DETAILED ACTION
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 .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 02/02/2026 has been entered.
Examiner’s Note
In regards to the 35 USC § 101 rejection, has been withdrawn in light of the instant amendments to the claim because the amended claim directed to improving how machine-learning models are trained and operate under time-varying data distributions by employing a transformer- discriminator architecture in which scaled feature representations are explicitly compared to reference scaled features derived from a defined prior time period, and discriminator feedback is used to mitigate temporal distribution mismatch. These features recite a concrete, technical approach to improving classifier stability and accuracy under non-stationary data conditions.
Applicant’s arguments (pgs. 10 – 11) with respect to amended claim(s) have been considered but are moot, because arguments/remarks are directed to amended claim limitations that were not previously examined by the examiner. The rejections are noted in the current office action to address amended claim limitations.
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.
21. Claim(s) 1 – 6 and 14 – 19 are rejected under 35 U.S.C. 103 as being unpatentable over Husein et al., "Generative adversarial networks time series models to forecast medicine daily sales in hospital", in view of Luus et al., Pub. No.: US20210334403A1, Clark et al., Pub. No.: US20220230276A1 and Song et al., Pub. No.: US20200134804A1.
22. Regarding claim 1, Husein teaches: A system of training a machine-learning classifier that accounts for variability in data distributions over time, comprising:
a processor programmed to: access features from a first dataset of available data, the first dataset relating to a first time period ending on a first date;
(Husein, page: 4, section IV.2 “Our goal is to predict six factors and get the stock of supplies needed to optimize the planning of drug purchases in the next one week through historical data on medicine use records [access features from a first dataset of available data]. Data is separated into two parts for training and testing models. We chose the data for the last 4 years (2015-2018) as training data [the first dataset relating to a first time period ending on a first date], and data 2019 for testing. The time period was used as input for GAN. Each weight of GAN training activated in the last layer used representation of discriminator features to generation distribution as a new representation of data.”)
train a transformer to scale the features;
(Husein, page: 3, section IV.1, “In this section an explanation of the results of implementing the GAN model [train a transformer] (i.e.: generator in GAN architecture is a transformer) for forecasting drug sales, as the main purpose of this paper is to build a model that can forecast drug sales a week (7th day) in future [to scale the features], for this purpose, we use stock card record data drugs use consisting of 20 tables. This data contains 58,819 patients from 82 care units with a record of medicinuse transactions totaling 163,739 for inpatients and 262,140 for outpatients. From this data, we divide into three datasets, namely drug data (Drug Code, Drug Name, Type, Unit), sales data (Date, Qty, Drug Code, ICD Code, Sales Price) and purchase data (Date, Drug Code, Qty, Purchase Price) with focuses on seven forecasting indicator variables, namely the transaction date, drug code, ICD Code, sales price, purchase price, usage amount and total drug purchase.”)
based on a transformer-discriminator architecture
(Husein, page: 4, section IV.2, “IV.2 Model Training and Testing Our goal is to predict six factors and get the stock of supplies needed to optimize the planning of drug purchases in the next one week through historical data on medicine use records. Data is separated into two parts for training and testing models. We chose the data for the last 4 years (2015-2018) as training data, and data 2019 for testing. The time period was used as input for GAN. Each weight of GAN training activated in the last layer used representation of discriminator features to generation distribution as a new representation of data [based on a transformer-discriminator architecture]. The results of the GAN model training are shown in Figure 2 and Figure 3”)
access a plurality of labels derived from a second dataset of the available data, the second dataset relating to a second time period starting after the first date; and
(Husein, page: 4, section IV.2, “Our goal is to predict six factors and get the stock of supplies needed to optimize the planning of drug purchases in the next one week through historical data on medicine use records. Data is separated into two parts for training and testing models. We chose the data for the last 4 years (2015-2018) as training data, and data 2019 for testing [access a plurality of labels derived from a second dataset of the available data, the second dataset relating to a second time period starting after the first date]. The time period was used as input for GAN. Each weight of GAN training activated in the last layer used representation of discriminator features to generation distribution as a new representation of data.”)
Husein does not teach:
generate a classifier that classifies input data based on the plurality of labels
in which the transformer adjusts feature weights based on discriminator scores from a discriminator to reduce loss resulting from variation in data distribution of the available data;
generate …the scaled features from the trained transformer.
