Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim Rejections - 35 USC § 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, 13, and 17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Neuhof et al. (US Pub. No. 2025/0209157, filed Dec. 2023, hereinafter “Neuhof”) in view of Xu et al. (NPL: Anomaly Transformer: Time Series Anomaly Detection with Association Discrepancy, published June 2022, hereinafter “Xu”).
Regarding claim 1, Neuhof teaches a method for managing an inference model, the method comprising:
obtaining input data from one or more data sources to generate a prediction using the inference model (Neuhof, [0018] – “ In the following examples, a transformer is trained to generate an activity prediction based on an observed activity sequence associated with a user account.” and in [0028] – “User activity data 110 is collected from one or more sensors 101 deployed within a computer infrastructure. For example, sensors may be deployed to endpoint devices (such as user devices, servers etc.) within a network, and collect and report user activity data. Such ‘raw’ activity data is a form of telemetry that may be stored in a telemetry database or databases.” – teaches obtaining input data from one or more data sources to generate a prediction using the inference model (user activity data is collected from one or more sensors to generative an activity prediction using a transformer));
obtaining a measure of anomalousness of the input data using an anomaly detector (Neuhof, [0018] – “An activity sequence means a time-series of activities. Each activity is an element of the activity sequence, and may for example take the form of an activity feature vector or scalar associated with the user account in a particular observation period (such as a one day period). The activity prediction is then compared with a further observed activity corresponding in time to the predicted activity (e.g. both relating to a common observation period, which may be earlier or later in time than the observed activity sequence used to make the prediction). The observed activity is classified as anomalous or benign based on an extent to which it matches the predicted activity (e.g., which may be quantified by an anomaly score).” and in [0036] – “In the first stage, the lightweight anomaly detector 102 is applied to the activity sequences 116A, . . . 116E held in the activity databases, in order to initially classify each activity feature vector for the current observation period T as anomalous or benign… For example, with a one day observation period, the process could be run every few minutes to update the anomaly feature vectors 130 in the current observation window, and the anomaly scores/classifications for the current observation period as new activity data comes in.” – teaches obtaining a measure of anomalousness of the input data using an anomaly detector (observed activity is classified as anomalous or benign on an extent to which it matches the predicted activity, which may be quantified by an anomaly score, thus obtaining a measure of anomalousness of the input data using an anomaly detector, such as anomaly detector 102));
obtaining an anomaly condition associated with the input data using the measure of anomalousness (Neuhof, [0018] – “The observed activity is classified as anomalous or benign based on an extent to which it matches the predicted activity (e.g., which may be quantified by an anomaly score).” – teaches obtaining an anomaly condition associated with the input data using the measure of anomalousness (observed activity is classified as anomalous or benign based on the anomaly score, thus obtaining an anomaly condition associated with the input data using the measure of anomalousness));
obtaining a prediction using the [[an]] updated inference model and the input data (Neuhof, [0037] – “Accounts with anomalous behaviors identified in the first stage are sent to a second stage for further processing. ”, and in [0038] – “The second stage employs fine behavior monitoring by leveraging individual account activity history to detect deviations. An activity predictor 103 is used in the second stage. In the present example, the activity predictor 103 takes the form of a single Deep Neural Network (DNN)-based time-series prediction model. As described above, the activity predictor 103 is used to model normal behaviors of individual accounts and to assess an extent to which a current behavior associated with differs from its past normal behavior.” – teaches obtaining a prediction using an updated inference model and the input data (identified accounts are sent to second stage, second stage employs activity predictor 103 as a DNN-based time-series prediction model which is used to model normal behaviors of individual accounts and assess extent to which current behavior differs from past behavior, thus obtaining a prediction using the updated inference model and the input data)); and
providing a computer-implemented service based, at least in part, on the prediction (Neuhof, [0021] – “Activity is only escalated to a second-stage anomaly detection if it is classed as anomalous in the first stage (that is, only positive detections are escalated). In the second stage, time series-based anomaly detection is used to refine the positive detections obtained in the first stage, which may result in activity that was classed as anomalous in the first stage being re-classified as benign in the second stage.” and in [0022] – “Activity classed as anomalous using time series-based prediction (e.g., whose anomalous classification is confirmed in the second stage) may be escalated to a third stage, which means triggering a cybersecurity mitigation action in relation to a user account associated with the activity in question. Such action may, for example, comprise automatically locking or isolating the user account, removing or restricting privileges associated with the user account, or generating an alert identifying the user account at a user interface (e.g., to prompt a review by a human security expert).” – teaches providing a computer-implemented service based, at least in part, on the prediction (activity classified as anomalous using time series-based prediction in stage two is escalated and triggers a cybersecurity mitigation action, thus providing a computer-implemented service based at least in part on the prediction)).
