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 .
Drawings
The drawings are objected to because Figure 5 (Synthetic Samples Generation Algorithm 500) numbers its steps 1 through 13, and then resumes numbering at steps 16 and 17, omitting steps 14 and 15 without explanation, rendering the drawing incomplete as to the algorithm it purports to illustrate. Corrected drawing sheets in compliance with 37 CFR 1.121(d) are required in reply to the Office action to avoid abandonment of the application. Any amended replacement drawing sheet should include all of the figures appearing on the immediate prior version of the sheet, even if only one figure is being amended. The figure or figure number of an amended drawing should not be labeled as "amended." If a drawing figure is to be canceled, the appropriate figure must be removed from the replacement sheet, and where necessary, the remaining figures must be renumbered and appropriate changes made to the brief description of the several views of the drawings for consistency. Additional replacement sheets may be necessary to show the renumbering of the remaining figures. Each drawing sheet submitted after the filing date of an application must be labeled in the top margin as either "Replacement Sheet" or "New Sheet" pursuant to 37 CFR 1.121(d). If the changes are not accepted by the examiner, the applicant will be notified and informed of any required corrective action in the next Office action. The objection to the drawings will not be held in abeyance.
Specification
The disclosure is objected to because of the following informalities: (a) ¶[0043] contains the misspelling "sequencies," which should read "sequences"; (b) ¶[0020] contains the typographical error "Once explanation," which should read "One explanation"; and (c) the specification inconsistently uses the terms "abnormality threshold" (¶[0047], [0054], and the heading preceding ¶[0050]) and "anomaly threshold" (¶[0039], [0056], and the captions of Figs. 2 and 4) to refer to the same threshold r; consistent terminology should be used throughout the specification.
Appropriate correction is required.
Claim Objections
Applicant is advised that should claim 10, as further limited by claim 15, be found allowable, claim 18 will be objected to under 37 CFR 1.75 as being a substantial duplicate thereof. When two claims in an application are duplicates or else are so close in content that they both cover the same thing, despite a slight difference in wording, it is proper after allowing one claim to object to the other as being a substantial duplicate of the allowed claim. See MPEP § 608.01(m).
Claim Rejections - 35 USC § 112
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention.
Claim 12 is rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claim contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, at the time the application was filed, had possession of the claimed invention.
Claim 12 recites "wherein the reconstruction errors are backpropagated through a network of the previously trained anomaly detection VAE model." Claim 12 depends from claim 10, which recites causing "a previously trained anomaly detection variational autoencoder (VAE) model to operate on a test set comprising non-anomalous samples" and computing "reconstruction errors based on the output of the VAE." The recited "the reconstruction errors" of claim 12 therefore draw antecedent basis from, and are limited to, the reconstruction errors that claim 10 computes from the output of the already-trained VAE model as it operates on the non-anomalous test set. Claim 12 accordingly requires that those test-phase reconstruction errors be backpropagated through the network of a model whose training is already complete.
The specification does not describe that operation. Backpropagation is disclosed in a single context — the training of the VAE model, performed on training data before the model becomes the "previously trained" model of claim 10. Specifically, the specification at ¶[0052] describes the VAE training process 300 of Figure 3 as including "an act (act 305) of randomly initializing the VAE weights," after which "[a]cts 310 through 330 are then performed for each non-anomaly sample in Xtrain," and ¶[0053] describes those acts as coding the sample, sampling a new point in latent space, decoding the new point, "computing the reconstruction error" (act 325), and "backpropagating the error through the network" (act 330). The errors so backpropagated are the errors of the training samples in Xtrain, not the test-set errors recited in claim 10.
By contrast, the disclosure of the operations that actually produce the reconstruction errors recited in claim 10 contains no backpropagation step whatsoever. ¶[0055] states that "[d]uring the test step of the VAE model, for each non-anomalous sample, the embodiments compute its respective reconstruction error and store it in a list." ¶¶[0056]–[0057] set out the corresponding process 400 of Figure 4 in full — act 405 initializing the empty error list, act 410 obtaining the latent dimension D, act 415 defining the reconstruction error function, acts 420 through 455 encoding, computing the mean and log-variance, sampling, decoding, computing the reconstruction error, and adding that error to the error list, and act 460 defining the anomaly threshold r using the error list. Not one act of process 400 propagates any error back through the network. Figure 8A is to the same effect: ¶[0069] describes act 805 training the VAE model, act 810 testing the VAE model, act 815 computing reconstruction errors, and act 820 defining the anomaly detection threshold, with no test-phase backpropagation.
