DETAILED ACTION
This action is in response to the amendment filed 12/22/2025.
Claims 1-7 and 9-22 are currently pending.
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 .
Response to Arguments
Applicant's arguments directed to the 35 U.S.C. 101 Applicant’s arguments, see pages 7 and 8 of amendment, filed 12/22/2025, with respect to Claim Rejections under 35 U.S.C. 101 have been fully considered and are persuasive. The rejections of claims 1-7 and 9-22 under 35 U.S.C. 101 have been withdrawn.
Applicant’s arguments, see pages 8 and 9 of amendment, filed 7/11/2025, with respect to the rejection(s) of claim(s) 1 under 35 U.S.C. 103 have been fully considered and are persuasive. Therefore, the rejection has been withdrawn. However, upon further consideration, a new ground(s) of rejection is made in view of Ghahramani et al. (US (2018/0114113), hereinafter Ghahramani and Fujimaki et al. (US 2013/0325782), hereinafter Fujimaki.
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) 1-6, 11, 12, and 18-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Anunciacao et al. (US 2020/0286095), hereinafter Anunciacao, in view of Meek et al. (US 6,728,690), hereinafter Meek, in view of Fujimaki in view of Ghahramani.
As per claim 1, Anunciacao teaches the following:
a method, comprising:
receiving input data. As Anunciacao teaches in paragraph [0047], a transaction is classified, thus “received”;
providing the received input data to a trained discriminative machine learning model to determine an inference result. As Anunciacao teaches in paragraph [0051], a machine-learning model may be selected based on a classification of the input (transaction);
using at least a portion of the received input data to determine a utility measure. As Anunciacao teaches in paragraph [0047], a classification for the input is determined based upon a similarity measure (portion of input); and
using the version of the determined inference result and the utility measure as inputs to a decision module optimizing one or more decision metrics to determine a decision result. As Anunciacao teaches in paragraph [0047], the transaction is classified as either fraudulent or genuine by using labeling information.
However, Anunciacao does not explicitly teach of utilizing class posterior probability estimates to compensate for miscalibrations. In a similar field of endeavor, Meek teaches of a method of classifiers (see abstract). Meek further teaches the following:
correcting a class posterior probability estimate to determine a version of the determined inference result and compensate for miscalibration of the trained discriminative machine learning model including by adjusting a value of one or more decision metrics As Meek teaches in column 12, lines 18-38, and corresponding Fig. 6, a probability transducer acts as a small neural network, where a probability scoring function compares a desired output to a probability output. The mismatch between the desired and probability (miscalibration) is fed to an iterative optimization algorithm, which produces a new set of parameters (adjusting decision metrics to determine a version of value) to minimize the mismatch (correcting the estimate).
outputting the class posterior probability estimate. As Meek shows in Fig. 6, the probability from the probability transducer 625 is “output” to a probability scoring function.
It would have been obvious to one of ordinary skill in the art before the effective filing date of applicant’s claimed invention to have modified the classification method Anunciacao with the probability transducer of Meek. One of ordinary skill would have been motivated to have made such modification because it would benefit a user of Anunciacao in creating probable results closer to user desired results.
Furthermore, Anunciacao does note explicitly teach of maximizing a first metric while fixing a second metric. In a similar field of endeavor, Fujimaki teaches of variable model estimation using posterior probabilities (see paragraph [0002]. Fujimaki further teaches the following:
wherein optimizing the one or more decision metrics includes maximizing a first metric while fixing a second metric. As Fujimaki teaches in paragraph [0055], a optimization unit optimizes a model by maximizing one variable (G), while other variables (q and q.about.) are fixed.
It would have been obvious to one of ordinary skill in the art before the effective filing date of applicant’s claimed invention to have modified the optimization of Anunciacao with the fixing and maximizing of values of Fujimaki. One of ordinary skill would have been motivated to have made such modification because as Fujimaki teaches in paragraph [0055], such single value maximization benefits users in optimizing each component separately without considering combinations and avoid combinatorial explosion.