Neuhoff fails to explicitly teach ingesting the anomaly condition into an attention mechanism of the inference model to obtain an updated inference model.
However, analogous to the field of the claimed invention, Xu teaches:
ingesting the anomaly condition into an attention mechanism of the inference model to obtain an updated inference model (Xu, Section 3 Paragraph 1 – “The observed time series X is denoted by a set of time points {x1,x2,··· ,xN}, where xt ∈ Rd represents the observation of time t. The unsupervised time series anomaly detection problem is to determine whether xt is anomalous or not without labels.”, Section 3 Paragraph 2 – “We propose the Anomaly Transformer to discover more informative associations and tackle this problem by learning the Association Discrepancy, which is inherently normal-abnormal distinguishable. Technically, we propose the Anomaly-Attention to embody the prior-association and series-associations, along with a minimax optimization strategy to obtain a more distinguishable association discrepancy.”, Section 3.1 Paragraph 2 – “Anomaly Transformer is characterized by stacking the Anomaly-Attention blocks and feed-forward layers alternately. This stacking structure is conducive to learning underlying associations from deep multi-level features.”, and in Fig. 1 description – “Anomaly Transformer architecture. Anomaly-Attention (left) models the prior-association and series-association simultaneously. In addition to the reconstruction loss, our model is also optimized by the minimax strategy with a specially-designed stop-gradient mechanism (gray arrows) to constrain the prior- and series-associations for more distinguishable association discrepancy.” – teaches ingesting the anomaly condition into an attention mechanism of the inference model to obtain an updated inference model (observed time series is passed into anomaly attention mechanism to determine if observed segment is anomalous, attention mechanism models prior-association and series-association simultaneously, and uses reconstruction loss and minimax strategy to optimize model and update gradients, thus updating the inference model by ingesting the anomaly condition into an attention mechanism));
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the anomaly attention mechanism of Xu to the inference model, data, and services of Neuhof. Doing so would enable the model to learn informative representations from complex temporal dynamics through unsupervised tasks and incorporate series-association, which provides a more informative description of temporal context, into the attention mechanism of the model (Xu, Introduction).
Claims 13 and 17 incorporate substantively all the limitations of claim 1 in a non-transitory machine-readable medium and a system, and are rejected on similar grounds as above. Neuhof teaches the processors and memory of these claims at [0101] – “Computing system 600 includes a logic processor 602, volatile memory 604, and a non-volatile storage device 606.”
Regarding claim 6, the combination of Neuhof and Xu teaches the method of claim 1, wherein
the inference model is a neural network, the neural network being trained using a transformer architecture (Neuhof, [0018] – “ In the following examples, a transformer is trained to generate an activity prediction based on an observed activity sequence associated with a user account.” – teaches wherein the inference model is a neural network trained using a transformer architecture (transformer is trained to generate an activity prediction)).
Regarding claim 7, the combination of Neuhof and Xu teaches the method of claim 6, wherein
weights of the neural network are modified based on the anomaly condition to update the inference model (Xu, Section 3.1 Last Paragraph – “From previous observation, anomalies will present smaller AssDis(P,S;X) than normal time points, which makes AssDis inherently distinguishable.” and in Section 3.2 Paragraph 1 – “As an unsupervised task, we employ the reconstruction loss for optimizing our model. The reconstruction loss will guide the series-association to find the most informative associations. To further amplify the difference between normal and abnormal time points, we also use an additional loss to enlarge the association discrepancy. Due to the unimodal property of the prior-association, the discrepancy loss will guide the series-association to pay more attention to the non-adjacent area, which makes the reconstruction of anomalies harder and makes anomalies more identifiable. The loss function for input series X∈RN×d is formalized as: Eq. (4)” – teaches wherein weights of the neural network are modified based on the anomaly condition to update the inference model (anomalies present small Association Discrepancy than non-anomalous time points. Optimizes model with an additional loss, the Association Discrepancy, to further amplify difference between normal and abnormal time points, thus modifying the weights of the neural network, in Eq. (4) the loss function shows the weight modification, based on the anomaly condition, in Eq. (4) the anomaly condition is the association discrepancy)).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the weight modification of Xu to further modify the anomaly condition, inference model, and anomaly detectors of Neuhof and Xu. Doing so would enable the model to incorporate series-association, which provides a more informative description of temporal context, into the attention mechanism of the model and derive new association-based detection criterion (Xu, Introduction).
