Notice of Pre-AIA or AIA Status
1. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Detailed Action
2. This Non-Final Office Action is responsive to Applicants’ amendments and arguments, as received 4/27/26. Subject to Applicants’ reply, claims 1-18 and 20-30 are now pending, of which claims 1-3 are independent.
Claim Rejections - 35 USC § 103
3. 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.
4. The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
5. Claims 1, 6-8, 10, and 29 are rejected under 35 U.S.C. 103 as being unpatentable over Non-Patent Literature “Isolation Forest” (“Liu”) in view of “IsolationForest” (“Scikit-Learn”).
Regarding claim 1, LIU teaches A computer-implemented (page 418’s section 5 discussing the use of CPUs and threaded jobs, i.e., clear use of computers, to perform the experimentation and evaluation on the framework’s test and inference/evaluation stages discussed per section 4) method for training a machine learning algorithm to detect at least one anomaly in at least one process (pages 417-418, sections 4 and 4.1, discussing the training stage for the framework to teach/train it to perform anomaly detection, and further the implication is that this is applicable to “various application domains” such as discussed in section 1’s first paragraph including logged credit card transaction data or logged computer network traffic data (i.e., examples of a process as recited)), the computer-implemented method comprising:
providing a training dataset of a process, the training dataset including a first number of training usage sequences, and the first number being greater than one (page 417’s sections 4 and 4.1: “training set”, which the Examiner understands to be inclusive of logged data (as discussed per section 1’s first paragraph) such that this data can be evaluated per anomaly detection, and such that this data could be understood to include instances of data as logged that are equivalent to “a first number of ... usage sequences”, which then if used to train per sections 4-4.1 then amount to “training usage sequences” as recited, and where all the data as logged and available could be understood to represent a recited “first number” of such data instances);
creating a second number of bootstrap datasets based on the training dataset (page 417’s section 4.1 discussing subsampling the training data set to create samples which the Examiner equates with the recited “bootstrap datasets”) by randomly drawing a third number of training usage sequences from the training dataset (page 416’s section 3, in its second paragraph, mentioning that “sub-sampling is conducted by random selection of instances”, and sections 4-4.1 further discuss a configurable sub-sampling size that is equivalent to how much data is drawn into each sub-sample), the second number being greater than one ... (section 4.1 on page 418 mentions a configurable “number of tree” which, if defined, is equivalent to how many times a sub-sample is drawn from the training data set, and where per Figure 1C’s graph, the number of trees could be anywhere between 1-1000); and
training a machine learning algorithm to detect at least one anomaly in at least one process (pages 417-418’s sections 4-4.1), the training including creating a number of process trees (“number of trees”/ “ensemble size” per section 4.1 on page 418 specifically) equal to the second number (where the number of trees simply matches the number of sub-samples, and it follows that this is the most basic way to leverage any of the sub-samples that Liu contemplates, e.g. a 1-1 relationship between a sub-sample and a tree in the ensemble) using a process mining algorithm (pages 417-418’s algorithms 1-2, which the Examiner understands to be the algorithms used to transform logged process data into an ensemble model of trees), each respective process tree among the number of process trees being created based on one corresponding bootstrap dataset among the second number of bootstrap datasets (where the number of trees simply matches the number of sub-samples, and it follows that this is the most basic way to leverage any of the sub-samples that Liu contemplates, e.g. a 1-1 relationship between a sub-sample and a tree in the ensemble).
As Applicants have argued (Reply received 4/27/26, pages 11-12), Liu does not teach that each draw of the randomly drawing the third number is being made from among the first number of training usage sequences, because Liu teaches random selection without replacement to conduct its sub-sampling. Hence, Liu teaches a sub sampling where with each draw, the pool as it were being drawn from is modified, and hence that pool does not persistent in a way that fairly reads on “the first number of training usage sequences” as recited. Rather, to teach what Liu is missing, the Examiner then relies upon SCIKIT-LEARN, see e.g., an Isolation Forest implementation for scoring anomalies that, like Liu, functions in part by random selection, and specifically provides a bootstrap Boolean argument permits a user to draw with or without replacement.