Furthermore, Anunciacao in view of Meek does not explicitly teach of adjusting a value of a decision metric without re-training the machine learning model. In a similar field of endeavor, Ghahramani teaches of a method utilizing trained computer models (see abstract). Ghahramani further teaches in the abstract of the model including an direct and indirect network, where the indirect network generates expected weights for nodes and layers of the direct network. Ghahramani further teaches in paragraph [0034], that when portions of the direct network no longer describe current data well, the indirect network can provide a starting place for generating weights without complete re-training of the direct network.
It would have been obvious to one of ordinary skill in the art before the effective filing date of applicant’s claimed invention to have modified the value adjustments of Anunciacao in view of Meek with the direct and indirect networks of Ghahramani. One of ordinary skill would have been motivated to have made such modification because as Ghahramani teaches in paragraph [0034], use of an indirect network may benefit users when certain portions of data may be missing.
Regarding claim 2, Anunciacao teaches the method of claim 1 as described above. Anunciacao further teaches the following:
the input data includes information associated with a transaction being analyzed for detection of fraud, money laundering, account takeover, inappropriate account opening, or other non-legitimate account activity behavior. See paragraph [0003].
Regarding claim 3, Anunciacao teaches the method of claim 1 as described above. Anunciacao further teaches the following:
the trained discriminative machine learning model is configured to perform a binary classification task. As Anunciacao teaches in paragraph [0047], the transaction is classified as either fraudulent or genuine by using labeling information, where fraud or genuine are interpreted as being a binary classification, i.e., one or the other
Regarding claim 4, Anunciacao teaches the method of claim 1 as described above. Anunciacao further teaches the following:
the trained discriminative machine learning model has been trained utilizing training data that includes information associated with a plurality of transactions, including, for each transaction of the plurality of transactions, a set of labeled transaction-related features and a labeled outcome as to whether fraudulent activity is present. As Aununciacao teaches in the abstract, “each training transaction being associated with labelling information that indicates whether the training transaction is either genuine or fraudulent”.
Regarding claim 5, Anunciacao teaches the method of claim 1 as described above. Anunciacao further teaches the following:
the inference result is a probability estimate. As Anunciacao teaches in paragraph [0050], the output of a machine-learning model may provide a numerical value that represents a probability that the transaction is fraudulent.
Regarding claim 6, Anunciacao teaches the method of claim 1 as described above. Anunciacao further teaches the following:
the utility measure is associated with a rate at which the trained discriminative machine learning model correctly predicts a positive class associated with the received input data. As Anunciacao teaches in paragraph [0043], a performance metrics may be determined based upon different machine-learning models, where the performance metric may be that of classification accuracy, i.e. rate at which classification is performed properly. The classification algorithm may then be selected based upon the performance metric. Therefore, the limitation of “providing the received input data to a trained discriminative machine learning model to determine an inference result” and the utility measure may be interpreted as Anunciacao’s teaching of utilizing the input and performance metric to select a classification algorithm, then submit the input into the classification’s machine-learning model.
Regarding claim 11, Anunciacao teaches the method of claim 1 as described above. Anunciacao further teaches the following:
the decision module includes a scoring function component that outputs a score based at least in part on the version of the determined inference result and the utility measure. As Anunciacao teaches in paragraph [0050], the output of a machine-learning model may provide a numerical value (i.e. score) that represents a probability that the transaction is fraudulent.
Regarding claim 12, Anunciacao teaches the method of claim 1 as described above. Anunciacao further teaches the following:
the one or more decision metrics includes a constraint associated with one of the following: a false positive rate, an alert rate, or a precision metric that is based on true positive and false positive measures. As Anunciacao teaches in paragraph [0031], the selection of a cluster is utilized to prevent false positives, which is interpreted as being “associated with” a false positive rate.
Regarding claim 18, Anunciacao teaches the method of claim 1 as described above. Anunciacao further teaches the following:
the decision result is a selection of one of two possible outcomes for the received input data. As Anunciacao teaches in paragraph [0047], the transaction is classified as either fraudulent or genuine by using labeling information.
As per claim 19, Anunciacao teaches the following:
A system, comprising:
one or more processors. See paragraph [0012].