(iii) the transformer adjusts the feature weights based on discriminator scores generated from the comparison to mitigate distribution mismatch between feature representations across time
(ii) the discriminator compares the scaled features to reference scaled features derived from data corresponding to the first time period;
Luus teaches:
generate a classifier that classifies input data based on the plurality of labels.
(Luus, [0136] In one example embodiment, the system further comprises a classifier 412, the generator, the discriminator, the privacy adversary, and the classifier being in data communication, wherein the classifier 412 [generate a classifier that classifies input data] is configured to be trained to indicate that the generated records 408 are plausible, based on conditioning, and to generate a class label [based on the plurality of labels] as input to the generator 420. In one example embodiment, the system further comprises a classifier 412, the generator, the discriminator, the privacy adversary, and the classifier being in data communication, wherein the system is configured to sequentially train the generator 420, the discriminator 424, the classifier 412, and the privacy adversary 436. In one example embodiment, the discriminator 424 is configured to be trained to detect that a given generated record 408 is distinguished from the training records 416 and that a given training sample originates from the training records 416. In one example embodiment, the discriminator 424 is configured to be trained based on the discriminator 424 configured with output of the generator as input, and the discriminator 424 configured with a given training sample as input.”)
Luus and Husein are related to the same field of endeavor (i.e.: generative adversarial networks architecture). It would have been obvious to a person having ordinary skill in the art before the effective filing date of the invention to combine the teachings of Luus with teachings of Husein to adds privacy-preserving synthetic data generation to classify card members based on interchange fees, ensuring accurate predictions without exposing sensitive cardholder data (Luus, Abstract).
Husein in view of Luus do not teach:
in which the transformer adjusts feature weights based on discriminator scores from a discriminator to reduce loss resulting from variation in data distribution of the available data;
generate …the scaled features from the trained transformer.
(iii) the transformer adjusts the feature weights based on discriminator scores generated from the comparison to mitigate distribution mismatch between feature representations across time
(ii) the discriminator compares the scaled features to reference scaled features derived from data corresponding to the first time period;
Clark teaches:
in which the transformer adjusts feature weights based on discriminator scores from a discriminator
(Clark, “[0044] Updating the weights for the generator network may comprise forming a combined discriminator score based on the first and second discriminator scores and updating the weights for the generator network [in which the transformer adjusts feature weights] based on the combined discriminator score [based on discriminator scores from a discriminator].”)
to reduce loss resulting from variation in data distribution of the available data;
(Clark, “[0070] Each of the generator and discriminator may be implemented via neural networks, wherein their output is determined through parameters of the respective neural network. These neural networks can be trained by adjusting the parameters to improve the result of an objective function (e.g. to reduce prediction error) [to reduce loss resulting from variation in data distribution of the available data].”)
generate …the scaled features from the trained transformer
(Clark, “[0044] Updating the weights [generate …the scaled features from the trained transformer] for the generator network may comprise forming a combined discriminator score based on the first and second discriminator scores and updating the weights for the generator network based on the combined discriminator score.”)
(iii) the transformer adjusts the feature weights based on discriminator scores generated from the comparison to mitigate distribution mismatch between feature representations across time:
(Clark, “[0042] training a discriminator network according any of the discriminator training methods described herein utilizing the generated sequence of images as the input sequence of images for the discriminator network; and updating weights for the generator network [(iii) the transformer adjusts the feature weights] based on the first discriminator score and the second discriminator score [based on discriminator scores generated from the comparison].”)
to mitigate distribution mismatch between feature representations across time
(Clark, “[0018] … This might be based on corresponding loss functions for the spatial and temporal discriminator network. Each discriminator network may be trained based on an objective function that aims to adjust (optimize) the parameters of the corresponding discriminator network to more accurately classify the image(s) input into the discriminator network as either generated by the generator network or not generated by the generator network (e.g. “real” images of an environment as opposed to generated images). The generator network might be trained with an objective function that aims to cause each discriminator network to misclassify the generated sequence of images [to mitigate distribution mismatch between feature representations across time]. The generator network may be configured to generate sequences of images based solely on learned distributions without relying on any predefined prior distributions for foreground, background or motion (e.g. without making use of any predefined models for optical flow).”)