Both Liu, as previously cited, and now Scikit-Learn, as newly presented, are directed to an isolation forest implementation that is explicitly directed to anomaly detection. Hence, they are similarly directed and therefore analogous. Subject to the Examiner’s further search and consideration, the Examiner understands that Isolation Forests are capable of implementation both with or without replacement. See, e.g., the 892 form and the list of references cited but not relied upon in the Conclusion section of this present Office Action. Hence, in the Examiner’s view, the selection by a user to randomly draw with or without replacement amounts to a design/implementation choice, as the Scikit-Learn reference explicitly provides with its parameterization, and hence- it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to implement an isolation forest as taught by Liu with a selection to draw with replacement, as permitted by Scikit-Learn, with a reasonable expectation of success, to tune the bootstrapping and anomaly detection model implementation in a manner that appropriately suits the underlying model training and available training data as judged by the end user.
Regarding claim 6, Liu in view of Scikit-Learn teach the computer-implemented method of claim 1, as discussed above. The aforementioned references further teach the additional limitation wherein the machine learning algorithm is a random forest algorithm comprising the number of process trees (Liu: page 416’s section 3, in its second paragraph, mentioning that “sub-sampling is conducted by random selection of instances”, where this is understood to generate the trees in the ensemble relating to underlying logged process data (as the Examiner has discussed in relation to claim 1)). The motivation for combining the references is as discussed above in relation to claim 1.
Regarding claim 7, Liu in view of Scikit-Learn teach the computer-implemented method of claim 1, as discussed above. The aforementioned references further teach the additional limitation wherein the second number lies in a range of 50 to 200 (Liu: the number of trees is graphed in Figure 1C as shown on page 414, where the number as graphed ranges from 1 to 1000, thereby inclusive of the recited range of 50-200). The motivation for combining the references is as discussed above in relation to claim 1.
Regarding claim 8, Liu in view of Scikit-Learn teach the computer-implemented method of claim 1, as discussed above. The aforementioned references further teach the additional limitation wherein the third number is equal to or less than the first number (Liu: as discussed per claim 1, subsampling as taught may involve the drawing of a sub-sampling size number of data instances from the training dataset (section 4.1), and it reasons that the number drawn in a sub-sample would be less than the whole amount (e.g., “first number” as recited)). The motivation for combining the references is as discussed above in relation to claim 1.
Regarding claim 10, Liu in view of Scikit-Learn teach the computer-implemented method of claim 1, as discussed above. The aforementioned references further teach the additional limitation for A data processing system, comprising at least one processor or electronic circuit for performing at least the computer-implemented method of claim 1 (Liu: section 5’s first paragraph on page 418, discussing the use of CPUs in relation to the section 4 stages for training and evaluation). The motivation for combining the references is as discussed above in relation to claim 1.
Regarding claim 29, Liu in view of Scikit-Learn teach the computer-implemented method of claim 1, as discussed above. The aforementioned references further teach the additional limitation wherein a trained machine learning algorithm including the number of process trees is configured to output a plurality of output values for classifying the process as one of normal or abnormal, the plurality of output values being output in response to input of a first usage sequence corresponding to the process (Liu: section 4 on page 417 discussing that each test instance as evaluated by the trained algorithm results in an anomaly score for the instance, and per section 5 on pages 418-419, anomalies (i.e., anomalous data instances) can be ranked in accordance to that score where the top n can be selected for reporting (i.e., providing a type of threshold for comparatively designating anomaly based on their score)). The motivation for combining the references is as discussed above in relation to claim 1.
6. Claims 4 and 21 are rejected under 35 U.S.C. 103 as being unpatentable over Liu in view of Scikit-Learn and further in view of U.S. Patent Application Publication No. 2022/0075705 (“Scheepens”).