The remaining limitations of claim 18 are substantially similar to those of claim 1 and are rejected using the same reasoning.
As per claim 20, Anunciacao teaches the following:
a computer program product embodied in a non-transitory computer readable medium and comprising computer instructions. See paragraph [0012].
The remaining limitations of claim 20 are substantially similar to those of claim 1 and are rejected using the same reasoning.
Claim(s) 7 and 13-16 is/are rejected under 35 U.S.C. 103 as being unpatentable over Anunciacao in view of Meek in view of Fujimaki in view of Ghahramani as applied to claim 1, and further in view of Adjaoute (US 2002/0133721).
Regarding claim 7, Anunciacao teaches the method of claim 1 as described above. While Anunciacao teaches in paragraph [0006] of utilizing relations between attributes of input transactions, Anunciacao does not explicitly teach of the input being associated with a monetary amount. In a similar field of endeavor, Adjaoute teaches of a method of utilizing intelligent systems to detect fraud (see abstract). Adjaoute further teaches the following:
the utility measure is associated with a monetary amount associated with the received input data. As Adjaoute teaches in paragraph [0019], a neural network for detecting fraud may take into account several variables of a transaction, including “transaction amount”.
It would have been obvious to one of ordinary skill in the art at the time the application was filed to have modified the utility measure of Anunciacao with the transaction amount parameter of Adjaoute. One of ordinary skill would have been motivated to have made such modification because while Anunciacao does not explicitly state the use of monetary amounts in transactions, the parameter is suggested in Anunciacao’s desire to detect monetary fraud (see paragraph [0003]).
Regarding claim 13, Anunciacao teaches the method of claim 1 as described above. While Anunciacao teaches in paragraph [0043] of comparing different classification model results, Anunciacao does not explicitly teach of comparing results to a threshold. In a similar field of endeavor, Adjaoute teaches of a method of utilizing intelligent systems to detect fraud (see abstract). Adjaoute further teaches the following:
the decision module optimizing the one or more decision metrics includes a component comparing a value based on the version of the determined inference result and the utility measure with a specified threshold. As Adjaoute teaches in paragraph [0027], a transaction is given an overall fraud risk score which is compared to a threshold to determine whether the transaction is fraudulent or not.
It would have been obvious to one of ordinary skill in the art at the time the application was filed to have modified the scores of Anunciacao with the threshold values of Adjaoute. One of ordinary skill would have been motivated to have made such modification because threshold values were known to benefit systems in allowing classifications of scores based on severity.
Regarding claim 14, Anunciacao teaches the method of claim 1 as described above. However, as described above, Anunciacao does not explicitly teach of comparing results to a threshold. Adjaoute further teaches the following:
the specified threshold is adapted to a type of constraint associated with optimizing the one or more decision metrics. As Adjaoute teaches in paragraph [0143], constraints may be specified about which combinations of values are allowed and which are not, thus setting a threshold. Further see paragraph [0146] where which constraints are determined to apply to a transaction.
It would have been obvious to one of ordinary skill in the art at the time the application was filed to have modified the scores of Anunciacao with the threshold values of Adjaoute. One of ordinary skill would have been motivated to have made such modification because threshold values were known to benefit systems in allowing classifications of scores based on severity.
Regarding claim 15, Anunciacao teaches the method of claim 1 as described above. Anunciacao further teaches the following:
the one or more decision metrics include both a metric associated with correctly predicting a positive class associated with the received input data As Anunciacao teaches in paragraph [0043], a performance metrics may be determined based upon different machine-learning models, where the performance metric may be that of classification accuracy, i.e. rate at which classification is performed properly. The classification algorithm may then be selected based upon the performance metric. Therefore, the limitation of “providing the received input data to a trained discriminative machine learning model to determine an inference result” and the utility measure may be interpreted as Anunciacao’s teaching of utilizing the input and performance metric to select a classification algorithm, then submit the input into the classification’s machine-learning model.