Regarding claim 8, the combination of Neuhof and Xu teaches the method of claim 7, wherein
modifying the weights contextualizes the prediction with respect to the anomaly condition (Xu, Section 3.1 Last Paragraph – “From previous observation, anomalies will present smaller AssDis(P,S;X) than normal time points, which makes AssDis inherently distinguishable.” and in Section 3.2 Paragraph 1 – “As an unsupervised task, we employ the reconstruction loss for optimizing our model. The reconstruction loss will guide the series-association to find the most informative associations. To further amplify the difference between normal and abnormal time points, we also use an additional loss to enlarge the association discrepancy. Due to the unimodal property of the prior-association, the discrepancy loss will guide the series-association to pay more attention to the non-adjacent area, which makes the reconstruction of anomalies harder and makes anomalies more identifiable. The loss function for input series X∈RN×d is formalized as: Eq. (4)” – teaches wherein modifying the weights contextualizes the prediction with respect to the anomaly condition (model is optimized with respect to the Association Discrepancy which amplifies the difference between normal and abnormal time points, thus the loss function as in Eq. (4) provides updates to the model to contextualize predictions with respect to the anomaly conditions)) .
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the weight modification of Xu to further modify the anomaly condition, inference model, and anomaly detectors of Neuhof and Xu. Doing so would enable the model to incorporate series-association, which provides a more informative description of temporal context, into the attention mechanism of the model and derive new association-based detection criterion (Xu, Introduction).
Regarding claim 9, the combination of Neuhof and Xu teaches the method of claim 8, wherein
the attention mechanism impacts operation of an input layer of the neural network (Xu, Section 3.1 Paragraph 2 – “Anomaly Transformer is characterized by stacking the Anomaly-Attention blocks and feed-forward layers alternately. This stacking structure is conducive to learning underlying associations from deep multi-level features. Suppose the model contains L layers with length-N input time series X∈RN×d. The overall equations of the l-th layer are formalized as: Eq. (1)” and Fig. 1 – teaches wherein the attention mechanism impacts operation of an input layer of the neural network (Fig. 1 (left) shows input X, which comes from input layer, input into Anomaly-Attention mechanism, thus the attention mechanism impacts operation of an input layer of the neural network)).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the attention mechanism impacting the input layer of Xu to further modify the inference model, neural network, and anomaly detector of Neuhof and Xu. Doing so would apply Anomaly-Attention to the embedded raw series to embody the prior-association and series-associations (Xu, Section 3.1).
Regarding claim 10, the combination of Neuhof and Xu teaches the method of claim 8, wherein
the attention mechanism impacts operation of at least one hidden layer of the neural network (Xu, Section 3.1 Paragraph 2 – “Anomaly Transformer is characterized by stacking the Anomaly-Attention blocks and feed-forward layers alternately” and Fig. 1 – teaches wherein the attention mechanism impacts operation of at least one hidden layer of the neural network (transformer architecture is characterized by stacking Anomaly-Attention blocks and feed-forward layers alternately, also shown in Fig. 1 (right), thus the Anomaly-Attention mechanism impacts operation of at least one hidden layer of the neural network)).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the attention mechanism impacting the hidden layer of Xu to further modify the inference model, neural network, and anomaly detector of Neuhof and Xu. Doing so would apply Anomaly-Attention to the l-th layer to embody the prior-association and series-associations (Xu, Section 3.1).