Regarding claim 4, Liu in view of Scikit-Learn teach the computer-implemented method of claim 1, as discussed above. The aforementioned references do not teach the further limitation specifically wherein the process mining algorithm is an Inductive Miner algorithm. Rather, the Examiner relies upon SCHEEPENS to teach what Liu etc. otherwise lacks, see e.g., Scheepens’s [0001] discussing the applicability of inductive miner to process tree discovery tasks.
Like Liu, Scheepens is involved with data mining and relatedly data discovery, e.g. to generate process trees/graphs and the like. Hence, the references are similarly directed and therefore analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Scheepens’s specific algorithm into Liu’s similar feature to achieve the same or similar result, with a reasonable expectation of success, to realize the usability, understandability, and accuracy objectives discussed per Scheepens’s [0023] in Liu’s modified framework.
Regarding claim 21, Liu in view of Scikit-Learn and further in view of Scheepens teach the computer-implemented method of claim 4, as discussed above. The aforementioned references further teach the additional limitations wherein the machine learning algorithm is a random forest algorithm comprising the number of process trees (Liu: page 416’s section 3, in its second paragraph, mentioning that “sub-sampling is conducted by random selection of instances”, where this is understood to generate the trees in the ensemble relating to underlying logged process data (as the Examiner has discussed in relation to claim 1)). The motivation for combining the references is as discussed above in relation to claim 4.
7. Claims 5, 9, and 12 are rejected under 35 U.S.C. 103 as being unpatentable over Liu in view of Scikit-Learn and further in view of U.S. Patent Application Publication No. 2019/0138542 (“Van Beest”).
Regarding claim 5, Liu in view of Scikit-Learn teach the computer-implemented method of claim 1, as discussed above. The aforementioned references do not teach the further limitation wherein each among the first number of training usage sequences includes a sequence of activities during use of a medical device for at least one of diagnosis of a patient or treatment of the patient. Rather, the Examiner relies upon VAN BEEST to teach what Liu etc. otherwise lacks, see e.g., Van Beest’s column 6 lines 60-67 discussing the use of medical devices explicitly in a person monitoring system resulting in the generation of event/log data, such as the type considered by Liu for example.
Like Liu, Van Beest relates to event/log data that can be evaluated for insights. Hence, the references are similarly directed and therefore analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to extend the event logging generated in application uses per Liu to be inclusive of event logging in clinical contexts from a medical device, per Van Beest, with a reasonable expectation of success, such that the information/data pertinent to a patient’s treatment/care can be readily ingested from a device and subject to analysis and use, thereby promoting advantages typically associated with automation in the state of the art.
Regarding claim 9, Liu in view of Scikit-Learn teach the computer-implemented method of claim 1, as discussed above. The aforementioned references do not teach the further limitation wherein the first number of training usage sequences are usage sequences of a medical device. Rather, the Examiner relies upon VAN BEEST to teach what Liu etc. otherwise lacks, see e.g., Van Beest’s column 6 lines 60-67 discussing the use of medical devices explicitly in a person monitoring system resulting in the generation of event/log data, such as the type considered by Liu for example.
Like Liu, Van Beest relates to event/log data that can be evaluated for insights. Hence, the references are similarly directed and therefore analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to extend the event logging generated in application uses per Liu to be inclusive of event logging in clinical contexts from a medical device, per Van Beest, with a reasonable expectation of success, such that the information/data pertinent to a patient’s treatment/care can be readily ingested from a device and subject to analysis and use, thereby promoting advantages typically associated with automation in the state of the art.