Clark, Husein and Luus are related to the same field of endeavor (i.e.: generative adversarial networks architecture). It would have been obvious to a person having ordinary skill in the art before the effective filing date of the invention to combine the teachings of Clark with teachings of Husein and Luus to refine predictions dynamically as new data becomes available to enable more accurate classification (Clark, ¶[0004]).
Husein in view of Luus and Clark do not teach:
(ii) the discriminator compares the scaled features to reference scaled features derived from data corresponding to the first time period;
Song teaches:
(ii) the discriminator compares the scaled features to reference scaled features derived from data corresponding to the first time period: and
(Song, “[0042]… The generator 310 takes past frames 305 as inputs and generates a (for example, predicted) future frame 315. The future frame 315 can include an extrapolated frame consistent with a sequence of past frames. For example, the generator 310 can extract spatio-temporal features of video frames from t−T to t−1 efficiently and generate a future frame at time t [the scaled features to reference scaled features derived from data corresponding to the first time period].
[0043]The image discriminator 325 [the discriminator compares] discriminates the generated frame and real frame. For example, the image discriminator 325 can output a single number, for example, between 0 and 1, representing “how realistic” the image looks. The image discriminator 325 can include a (for example, frame level in a frame by frame video) convolutional filter generated to classify images as true or false.”)
Song, Husein, Luus and Clark are related to the same field of endeavor (i.e.: generative adversarial networks architecture). It would have been obvious to a person having ordinary skill in the art before the effective filing date of the invention to combine the teachings of Song with teachings of Husein, Luus and Clark to improve detection accuracy and robustness (Song, Abstract).
23. Claim 14 recites analogous limitation as claim 1, so is rejected under the same rationale.
Regarding claim 2, Husein in view of Luus, Clark and Song teach the method of claim 1.
Husein further teaches: wherein to train the transformer, the processor is further programmed to implement a discriminator that operates in an adversary manner with the transformer
(Husein, page: 2 – 3, section III, “In this section we describe the proposed method using the Generative Adversarial Networks (GAN) framework [that operates in an adversary manner with the transformer] approach for forecasting sales data in hospitals. GAN is introduced by Goodfellow [42] as deep learning model framework for capturing the distribution of training data by generating new data from the same distribution using generator and discriminator models. Architecture GAN learns unsupervised features with a competitive learning process. GAN will produce more feature space that may be exploited, thereby reducing the potential for excess features during training. The G (Generator) model is trained to produce data that looks like sales data from the target stock, while the D model (Discriminator) [implement a discriminator] is trained to tell the difference between data from the Generator and real data. Errors from D are used to train G to defeat D.”)
Clark further teaches: to adjust feature weights of the transformer
(Clark, [0018] Varying weights of the discriminator network might comprise varying weights of the spatial discriminator network based on the first discriminator score and varying weights of the temporal discriminator network based on the second discriminator score. That is, the spatial discriminator network and the temporal discriminator network might be trained independently of each other based on their corresponding discriminator scores. This might be based on corresponding loss functions for the spatial and temporal discriminator network. Each discriminator network [of the transformer] may be trained based on an objective function that aims to adjust (optimize) [to adjust feature weights] the parameters of the corresponding discriminator network to more accurately classify the image(s) input into the discriminator network as either generated by the generator network or not generated by the generator network (e.g. “real” images of an environment as opposed to generated images). The generator network might be trained with an objective function that aims to cause each discriminator network to misclassify the generated sequence of images. The generator network may be configured to generate sequences of images based solely on learned distributions without relying on any predefined prior distributions for foreground, background or motion (e.g. without making use of any predefined models for optical flow).”)
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Clark with teachings of Husein, Luus and Song for the same reasons disclosed for claim 1.
25. Claim 15 recites analogous limitation as claim 2, so is rejected under the same rationale.
Regarding claim 3, Husein in view of Luus, Clark and Song teach the method of claim 2.