While Anunciacao teaches in paragraph [0006] of utilizing relations between attributes of input transactions, Anunciacao does not explicitly teach of the input being associated with a monetary amount. In a similar field of endeavor, Adjaoute teaches of a method of utilizing intelligent systems to detect fraud (see abstract). Adjaoute further teaches in paragraph [0019], a neural network for detecting fraud may take into account several variables of a transaction, including “transaction amount”.
It would have been obvious to one of ordinary skill in the art at the time the application was filed to have modified the utility measure of Anunciacao with the transaction amount parameter of Adjaoute. One of ordinary skill would have been motivated to have made such modification because while Anunciacao does not explicitly state the use of monetary amounts in transactions, the parameter is suggested in Anunciacao’s desire to detect monetary fraud (see paragraph [0003]).
Regarding claim 16, modified Anunciacao teaches the method of claim 15 as described above. However, Anunciacao does not explicitly teach of formulating a score with the parameters’ respect for one another. Adjaoute further teaches the following:
the metric associated with correctly predicting the positive class and the metric associated with the monetary amount are formulated with respect to each other in terms of a parameterized scoring function. As Adjaoute teaches in paragraph [0019], fraud may be determined based upon the relationships among many variables, i.e. the relationship of parameters with respect to one another.
It would have been obvious to one of ordinary skill in the art at the time the application was filed to have modified the anomaly detection of Anunciacao with the parameter relationships of Adjaoute. One of ordinary skill would have been motivated to have made such modification because as Adjaoute teaches in paragraph [0019], such relationships may beneficially enable the estimation of probability in determination of fraud.
Claim(s) 9, 11 and 17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Anunciacao in view of Meek in view of Fujimaki in view of Ghahramani as applied to claim 1, and further in view of Patton et al. (US 10,209,974), hereinafter Patton.
Regarding claim 9, modified Anunciacao teaches the method of claim 1 as described above. However, Anunciacao does not explicitly teach of the correction to a probability estimate being a disparity between values. Patton further teaches the following:
the correction to the probability estimate is associated with a disparity between a rate of occurrence of a data class in training data utilized to train the discriminative machine learning model and the rate of occurrence of the data class in data upon which the discriminative machine learning model operates after it is deployed. Patton gives several examples of how the concept drift parameter may be determined in column 16, lines 3-12, including “an anticipated class occurrence frequency”.
It would have been obvious to one of ordinary skill in the art at the time the application was filed to have modified the classification of Anunciacao to include the drift parameter of Patton. One of ordinary skill would have been motivated to have made such modification because as Patton teaches in column 4, lines 11-36, models may go obsolete over time as concepts change.
Regarding claim 17, modified Anunciacao teaches the method of claim 1 as described above. However, Anunciacao does not explicitly teach of maximizing true positives while maintaining a threshold false positive. In a similar field of endeavor, Patton teaches of a method of managing learning models (see abstract). Patton further teaches the following:
the decision module optimizing the one or more decision metrics includes a component maximizing a specified true positive rate of the trained discriminative machine learning model while maintaining a specified false positive rate of the trained discriminative machine learning model below a specified threshold. As Patton teaches in column 21, lines 26-53), models are chosen based upon challenging the models against one another for better “true results” (maximize true positive). Patton teaches in column 12, lines 1-9, a run condition may be when a model outputs false results at a rate exceeding a predetermined threshold (false positive rate must remain below threshold). Further see column 11, lines 38-59.
It would have been obvious to one of ordinary skill in the art at the time the application was filed to have modified the fraud detection models of Anunciacao with the true/false positives of Patton. One of ordinary skill would have been motivated to have made such modification because as Patton teaches in column 4, lines 11-36, models may go obsolete over time as concepts change.
Claim(s) 10 is/are rejected under 35 U.S.C. 103 as being unpatentable over Anunciacao in view of Meek in view of Fujimaki in view of Ghahramani as applied to claim 1, and further in view of Chang (US 2005/0265607).