Regarding claim 12, Liu in view of Scikit-Learn teach the data processing system of claim 10, as discussed above. The aforementioned references do not teach the additional limitation for A system for detecting at least one anomaly in at least one process (as discussed per claim 1, Liu’s pages 417-418, sections 4 and 4.1, discussing the training stage for the framework to teach/train it to perform anomaly detection, and further the implication is that this is applicable to “various application domains” such as discussed in section 1’s first paragraph including logged credit card transaction data or logged computer network traffic data (i.e., examples of a process as recited)), comprising: a ... device configured to log usage sequences relating to processes and the data processing system of claim 10, the data processing system being communicatively connected to the ... device, and the data processing system being configured to receive the log usage sequences from the ... device (as discussed per claim 1, but also as just mentioned above, Liu’s page 413, section 1, 1st paragraph mentions the logging of process data and the use thereof for teaching/training and inference/evaluation aspects of a machine-learned framework to detect anomalies). Liu, as discussed thus far, does not explicitly teach that the device is a medical device, although it is open-ended in terms of entertaining extensibility to “various application domains.” Rather, the Examiner relies upon VAN BEEST to teach what Liu etc. otherwise lack, see e.g., Van Beest’s column 6 lines 60-67 discussing the use of medical devices explicitly in a person monitoring system resulting in the generation of event/log data, such as the type considered by Liu for example.
Like Liu, Van Beest relates to event/log data that can be evaluated for insights. Hence, the references are similarly directed and therefore analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to extend the event logging generated in application uses per Liu to be inclusive of event logging in clinical contexts from a medical device, per Van Beest, with a reasonable expectation of success, such that the information/data pertinent to a patient’s treatment/care can be readily ingested from a device and subject to analysis and use, thereby promoting advantages typically associated with automation in the state of the art.
8. Claim 20 is rejected under 35 U.S.C. 103 as being unpatentable over Liu in view of Scikit-Learn and Scheepens and further in view of Van Beest.
Regarding claim 20, Liu in view of Scikit-Learn and Scheepens teach the computer-implemented method of claim 4, as discussed above, but not the further limitation wherein each among the first number of training usage sequences includes a sequence of activities during use of a medical device for at least one of diagnosis of a patient or treatment of the patient. Rather, the Examiner relies upon VAN BEEST to teach what Liu etc. otherwise lack, see e.g., Van Beest’s column 6 lines 60-67 discussing the use of medical devices explicitly in a person monitoring system resulting in the generation of event/log data, such as the type considered by Liu for example.
Like Liu, Van Beest relates to event/log data that can be evaluated for insights. Hence, the references are similarly directed and therefore analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to extend the event logging generated in application uses per Liu to be inclusive of event logging in clinical contexts from a medical device, per Van Beest, with a reasonable expectation of success, such that the information/data pertinent to a patient’s treatment/care can be readily ingested from a device and subject to analysis and use, thereby promoting advantages typically associated with automation in the state of the art.
Allowable Subject Matter
9. Claims 2-3, 13-18, 22-23, 25-28 and 30 are allowed.
10. Claims 11 and 24 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Response to Arguments
11. With respect to some of the claims as argued, Applicants’ arguments filed 4/27/26 have been fully considered but have not been found persuasive. The Examiner takes this opportunity to provide some feedback.
Regarding Applicants’ arguments per independent claim 1 and its dependents, the Examiner believes the grounds for rejection as reformulated in this present Action are responsive to Applicants’ characterization of the primary reference Liu. Liu’s mention of sub-sampling by random selection without replacement, as Applicants note, is a design choice for Liu’s experiment and nothing more as far as the Examiner understands the reference as a whole. Hence, toggling the type of random selection, as an end user per the Scikit-Learn NPL might decide (e.g., per the Examiner’s 103 rationale), does not appear to break or frustrate Liu’s operability as a framework. Nor does the Examiner understand Liu to effectively teach away from sampling with replacement. Finally, the Examiner notes Liu’s date, and then notes the many references that span the time between Liu and Applicants’ effective filing date that do perform anomaly/outlier detection using isolation forest or some variant involving random selection with replacement. See for example the Conclusion section at the end of this present Office Action.
Regarding Applicants’ arguments per claim 4, the Examiner is not persuaded.