Husein further teaches: wherein the transformer is to: generate a scaled representation of the features based on the feature weights; and
(Husein, page: 4, section IV.2, “Our goal is to predict six factors and get the stock of supplies needed to optimize the planning of drug purchases in the next one week through historical data on medicine use records. Data is separated into two parts for training and testing models. We chose the data for the last 4 years (2015-2018) as training data, and data 2019 for testing. The time period was used as input for GAN. Each weight of GAN training activated in the last layer used representation of discriminator features [based on the feature weights] to generation distribution as a new representation of data [generate a scaled representation of the features].”)
wherein the discriminator is to: compare the scaled representation of the features with reference scaled features corresponding to the first dataset;
(Husein, page: 2, “In this paper, we propose the GAN model for forecasting drug sales using generators and discriminators [wherein the discriminator is to] GAN trained on drug sales history data. GAN model training is trained in drug sales history data, each stock data is normalized [the scaled representation of the features], a time period of six days is constructed as input for GAN. After the GAN has finished training, the activated weight of the last layer puts the convolution used as a new representation of the data. Testing Data held in the training phase is run through the discriminator GAN section and the activated weights of the last convolutional layer are extracted. The extracted features are then classified to get predictive results. The model classification results will be evaluated [compare] with actual data [with reference scaled features corresponding to the first dataset] indicators using evaluating Mean Absolute Error (MAE), Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE).”)
Clark further teaches: generate the discrimination scores based on the comparison, each discrimination score indicating a level of difference between a scaled representation of a feature from among the scaled representation of the features and a corresponding reference scaled feature among the reference scaled features;
(Clark, “[0010] forming, from the input sequence [a corresponding reference scaled feature among the reference scaled features], a first set of one or more images having a lower temporal resolution than the input sequence, and inputting the first set into the spatial discriminator network to determine, based on the spatial features of each image in the first set [each discrimination score indicating a level of difference between a scaled representation of a feature from among the scaled representation of the features] (i.e.: discriminator score represents a probability of the input sequence being generated, which effectively measures the level of difference between the generated and real data), a first discriminator score [generate the discrimination scores] (i.e.: this score reflects the level of difference or similarity between the generated and real (reference) data) representing a probability that the input sequence has been generated by the generator network; [based on the comparison]”).
provide the discrimination scores to the transformer, wherein the transformer is to adjust the feature weights based on the discrimination scores to adjust generation of the scaled representation of the features.
(Clark, “[0044] Updating the weights for the generator network [to the transformer] may comprise forming a combined discriminator score based on the first and second discriminator scores [provide the discrimination scores] and updating the weights for the generator network based on the combined discriminator score [wherein the transformer is to adjust the feature weights based on the discrimination scores to adjust generation of the scaled representation of the features].)
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Clark with teachings of Husein, Luus and Song for the same reasons disclosed for claim 1.
Claim 16 recites analogous limitation as claim 3, so is rejected under the same rationale.
Regarding claim 4, Husein in view of Luus, Clark and Song teach the method of claim 3.
Clark further teaches: wherein the transformer is to adjust feature weights until one or more of the discrimination scores are each within a threshold level of error.
(Clark, “[0070] Each of the generator and discriminator may be implemented via neural networks, wherein their output is determined through parameters of the respective neural network. These neural networks can be trained by adjusting the parameters to improve the result of an objective function (e.g. to reduce prediction error) [wherein the transformer is to adjust feature weights until one or more of the discrimination scores are each within a threshold level of error].”)
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Clark with teachings of Husein, Luus and Song for the same reasons disclosed for claim 1.
Claim 17 recites analogous limitation as claim 4, so is rejected under the same rationale.
Regarding claim 5, Husein in view of Luus, Clark and Song teach the method of claim 3.
Husein further teaches: wherein to train the transformer, the processor is further programmed to: train the transformer and the discriminator to generate the scaled representation of the features;
(Husein, page: 3, “The G (Generator) model is trained [train the transformer] to produce data [to generate the scaled representation of the features] that looks like sales data from the target stock, while the D model (Discriminator) is trained [and the discriminator] to tell the difference between data from the Generator and real data. Errors from D are used to train G to defeat D. The competition between G and D forces D to randomly distinguish from real variability, formally GAN solves the min-max game with the following equation:”)
Luus further teaches: train the classifier after the transformer and the discriminator are trained.