Regarding claim 11, the combination of Neuhof and Xu teaches the method of claim 1, wherein
the prediction comprises information usable to manage a condition impacting a business at a future point in time (Neuhof, [0017] – “To address the challenges of diverse and evasive activities in enterprise accounts, an insight that most accounts exhibit fixed and repetitive patterns is leveraged. When an account is compromised during lateral movement, the attacker's behavior should deviate significantly from the normal user's behavior. An approach is described, which models each account's behavior individually and detects deviations to identify compromised accounts.” – teaches wherein the prediction comprises information usable to manage a condition impacting a business at a future point in time (prediction models normal user behavior and detects significant deviations to identify compromised enterprise accounts, thus the prediction comprises information usable to manage a condition impacting a business at a future point in time)).
Claim(s) 2-5, 14-16, and 18-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Neuhof and Xu as applied to claims 1, 13, and 17 above, and further in view of Tuli et al. (NPL: TranAD: Deep Transformer Networks for Anomaly Detection in Multivariate Time Series Data, published May 2022, hereinafter “Tuli”).
Regarding claim 2, the combination of Neuhof and Xu teaches the method of claim 1.
The combination of Neuhof and Xu fails to explicitly teach wherein the anomaly detector comprises a fixed-vector inference model trained to generate a fixed output upon ingesting non-anomalous input data.
However, analogous to the field of the claimed invention, Tuli teaches: wherein
the anomaly detector comprises a fixed-vector inference model trained to generate a fixed output upon ingesting non-anomalous input data (Tuli, Section 3.1 Paragraph 1 – “We consider a multivariate time-series, which is a timestamped sequence of observations/datapoints of size T T = {x1…xT} where each datapoint 𝑥𝑡 is collected at a specific timestamp 𝑡 and 𝑥𝑡 ∈ IR𝑚, ∀𝑡.”, Section 3.1 Paragraph 2 – “Given a training input time-series T, for any unseen test time-series ˆ T of length ˆ𝑇 and same modality as the training series, we need to predict Y = {𝑦1,. . .,𝑦ˆ𝑇}, where we use 𝑦𝑡 ∈ {0,1} to denote whether the datapoint at the 𝑡-th timestamp of the test set is anomalous (1 denotes an anomalous datapoint).”, and in Section 3.2 Paragraph 2 – “To model the dependence of a data point 𝑥𝑡 at a timestamp 𝑡, we consider a local contextual window of length 𝐾 as Wt = {xt-K+1…,xt}” – teaches wherein the anomaly detector comprises a fixed-vector inference model (predicts Y based on a time-series of size T, thus the anomaly detector comprises a fixed-vector inference model) trained to generate a fixed output upon ingesting non-anomalous input data (outputs window Wt of fixed-length and predicts labels Y of the form 𝑦𝑡 ∈ {0,1} where 0 denotes a non-anomalous input, thus the fixed-vector inference model is trained to generate a fixed output upon ingesting non-anomalous input data)).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the fixed-vector inference model of Tuli to the anomaly detector, input data, and methods of Neuhof and Xu. Doing so would provide an unsupervised method for anomaly detection that is fast for training and testing while maintaining stability with large input sequences (Tuli, Introduction).
Claims 14 and 18 are similar to claim 2, hence similarly rejected.
Regarding claim 3, the combination of Neuhof, Xu, and Tuli teaches the method of claim 2, wherein obtaining the measure of anomalousness comprises:
ingesting the input data into the fixed-vector inference model (Tuli, Section 3.1 Paragraph 1 – “We consider a multivariate time-series, which is a timestamped sequence of observations/datapoints of size T T = {x1…xT} where each datapoint 𝑥𝑡 is collected at a specific timestamp 𝑡 and 𝑥𝑡 ∈ IR𝑚, ∀𝑡.”, Section 3.1 Paragraph 2 – “Given a training input time-series T, for any unseen test time-series ˆ T of length ˆ𝑇 and same modality as the training series, we need to predict Y = {𝑦1,. . .,𝑦ˆ𝑇}, where we use 𝑦𝑡 ∈ {0,1} to denote whether the datapoint at the 𝑡-th timestamp of the test set is anomalous (1 denotes an anomalous datapoint).”, and in Section 3.4 Paragraph 2 – “The Transformer model enables us to predict the reconstruction of each input time-series window.” – teaches wherein obtaining the measure of anomalousness comprises ingesting the input data into the fixed-vector inference model (multivariate time-series of size T is fed into fixed-vector inference model to predict reconstructed input Ot and predict labels Y of the form 𝑦𝑡 ∈ {0,1}, where a value of 1 denotes an anomalous datapoint, thus ingesting the input data into the fixed-vector inference model to obtain a measure of anomalousness)).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the fixed-vector inference model of Tuli to further modify the anomaly detector, input data, and methods of Neuhof and Xu. Doing so would provide an unsupervised method for anomaly detection that is fast for training and testing while maintaining stability with large input sequences, and provide detection of behavior that does not conform to expected trends (Tuli, Introduction).
