023 - receive the experiment request associated with a target dataset, wherein the experiment request includes the information of the target dataset for training a machine learning model. More particularly, a user may upload files of multiple datasets to a file system which can also be called Dataset Store in advance, and the description of the target dataset may be stored in the database 16. When a user would like to train a machine learning model using the target dataset, the user may input a selection instruction of the target dataset through input interface 11. The experiment generator 12 asks the file system for the file of the target dataset according to the selection instruction, and the file system accordingly searches for the description of the target dataset in the database 16. Or, the user may directly provide the file of the target dataset to the machine learning system 1 through the target dataset.).
Accordingly, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Chen with those of Lang. One would have been motivated to do so for the purpose efficiently enabling a user to configure model training including configuring the training data used, improving model tailoring, as taught by Chen (0023).
Claim(s) 10 and 19 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lang, and further in view of Banerjee et al., US Publication 2021/0201003 (“Banerjee”).
Claim 10:
Banerjee further teaches or suggests determine that the training corpus used to train the first instance of the first AI algorithm does not have enough training data for the training scope; retrieve training data that is within the training scope of the first instance of the first artificial intelligence algorithm; generate new synthetic training data that is within the training scope of the first instance of the first AI algorithm; and add the new synthetic training data to the training corpus used to train the first instance of the first AI algorithm (see Fig. 9; para. 0005 - results of the neural network training based on the augmenting with the synthetic data can be used to further train the neural network; para. 0006 - augmented with generated synthetic vectors. The generated synthetic vectors can augment sparse classes; para. 0028 – sparse class of a facial expression can differ from a more abundant class of a facial expression in that the sparse class is uncommon or difficult to identify in potential training image data; para. 0039 - synthetic data can be generated to address an imbalance or "sparsity" within a training data set. generating the synthetic data can provide or "fill in" otherwise sparse training data such that a classifier trained using the training data can be better balanced; para. 0040 - samples can be sparse or insufficient. Discussed throughout, training, using the example dataset, can result in a biased classifier model. Synthesizes artificial or synthetic samples of a particular domain; Claim 28 - generating synthetic data to augment the sparse classes … augmenting the training dataset using the synthetic data.).
Accordingly, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Banerjee with those of Lang. One would have been motivated to do so for the purpose efficiently addressing training data imbalance by generating synthetic data for a particular domain, lessening model bias and improving model performance, as taught by Banerjee (0039, 0040).
Claim(s) 19:
Claim(s) 19 correspond to claim 10, and thus, Lang and Banerjee teach or suggest discloses the limitations of claim(s) 19 as well.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Andrew T McIntosh whose telephone number is (571)270-7790. The examiner can normally be reached M-Th 8:00am-5:30pm.
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, Tamara Kyle can be reached at 571-272-4241. 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.
/ANDREW T MCINTOSH/Primary Examiner, Art Unit 2144
Prosecution Insights