As a first matter, if the Examiner is to follow Applicants’ reasoning about how an inductive mining algorithm does not sufficiently gel in the same framework that uses a random selection bootstrapping method, then the Examiner suggests that Applicants’ own invention and the teaching found in Applicants’ own specification would be likewise deficient – since that is essentially what is taught in Applicants’ specification and per Applicants’ claims 1 and 4 when considered together. If Applicants are earnest with this reasoning, then they need to provide further detailed argumentation as to why this reasoning should dissuade the Examiner’s 103 rationale but also enable Applicants’ own invention to a standard that warrants a finding of patentability.
As a second matter, the Examiner is uncertain whether Applicants’ argument is actually in scope with the invention as recited and as taught in Applicants’ specification. As the Examiner understands it, random sampling and bootstrapping provide for the creation of subsample datasets, and then each subsample dataset is subject to the process mining algorithm (per claim 1) / inductive mining algorithm (per claim 4), with the output of that being n trees corresponding to n subsamples. Hence, the employment of random sampling to create subsamples, per Liu for example, only serves to create subsamples that are then fed into the mining algorithm, which in turn has the effect of processing the subsample to generate a tree. In contrast, as the Examiner understands it, Applicants argue that the random selection process to create the subsample thwarts the use/application of the mining algorithm as applied to the subsample. The Examiner is open to be clarified by Applicants’ next reply, but this argument does not track with Examiner’s understanding of the claims or the specification. Moreover, it is unclear why the randomness involved in sampling to create sub-sample data sets for input into an algorithm is in any actual tension, as Applicants appear to argue, with the applicability of that algorithm to that sub-sample data set.
Conclusion
12. The prior art made of record and not relied upon is considered pertinent to Applicants’ disclosure:
US 11816550 B1: Gupta
Column 5 lines 23-30 discussing outlier prediction using isolation forests, which are clarified to involve sub-sampling with replacement as discussed per column 6 lines 32-39.
US 11501101 B1: Ganesan
Column 10 lines 47-56 discussing random sample generation via sampling with replacement, pursuant to performing the steps shown in FIG. 6, and which can be applied specifically in leveraging isolation trees per column 9 lines 5-26.
US 11348034 B1: Jain
Column 8 line 31 – column 9 line 10 discussing random data selection as part of a bagging method, one that is described as sampling with replacement, which the reference clarifies could be used pursuant to training an isolation forest model per column 1 line 58 – column 2 line 24, column 2 lines 33-36, column 8 lines 31-42, and column 9 line 62 – column 10 line 31
WO 2019043163 A1: Matyska Tomas
Non-Patent Literature “Isolation Forest algorithm for anomaly detection” (“Arpit”)
Discussing anomaly detection using isolation forest (page 1), where all the features of independent claim 1 appear to be taught herein with the caveat that subsampling as discussed is mentioned with any qualification as to whether it is with or without replacement.
Non-Patent Literature “Bootstrapping the out-of-sample predictions for efficient and accurate cross-validation” (“Tsamardinos”)
Non-Patent Literature “A synopsis of resampling techniques” (“Brunelli”)
Page 2 discussing bootstrapping through random selection, specifically sampling with replacement.
Non-Patent Literature “The average bootstrap sample omits 36.8% of the data” (“Wicklin”)
Non-Patent Literature “isolation.forest function” (“RDocumentation”)
Discussing an isolation forest approach for outlier detection (page 2)
Non-Patent Literature “Tree-based Machine Learning Models for Handling Imbalanced Datasets” (“Li”)
Pages 7-8 discussing a balanced random forest approach involving resampling randomly that is specifically sampling with replacement to construct individual decision trees in a forest.
13. Any inquiry concerning this communication or earlier communications from the examiner should be directed to SHOURJO DASGUPTA whose telephone number is (571)272-7207. The examiner can normally be reached M-F 8am-5pm CST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Tamara Kyle can be reached at 571 272 4241. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/SHOURJO DASGUPTA/Primary Examiner, Art Unit 2144
Prosecution Insights