Claims 15 and 19 are similar to claim 3, hence similarly rejected.
Regarding claim 4, the combination of Neuhof, Xu, and Tuli teaches the method of claim 3, wherein obtaining the anomaly condition comprises:
making a determination regarding whether the [[a]] difference exceeds an anomaly threshold (Neuhoff, [0018] – “The observed activity is classified as anomalous or benign based on an extent to which it matches the predicted activity (e.g., which may be quantified by an anomaly score).” and in [0098] – “If the anomaly score M exceeds a threshold, the system 100 flags the account as positive (potentially compromised), and passed it to the third stage for further scrutiny.” – teaches making a determination regarding whether the difference exceeds an anomaly threshold (makes a determination regarding whether the anomaly score M, which quantifies the difference between the observed and predicted activity, exceeds a threshold)); and
in a first instance of the determination where the difference exceeds the anomaly threshold: treating the input data as anomalous (Neuhof, [0098] – “If the anomaly score M exceeds a threshold, the system 100 flags the account as positive (potentially compromised), and passed it to the third stage for further scrutiny.” – teaches in a first instance of the determination where the difference exceeds the anomaly threshold, treating the input data as anomalous (if anomaly score M exceeds a threshold, system flags the account as positive, thus treating the input data of the account as anomalous)), and
in a second instance of the determination where the difference does not exceed the anomaly threshold: treating the input data as non-anomalous (Neuhof, [0098] – “If the anomaly score M exceeds a threshold, the system 100 flags the account as positive (potentially compromised), and passed it to the third stage for further scrutiny.” – teaches in a second instance of the determination where the difference does not exceed the anomaly threshold, treating the input data as non-anomalous (if anomaly score M fails to exceed the threshold, then the account is flagged as negative, and thus treating the input data of the account as non-anomalous)).
The combination of Neuhof and Xu fails to explicitly teach obtaining a difference between the measure of anomalousness and the fixed output.
However, analogous to the field of the claimed invention, Tuli teaches:
obtaining a difference between the measure of anomalousness and the fixed output (Tuli, Section 3.2 Paragraph 4 – “To calculate the anomaly score 𝑠𝑡, we reconstruct the input window as 𝑂𝑡 and use the deviation between 𝑊𝑡 and 𝑂𝑡.” and in Section 3.4 Paragraph 2 – “The Transformer model enables us to predict the reconstruction of each input time-series window.” – teaches obtaining a difference between the measure of anomalousness and the fixed output (determines difference between predicted reconstruction of input time-series window Ot and the fixed window Wt, anomaly score st is the difference between the measure of anomalousness and the fixed output));
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the calculated difference of Tuli to further modify the anomaly detector, input data, and anomaly thresholds of Neuhof and Xu. Doing so would provide an unsupervised method for anomaly detection that is fast for training and testing while maintaining stability with large input sequences, and provide detection of behavior that does not conform to expected trends (Tuli, Introduction).
Claims 16 and 20 are similar to claim 4, hence similarly rejected.