(Luus, [0136] In one example embodiment, the system further comprises a classifier 412, the generator, the discriminator, the privacy adversary, and the classifier being in data communication, wherein the classifier 412 is configured to be trained to indicate that the generated records 408 are plausible, based on conditioning, and to generate a class label as input to the generator 420. In one example embodiment, the system further comprises a classifier 412, the generator, the discriminator, the privacy adversary, and the classifier being in data communication, wherein the system is configured to sequentially train the generator 420, the discriminator 424, the classifier 412 [train the classifier after the transformer and the discriminator are trained], and the privacy adversary 436. In one example embodiment, the discriminator 424 is configured to be trained to detect that a given generated record 408 is distinguished from the training records 416 and that a given training sample originates from the training records 416. In one example embodiment, the discriminator 424 is configured to be trained based on the discriminator 424 configured with output of the generator as input, and the discriminator 424 configured with a given training sample as input.”)
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Luus with teachings of Husein, Clark and Song for the same reasons disclosed for claim 1.
Claim 18 recites analogous limitation as claim 5, so is rejected under the same rationale.
Regarding claim 6, Husein in view of Luus, Clark and Song teach the method of claim 3.
Luus further teaches: wherein to train the transformer, the processor is further programmed to: train the transformer, the discriminator, and the classifier simultaneously.
(Luus, “[0173] In one example embodiment, the training records 416 and the reference records 440 originate from a same larger common database 428 and the reference records 440 are composed of data excluded from the training records 416. [0174] In one example embodiment, the system further comprises a classifier 412, and the generator, the discriminator, the privacy adversary, and the classifier are in data communication. The generator 420 is configured to be trained during a generator update phase 604 based on the discriminator 424, the classifier 412, and the privacy adversary 436 [wherein to train the transformer, the processor is further programmed to:
train the transformer, the discriminator, and the classifier simultaneously] (i.e.: the generator, discriminator and classifier are involved in different update phases where they interact with each other. Specifically, the generator is trained based on the discriminator, classifier, and privacy adversary, while the classifier and discriminator are also trained in their respective update phases). The discriminator 424 is configured to be trained during a discriminator update phase 608 based on the generator 420. The classifier 412 is configured to be trained during a classifier update phase 612 based on the generator 420 and the privacy adversary 436. The privacy adversary 436 is configured to be trained during an attack model update phase 616 based on the generator 420 and the classifier 412.”)
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Luus with teachings of Husein, Clark and Song for the same reasons disclosed for claim 1.
Claim 19 recites analogous limitation as claim 6, so is rejected under the same rationale.
Claim(s) 7 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Husein in view of Luus, Clark, Song and in further view of Wang et al., Pub. No.: US20220343638A1.
Regarding claim 7, Husein in view of Luus, Clark and Song teach the method of claim 2.
Husein further teaches: [training data] derived from the available data based on a second set of available data that is less in quantity than the available data
(Husein, page: 4, section IV.2 “Our goal is to predict six factors and get the stock of supplies needed to optimize the planning of drug purchases in the next one week through historical data on medicine use records. Data is separated into two parts for training and testing models. We chose the data for the last 4 years (2015-2018) as training data, and data 2019 for testing [[training data] derived from the available data based on a second set of available data that is less in quantity than the available data] (i.e.: second dataset contains 1 year(2019), which is less than available training data(2015-2018) 4 years). The time period was used as input for GAN. Each weight of GAN training activated in the last layer used representation of discriminator features to generation distribution as a new representation of data.”)
Husein in view of Luus, Clark and Song do not teach:
wherein the processor is further programmed to fine-tune classifier weights, wherein to fine-tune, the processor is programmed to: access the classifier weights; and adjust the classifier weights based on the second set of available data.
Wang teaches:
wherein the processor is further programmed to fine-tune classifier weights, wherein to fine-tune, the processor is programmed to: access the classifier weights; and adjust the classifier weights based on the second set of available data.
(Wang, “[0166] The terminal updates the network parameters respectively corresponding to the preset generator model, the preset discriminator model and the preset classifier model through the gradient descent of the backpropagation algorithm respectively based on the calculated first loss function, second loss function and third loss function. For example, weight values (i.e.: classifier model weight) etc. [adjust the classifier weights] of each network layer in each of the preset generator model, the preset discriminator model and the preset classifier model (i.e.: classifier model) [wherein the processor is further programmed to fine-tune classifier weights, wherein to fine-tune, the processor is programmed to: access the classifier weights] are updated according to the first loss function, the second loss function and the third loss function. (i.e.: updating network parameters, including classifier weights, through gradient descent based on a loss function) Then, the training is continued based on the parameter-updated preset generator model, preset discriminator model and preset classifier model. That is, the sample images and the classification categories corresponding to the sample images [based on the second set of available data] (i.e.: the second set of data correspond to a new batch of sample images and classification categories used to continue training and updating the model’s parameters) continue to be trained based on the respective parameter-updated models.”)