The specification further confirms that the model's training is complete at the point where the recited reconstruction errors arise. ¶[0047] states that "[i]n phase one, the embodiments applied a previously trained anomaly detection VAE model to a test set with non-anomalous samples. Based on the reconstruction errors of the test samples, the embodiments defined an abnormality threshold." ¶[0054] likewise begins, "Now, given a trained VAE model, the embodiments compute the abnormality threshold used to classify anomalies." The specification thus describes the test-phase reconstruction errors as serving exactly one purpose — accumulation in a list from which the threshold r is derived — and describes the trained model's parameters as fixed at that point, such that the threshold and the subsequently recited latent-space navigation operate on a stable network. Backpropagating the test-phase errors through that network, as claim 12 requires, would alter the very weights the specification treats as settled, and the specification nowhere describes, contemplates, or reconciles that operation.
The specification affirmatively confines backpropagation to the training phase and affirmatively assigns the test-phase reconstruction errors a different and exclusive use. For these reasons, the specification fails to reasonably convey to one of ordinary skill in the art that the inventors had possession of the subject matter of claim 12 at the time the application was filed. See Ariad Pharmaceuticals, Inc. v. Eli Lilly & Co., 598 F.3d 1336, 1351 (Fed. Cir. 2010) (en banc); Vas-Cath Inc. v. Mahurkar, 935 F.2d 1555, 1563-64 (Fed. Cir. 1991); MPEP § 2163.
Appropriate correction is required. Applicant may consider amending claim 12 to recite that reconstruction errors computed during training of the anomaly detection VAE model are backpropagated through the network of the model, which would be consistent with the disclosure at ¶[0053] and act 330 of Figure 3. No new matter should be introduced.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-20 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claims 1, 10, and 18 each recite generating a set of synthetic samples "by navigating through a latent space that exists between an embedding of an anomalous input and an embedding of an anomalous denoised sample." The phrase "anomalous denoised sample" is facially self-contradictory: the specification describes the VAE's denoising characteristic as producing "a non-anomalous synthetic sample" from an anomalous input (¶[0019]), yet the claims modify "denoised sample" with the term "anomalous." It is unclear whether the recited sample is (i) the model's non-anomalous denoised reconstruction of the anomalous input, consistent with the specification's teaching of what denoising produces, or (ii) a sample that remains anomalous notwithstanding the denoising operation, a result nowhere described in the specification. Because this term defines one of the two latent-space endpoints between which synthetic samples are generated, the ambiguity directly affects the metes and bounds of the claimed navigation step. For purposes of examination, "an embedding of an anomalous denoised sample" is interpreted under BRI to mean the embedding of the VAE's denoised (i.e., non-anomalous) reconstruction of the anomalous input, as the specification at ¶[0019] describes the VAE's denoising output as "a non-anomalous synthetic sample" built from the anomalous input.
Claims 2-9 depend, directly or indirectly, from claim 1; claims 11-17 depend from claim 10; and claims 19-20 depend from claim 18. Each of these dependent claims incorporates by reference the "an embedding of an anomalous denoised sample" limitation of its respective independent claim and does not resolve the ambiguity identified above. Each dependent claim is therefore indefinite for the same reason as its parent independent claim.
Claim 8 recites "wherein, when the reconstruction error is above the threshold, data corresponding to the reconstruction error is considered to be anomalous." There is insufficient antecedent basis for "the reconstruction error" in the claim. Claim 1, from which claim 8 depends, recites only the plural "reconstruction errors" ("computing reconstruction errors based on the output of the VAE"); no singular "a reconstruction error" is introduced in claim 1 or claim 8. Because claim 1 contemplates a reconstruction error computed for each of multiple samples in the test set, it is unclear which particular reconstruction error claim 8 refers to. For purposes of examination, "the reconstruction error" in claim 8 is interpreted under BRI to refer to any individual reconstruction error among the plurality of "reconstruction errors" recited in claim 1, computed for a given sample, as the specification at ¶¶[0038]-[0039] describes each sample's reconstruction error being individually compared against the threshold r to classify that sample.