Regarding claim 10, modified Anunciacao teaches the method of claim 1 as described above. However, Anunciacao does not explicitly teach of the correction to a probability estimate being one of Platt scaling or isotonic regression. In a similar field of endeavor, Chang teaches a method of classifying information (see abstract). Chang teaches the following:
the correction to the probability estimate includes at least one of Platt scaling or isotonic regression. As Chang teaches in paragraph [0073], Platt’s formula is utilized to convert support vector machines, utilized as base classifiers, to posterior probabilities.
It would have been obvious to one of ordinary skill in the art at the time the application was filed to have modified the classification using probability estimates of Anunciacao in view of Meek to utilize Platt’s formula such as in Chang. One of ordinary skill would have been motivated to have made such modification because Platt’s formula benefits users by improving accuracy of probability estimates by transforming raw model scores into more reliable probabilities.
The examiner would like to further note that the Meek reference lists John C. Platt as an inventor. The individual who created “Platt scaling”. Thus providing further motivation for the modification.
Claim(s) 21 is/are rejected under 35 U.S.C. 103 as being unpatentable over Anunciacao in view of Meek in view of Fujimaki in view of Ghahramani as applied to claim 1, and further in view of Karmakar et al. (US 2020/0349575), hereinafter Kamakar.
Regarding claim 21, modified Anunciacao teaches the method of claim 1 as described above. While Anunciacao teaches of weighting nodes representing neurons in paragraph [0040], Anunciacao does not explicitly teach of weighting decision metrics. In a similar field of endeavor, Kamakar teaches of identifying fraudulent transactions utilizing machine learning (see abstract and paragraph [0039]). Kamakar further teaches the following:
the one or more decision metrics includes at least two decision metrics and a parameterized scoring function maximizes a first one of the at least two decision metrics and assigns a specified weight to a second one of the at least two decision metrics. As Kamakar teaches in paragraphs [0022] – [0033], several example decision metrics are described. Kamakar teaches in paragraph [0039] that weights may be assigned to the different metrics. As Kamakar describes, a certain metric with low variance may have a high weight such that any deviation from the norm may signal fraud (i.e., maximized metric), while other metrics may be given lower weights.
It would have been obvious to one of ordinary skill in the art at the time the application was filed to have modified the machine learning of Anunciacao with the maximizing of metrics of Kamakar. One of ordinary skill would have been motivated to have made such modification because multiple metrics benefit users in allowing different parameters of transactions to be correlated and patterns formed for identifying fraud. Furthermore, the maximization of weight of certain metrics would benefit users in more easily identifying fraudulent activity when a value varies from the norm.
Claim(s) 22 is/are rejected under 35 U.S.C. 103 as being unpatentable over Anunciacao in view of Meek in view of Fujimaki in view of Ghahramani as applied to claim 1, and further in view of Yang (US 2016/0259896).
Regarding claim 22, modified Anunciacao teaches the method of claim 1 as described above. However, neither Anunciacao nor Meek explicitly teach of presenting the probability estimate on an interface. In a similar field of endeavor, Yang teaches of a method of identifying fraud (see abstract). Yang further teaches the following:
outputting the class posterior probability estimate includes presenting the class posterior probability estimate on a user interface to provide insight into a decision-making process associated with the decision module. As Yang teaches in paragraphs [0028] and [0029], and corresponding Fig. 7, calculated probabilities may be displayed to a user.
It would have been obvious to one of ordinary skill in the art at the time the application was filed to have further modified the probability estimate of Anunciacao in view of Meek with the display of Yang. One of ordinary skill would have been motivated to have made such further modification because as Yang teaches in paragraph [0005], such display benefits users in allowing the user to visualize the chances of fraud and fraudulent patterns.
The examiner would like to further note that simply “displaying data” which has been calculated by a computer would fall under generic computer practices. Displaying calculated data is a well known and established technique in the computer science field.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to GREGORY A DISTEFANO whose telephone number is (571)270-1644. The examiner can normally be reached Monday - Friday: 9 am - 5 pm.
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, William Bashore can be reached at 5712424088. 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.
/GREGORY A. DISTEFANO/
Examiner
Art Unit 2174
/WILLIAM L BASHORE/ Supervisory Patent Examiner, Art Unit 2174
Prosecution Insights