Wang, Husein, Luus, Clark and Song are related to the same field of endeavor (i.e.: generative adversarial networks architecture). It would have been obvious to a person having ordinary skill in the art before the effective filing date of the invention to combine the teachings of Wang with teachings of Husein, Luus, Clark and Song to improve the system’s ability to prioritize important features during classification by calculating a weight vector, resulting in more accurate and representative global high-order feature maps (Wang, [0115] – [0117]).
Claim 20 recites analogous limitation as claim 7, so is rejected under the same rationale.
Claim 8 is rejected under 35 U.S.C. 103 as being unpatentable over Husein in view of Luus, Clark, Song, Wang and in further view of Tiwari et al., Pub. No.: US20180082354A1.
Husein in view of Luus, Clark, Song and Wang teach the method of claim 7.
Husein in view of Luus, Clark, Song and Wang do not teach:
wherein the available data relates to a first card type of respective card members and the second set of available data relates to a second card type of respective card members.
Tiwari teaches:
wherein the available data relates to a first card type of respective card members and the second set of available data relates to a second card type of respective card members
(Tiwari, “[0029] The payment network data transactions database 110 stores first payment card type transaction data 112 [wherein the available data relates to a first card type of respective card members] and second payment card type transaction data 114 [and the second set of available data relates to a second card type of respective card members]. The first payment card type transaction data 112 comprises data on transactions carried out using a first payment card type. In this description the term payment card type is used to refer to a payment card product such as a gold credit card of issuing bank. As the payment network processes all types of transactions such as POS, e-commerce etc.; the payment network data transactions database 110 stores all transactions by transaction type and by issuer-product combination.”)
Tiwari, Husein, Luus, Clark, Song and Wang are related to the same field of endeavor (i.e.: generative adversarial networks architecture). It would have been obvious to a person having ordinary skill in the art before the effective filing date of the invention to combine the teachings of Tiwari with teachings of Husein, Luus, Clark, Song and Wang to add the ability to provide a more comprehensive assessment of payment card performance by combining electronic commerce and billing data to gain deeper insights into customer behavior and card usage, improving decision-making for marketing, fraud detection, or financial analysis (Tiwari, Abstract).
Claim(s) 9 and 10 are rejected under 35 U.S.C. 103 as being unpatentable over Husein in view of Luus, Clark, Song and Tiwari.
Regarding claim 9, Husein in view of Luus, Clark and Song teach the method of claim 1.
Husein in view of Luus, Clark and Song do not teach:
wherein the first dataset comprises univariate data relating to a plurality of card members and wherein the features comprise a time series of data relating to an amount of spending of each of the plurality of card members.
Tiwari teaches:
wherein the first dataset comprises univariate data relating to a plurality of card members, and
(Tiwari, “[0036] The data warehouse database 182 may also comprise records relating to individual cardholders [relating to a plurality of card members], which, for example, may associate demographic information such as age, gender, number of dependents and salary range [wherein the first dataset comprises univariate data] (i.e.: salary range represents a single variable associated with a cardholder, it can be categorized as univariate data) with a card identifier (e.g., a PAN), thereby permitting transaction data to be matched to demographic data. In some embodiments, each transaction record stored in the data warehouse database 182 may already have the matched demographic data stored as part thereof.”)
wherein the features comprise a time series of data relating to an amount of spending of each of the plurality of card members.