Claim 12 recites that "the reconstruction errors are backpropagated through a network of the previously trained anomaly detection VAE model." Claim 10, from which claim 12 depends, characterizes the VAE model as "previously trained" and recites computing the reconstruction errors as part of operating that already-trained model on a test set. The specification discloses backpropagation only as part of the training process (¶[0053], act 330, Figure 3), which is distinct from and precedes the test/threshold-definition process (¶¶[0055]-[0058], Figure 4) that computes reconstruction errors for the test set — a process that includes no backpropagation step. It is therefore unclear whether claim 12 requires further training-time weight updates to a model that claim 10 already designates as "previously trained," rendering the scope of claim 12 uncertain. For purposes of examination, claim 12 is interpreted under BRI as reciting that the reconstruction errors computed in claim 10 are used to further adjust the weights of the previously trained VAE model's network via a backpropagation operation, notwithstanding the apparent inconsistency with the "previously trained" characterization in claim 10, because ¶[0053] (act 330) is the only disclosed backpropagation step in the specification and it is described only in the context of training (Figure 3).
Appropriate correction is required.
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.
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.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1-10 and 15-20 are rejected under 35 USC 103 as being unpatentable over US Pat. Pub. No. 2021/0304067A1 to Kumar et al. (hereinafter Kumar) in view of Counterfactual Explanations via Latent Space Projection and Interpolation to Barr et al. (hereinafter Barr) and further in view of Diverse Counterfactual Explanations for Anomaly Detection in Time Series to Sulem et al. (hereinafter Sulem).
Per claim 1, Kumar discloses A method (Kumar: ¶[0014]…Kumar discloses a computer-implemented method for detecting anomalies during operation of a system using a machine-learning model, which constitutes the claimed method under BRI, "there is provided a method for detecting anomalies during the operation of an information technology system"), comprising:
causing a previously trained anomaly detection variational autoencoder (VAE) model to operate on a test set comprising non-anomalous samples, the VAE generating an output based on said operating (Kumar: ¶[0038]…Kumar’s machine-learning model is a variational autoencoder (per the reference title, “Variational Autoencoding for Anomaly Detection,” and ¶[0001]) that is previously trained on non-anomalous performance-metric sequences and then operated to encode and reconstruct collected sequences, which constitutes causing a previously trained anomaly-detection VAE to operate on a test set of non-anomalous samples under BRI, "the autoencoder may be trained based on training data that includes one or more sequences of non-anomalous performance metrics but excludes sequences of anomalous performance metrics. Trained in this manner, the autoencoder may be able to encode sequences of non-anomalous performance metrics with minimal reconstruction error"; ¶[0007]…Kumar’s decoder generates the reconstruction as the output of the operating step, which constitutes the VAE generating an output based on said operating, "a decoder trained to generate, based at least on the encoded representation, the reconstruction of the sequence of performance metrics");
computing reconstruction errors based on the output of the VAE (Kumar: ¶[0010]…Kumar computes a reconstruction error as the difference between the input sequence and the VAE reconstruction output, which constitutes computing reconstruction errors based on the output of the VAE, "the reconstruction error may include a difference between the sequence of performance metrics and the reconstruction of the sequence of performance metrics");
using the reconstruction errors to define a threshold that is usable to determine whether data is anomalous or is non-anomalous (Kumar: ¶[0011]…Kumar establishes a threshold value against which the reconstruction error is compared to determine whether an anomaly is present, such that the threshold is derived from and usable on the reconstruction errors to classify data as anomalous or non-anomalous, which constitutes the claimed defining of a threshold under BRI, "the occurrence of the anomaly at the information technology system may be detected based at least on the reconstruction error exceeding a threshold value");
Kumar does not expressly disclose, but Barr does teach:
generating a set of synthetic samples by navigating through a latent space that exists between an embedding of an anomalous input and an embedding of an anomalous denoised sample (Barr: Section 2…Barr uses a variational autoencoder to obtain a controllable latent space, which constitutes the claimed VAE latent space, "To gain greater control over the latent space we use variational autoencoders (VAE)"; Section 3.1…Barr generates a set of synthetic counterfactual candidates by linearly interpolating in the VAE latent space along the line between the embedding of a base (anomalous) input and the embedding of its target-class projection (the denoised, non-anomalous reconstruction), which constitutes generating synthetic samples by navigating a latent space between an embedding of an anomalous input and an embedding of an anomalous denoised sample under BRI, "We then use linear interpolation to sample along the line between a base sample and its TVAE-generated projection in UVAE space");
computing a corresponding difference between each synthetic sample in the set of synthetic samples and said each synthetic sample's neighbor in the latent space, resulting in generation of a set of differences (Barr: Sections 3.1-3.2…Barr samples an ordered succession of interpolated points along the latent-space line, each successive interpolated point differing from its neighbor by the incremental interpolation step, and decodes the ordered points, such that the incremental change between each synthetic sample and its neighbor constitutes computing a corresponding difference between neighbors resulting in a set of differences under BRI, "The interpolated points are then decoded by the UVAE and passed to the classifier to obtain a classification score…");
Kumar combined with Barr does not expressly disclose, but Sulem does teach:
using the set of differences to generate a temporal heatmap, wherein the temporal heatmap is structured to provide an actionable explanation, which details what changes are made to turn an anomalous sample into a non-anomalous sample (Sulem: Section 5.4, Fig. 1…Sulem uses the set of counterfactual perturbations (differences) applied across the anomalous time-series sub-window to produce an interpretable color-map visualization over time whose color encodes the model’s anomaly-probability score, which constitutes using the set of differences to generate a temporal heatmap under BRI, "the orange color map indicates their anomaly probability scores (between 0 and 1) given by the model"; Section 1…Sulem’s visualization details what can be changed in the input to avoid the anomaly, i.e., what changes turn an anomalous sample into a non-anomalous one, which constitutes the claimed actionable explanation, "a user of an anomaly detection model is interested in (a) knowing what can be changed in the input data to avoid encountering the anomaly again in the future (preferentially with minimal cost), and (b) understand the model’s sensitivity to a particular anomaly").