Regarding claim 5, the combination of Neuhof, Xu, and Tuli teaches the method of claim 4, wherein treating the input data as anomalous comprises:
obtaining a deviation for the measure of anomalousness based on the fixed output (Tuli, Section 3.2 Paragraph 4 – “To calculate the anomaly score 𝑠𝑡, we reconstruct the input window as 𝑂𝑡 and use the deviation between 𝑊𝑡 and 𝑂𝑡.” – teaches obtaining a deviation for the measure of anomalousness based on the fixed output (determines deviation between the predicted reconstructed of input time-series window Ot and the fixed window Wt)); and
using a classification schema keyed to the deviation to obtain the anomaly condition (Tuli, Section 3.2 Paragraph 4 – “Using anomaly scores for the past input windows, we calculate a threshold value 𝐷, above which we label the input window as anomalous, thus 𝑦𝑡 = 1(𝑠𝑡 ≥ 𝐷).” – teaches using a classification schema (calculated threshold) keyed to the deviation to obtain the anomaly condition (classifies input window as anomalous based on the anomaly score keyed to the calculated threshold, thus using a classifications schema keyed to the deviation to obtain the anomaly condition)).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the deviation and classification schema of Tuli to further modify the inference model, anomaly detector, and thresholds of Neuhof, Xu, and Tuli. Doing so would provide mechanisms for multi-class prediction for determining an anomaly and from which source it originates from (Tuli, Introduction).
Claim(s) 12 is/are rejected under 35 U.S.C. 103 as being unpatentable over Neuhof and Xu as applied to claims 1, 13, and 17 above, and further in view of Nguyen et al. (NPL: Forecasting and Anomaly Detection Approaches using LSTM and LSTM Autoencoder techniques with the applications in supply chain management, published April 2021, hereinafter “Nguyen”).
Regarding claim 12, the combination of Neuhof and Xu teaches the method of claim 11.
The combination of Neuhof and Xu fails to explicitly teach wherein the condition impacting the business at the future point in time is a change in availability of supply of a product from a supplier.
However, analogous to the field of the claimed invention, Nguyen teaches:
the condition impacting the business at the future point in time is a change in availability of supply of a product from a supplier (Nguyen, Section 3 Paragraph 3 – “In this study, we focus on the downstream supply chain of fashion retailers. As mentioned earlier, consumer demand very fluctuates. When the product variety is high, inventory allocations become very challenging for an extensive store network. Thus, companies rely on efficient and reactive information system to monitor POS data and compute replenishment of each store for the next day or next two days. Combined with efficient transportation and distribution logistics, this process enables companies to drive their local inventories in most situations… However, the product variety and extensive store network generate a huge number of situations which are as many sources of forecast errors. To deal with these issues, the proposed approach which combines new advances in forecasting with the LSTM network, the LSTM Autoencoder network, and the OCSVM algorithm. In this context, the aim of our method is not only to predict the exact sales by stock-keeping unit (SKU) and store but also to detect and anticipate exceptional sales in order to enable practitioners to make a suitable decision and adjust their replenishment for highlighted SKU/stores accordingly.” – teaches wherein the condition impacting the business at the future point in time is a change in the availability of supply of a product from a supplier (system forecasts timeseries with LSTM, LSTM Autoencoder, and OCSVM algorithm to perform anomaly detection, as in Section 5.2, wherein the anomaly detector identifies conditions affecting availability of supply of a product from a supplier in order to predict exactly sales by stock-keeping unit and adjust replenishment)).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the conditions impacting a business of Nguyen to the anomaly detector, inference models, and activity sequences of Neuhof and Xu. Doing so would allow determining quickly anomalies or unexpected patterns for making more effective decisions and enable application of anomaly detection models to supply chain management (Nguyen, Introduction).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Moschini et al. (NPL: Anomaly and Fraud Detection in Credit Card Transactions Using the ARIMA Model, published July 2021) teaches methods for anomaly and fraud detection in card transactions using the autoregressive integrated moving average (ARIMA) models. Teaches determining if time series are stationary or not using differencing. Teaches estimating coefficients, or parameters, of the model based at least in part on anomaly conditions.
Branisavljevic (NPL: Improved real-time data anomaly detection using context classification, published Jan. 2011) teaches methods for improving performance of real-time anomaly detection using data context classifications and tuning. Teaches a classification schema for classifying an anomaly condition based on anomalous measures. Teaches tuning models based on the identified context to further improve detection performance.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to LOUIS C NYE whose telephone number is 571-272-0636. The examiner can normally be reached Monday - Friday 9:00AM - 5:00PM.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, MATT ELL can be reached at 571-270-3264. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/LOUIS CHRISTOPHER NYE/Examiner, Art Unit 2141
/MATTHEW ELL/Supervisory Patent Examiner, Art Unit 2141
Prosecution Insights