(Tiwari, “[0030] The second payment card type transaction data 114 comprises data on transactions carried out using a second payment card type. It is envisaged that in embodiments of the present invention more than two payment card types may be analysed, however for the sake of simplicity two are shown in FIG. 1. Each of the first payment card type transaction data 112 and the second payment card type transaction data 114 comprise indications of transactions, which indicates information including the time and date of transactions [wherein the features comprise a time series of data]; transaction amount [relating to an amount of spending of each of the plurality of card members]; the card number of a payment card used for the transaction; and the merchant at which the transaction was carried out. The first payment card type transaction data 112 and the second payment card type transaction data may also comprise point of interest (POI) flags which indicate whether a transaction is an electronic commerce transaction. When a merchant gets associated with a payment network through an acquirer, the merchant is allocated a POI classification which identifies whether the merchant is a bricks & mortar merchant or an electronic commerce merchant.”)
Tiwari, Husein, Luus, Song and Clark are related to the same field of endeavor (i.e.: generative adversarial networks architecture). It would have been obvious to a person having ordinary skill in the art before the effective filing date of the invention to combine the teachings of Tiwari with teachings of Husein, Luus, Clark and Song to providing detailed transactional data such as merchant information, business category, currency, transaction environment, location, and ticket size, which can be used to analyze and segment cardholder behavior patterns. (Tiwari, ¶[0038]).
Regarding claim 10, Husein in view of Luus, Clark and Song teach the method of claim 1.
Husein in view of Luus, Clark and Song do not teach:
wherein the first dataset comprises multivariate data relating to a plurality of card members, and wherein the features comprise at least a time series of data relating to an amount of spending of each of the plurality of card members and at least one other characteristic of each of the plurality of card members.
Tiwari teaches:
wherein the first dataset comprises multivariate data relating to a plurality of card members, and
(Tiwari, “[0036] The data warehouse database 182 may also comprise records relating to individual cardholders [relating to a plurality of card members], which, for example, may associate demographic information such as age, gender [wherein the first dataset comprises multivariate data], number of dependents and salary range with a card identifier (e.g., a PAN), thereby permitting transaction data to be matched to demographic data. In some embodiments, each transaction record stored in the data warehouse database 182 may already have the matched demographic data stored as part thereof.”)
wherein the features comprise at least a time series of data relating to an amount of spending of each of the plurality of card members and at least one other characteristic of each of the plurality of card members.
(Tiwari, “[0038] The transaction records may comprise a plurality of fields, including acquirer identifier/card accepter identifier (the combination of which uniquely defines the merchant); merchant category code (also known as card acceptor business code), that is, an indication of the type of business the merchant is involved in (for example, a gas station); cardholder base currency (i.e., U.S. Dollars, Euros, Yen, etc.); the transaction environment or method being used to conduct the transaction; card identifier (e.g., card number); time and date [wherein the features comprise at least a time series of data]; location (full address and/or GPS data); transaction amount [relating to an amount of spending of each of the plurality of card members] (also referred to herein as ticket size); terminal identifier (e.g., merchant terminal identifier or ATM identifier); and response code (also referred to herein as authorization code). Other fields may be present in each transaction record [and at least one other characteristic of each of the plurality of card members].”)
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Tiwari with teachings of Husein, Luus, Clark and Song for the same reasons disclosed for claim 9.
Claim(s) 11 – 13 are rejected under 35 U.S.C. 103 as being unpatentable over Husein in view of Luus, Clark, Song and in further view of RAJAGOPAL et al., Pub. No.: US20190188771A1.
Regarding claim 11, Husein in view of Luus, Clark and Song teach the method of claim 1.
Husein in view of Luus, Clark and Song do not teach:
wherein each label of the plurality of labels comprises a card member category that is based on a level of spend of a card member.
RAJAGOPAL teaches:
wherein each label of the plurality of labels comprises a card member category that is based on a level of spend of a card member.
(RAJAGOPAL, “[0039] Customer profile database 270 may store information relating to one or more customers having an account with a financial services provider, such as a credit card company. For example, customer profile database 270 may store customer segmentation data that represents a plurality of customer segments [wherein each label of the plurality of labels]. A customer segment may be a group of customers having similar demographics that may have similar preferences. For example, service provider terminal 110 may analyze aggregated information about customers such as gender, age, income, average monthly spend using one or more financial accounts (e.g., credit cards), average monthly spend by category [comprises a card member category that is based on a level of spend of a card member], customer spend frequency by category, home zip code, credit score, indications of financial health (e.g., debt, savings, cash flow, etc.), and general spending habits (e.g., whether a customer prefers cheap fast food or expensive fine dining) across a plurality of customer accounts to create a plurality of customer segments that are associated with a set of preferences, such that each customer segment comprises a group of customers with similar demographic characteristics and similar purchasing preferences.”)
RAJAGOPAL, Husein, Luus, Clark and Song are related to the same field of endeavor (i.e.: generative adversarial networks architecture). It would have been obvious to a person having ordinary skill in the art before the effective filing date of the invention to combine the teachings of RAJAGOPAL with teachings of Husein, Luus, Clark and Song to enhance a system that classifies card members based on their spending levels by providing real-time purchase recommendations that align with the customer’s spending habits and preferences (RAJAGOPAL, ¶[0006]).
Regarding claim 12, Husein in view of Luus, Clark, Song and RAJAGOPAL teach the method of claim 11.
RAJAGOPAL further teaches: wherein the classifier generates a respective probability that the card member belongs to a given card member category.
(RAJAGOPAL, “[0051] According to some embodiments, service provider terminal 110 may rank or assign probabilities [wherein the classifier generates a respective probability] to the subset of connected category purchase pairs based on one or more of the subcategories associated with each connected category, the customer profile, and the customer segment associated with the customer [that the card member belongs to a given card member category]. For example, service provider terminal 110 may rank the subset of connected category purchase pairs based on the frequency with which connected category purchase pairs including the subcategory associated with the connected category appear in the historical transaction data. In some embodiments, if the customer profile indicates a preference for one subcategory over another, then service provider terminal 110 may weight the rankings based on the customer preference. Similarly, service provider terminal 110 may weight the rankings of the connected category purchase pairs based on preferences associated with a customer segment that a particular customer is associated with.”)
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of RAJAGOPAL with teachings of Husein, Luus, Clark and Song for the same reasons disclosed for claim 11.
Regarding claim 13, Husein in view of Luus, Clark, Song and RAJAGOPAL teach the method of claim 12.
RAJAGOPAL further teaches: wherein the processor is further programmed to: rank, within each card member category, each card member based on the respective probability that each card member CM belongs to the card member category.
(RAJAGOPAL, “[0051] According to some embodiments, service provider terminal 110 may rank [rank, within each card member category] or assign probabilities to the subset of connected category purchase pairs based on one or more of the subcategories associated with each connected category, the customer profile, and the customer segment associated with the customer [each card member based on the respective probability that each card member CM belongs to the card member category]. For example, service provider terminal 110 may rank the subset of connected category purchase pairs based on the frequency with which connected category purchase pairs including the subcategory associated with the connected category appear in the historical transaction data. In some embodiments, if the customer profile indicates a preference for one subcategory over another, then service provider terminal 110 may weight the rankings based on the customer preference. Similarly, service provider terminal 110 may weight the rankings of the connected category purchase pairs based on preferences associated with a customer segment that a particular customer is associated with.”)
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of RAJAGOPAL with teachings of Husein, Luus, Clark and Song for the same reasons disclosed for claim 11.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Kumar, et al., "ecommercegan: A generative adversarial network for e-commerce.", (2018).
This paper introduces a Generative Adversarial Network (GAN) specifically for e-commerce orders. The GAN's generator can create many plausible orders once it is trained. The main contributions of this work are: (a) developing a compact and detailed representation of e-commerce orders, (b) training an e-commerce GAN (ecGAN) with real order data to demonstrate its feasibility, and (c) training an e-commerce conditional GAN (ec2GAN) to produce plausible orders for specific products.
Zhou et al., "Misc-GAN: A multi-scale generative model for graphs.", (2019).
This paper introduces a multi-scale graph generative model called Misc-GAN, which captures graph structure distributions at different levels. It transfers this hierarchical distribution from the graphs of interest to a single graph representation. Results from seven real data sets show that this approach is effective.
Any inquiry concerning this communication or earlier communications from the examiner
should be directed to MATIYAS T MARU whose telephone number is (571) 270-0902 or via email: matiyas.maru@uspto.gov. The examiner can normally be reached Monday 8:00am - Friday 4:00pm EST.
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,
Michelle Bechtold can be reached on (571)431-0762. The fax phone number for the organization were 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.
/M.T.M./ Examiner, Art Unit 2148
/Ryan Barrett/Primary Examiner, Art Unit 2148
Prosecution Insights