Kumar, Barr and Sulem are analogous art because they are from the same field of endeavor, specifically machine-learning anomaly detection and the explanation of machine-learning model predictions. They address the same problem of detecting anomalies with autoencoder/variational-autoencoder models and explaining, in an actionable manner, what distinguishes an anomalous input from a non-anomalous one.
Before the effective filing date of the claimed invention, it would have been obvious to a PHOSITA to modify the VAE reconstruction-error anomaly detector of Kumar to generate counterfactual synthetic samples by interpolating through the VAE latent space between the embedding of an anomalous input and the embedding of its denoised reconstruction, as taught by Barr, and to compute and visualize the resulting differences as a temporal heatmap that details what changes turn an anomalous sample into a non-anomalous one, as taught by Sulem.
The suggestion/motivation for doing so is provided by the applied references themselves. Barr teaches that generating counterfactuals within the latent space of a VAE ensures the explanations “lie on the data manifold, and that they are close to their original sample” (Barr: Section 3.1…"we can ensure counterfactuals both lie on the data manifold, and that they are close to their original sample in latent distance"), and Sulem teaches that a user of an anomaly detection model is interested in “knowing what can be changed in the input data to avoid encountering the anomaly again in the future” (Sulem: Section 1…"a user of an anomaly detection model is interested in (a) knowing what can be changed in the input data to avoid encountering the anomaly again in the future (preferentially with minimal cost)"). Combining these express teachings with Kumar’s VAE detector yields the predictable result of an anomaly detector that additionally outputs an actionable, on-manifold explanation, per KSR rationale (A) combining prior-art elements according to known methods to yield predictable results and (G) an explicit teaching and suggestion found in the references.
Per claims 10 and 18, Kumar discloses A computer system comprising: one or more processors; and one or more hardware storage devices that store instructions that are executable by the one or more processors to cause the computer system to: (Kumar: ¶[0004]…Kumar discloses a system having at least one data processor and at least one memory storing instructions executed by the processor to perform the anomaly-detection operations, which constitutes the claimed one or more processors and one or more hardware storage devices storing executable instructions under BRI, "The system may include at least one data processor and at least one memory. The at least one memory may store instructions that result in operations when executed by the at least one data processor").
Kumar combined with Barr and Sulem discloses the recited operations — causing the VAE to operate on a non-anomalous test set and generate an output, computing reconstruction errors, using the reconstruction errors to define a threshold, generating a set of synthetic samples by navigating the latent space between the embedding of an anomalous input and the embedding of an anomalous denoised sample, computing a corresponding difference between each synthetic sample and its neighbor, and using the set of differences to generate a temporal heatmap — in the same manner set forth for claim 1 above, which claims 10 and 18 recites in commensurate scope.
Per claims 10 and 18, Kumar discloses A computer system comprising: one or more processors, and one or more hardware storage devices that store instructions that are executable by the one or more processors to cause the computer system to: (Kumar: ¶[0004]…same system hardware as set forth for claim 10, "The system may include at least one data processor and at least one memory. The at least one memory may store instructions that result in operations when executed by the at least one data processor").
Claims 10 and 18 are directed to a computer system that is substantially similar in scope and spirit to the method of claim 1. Therefore the rejection of claim 1 over Kumar, Barr and Sulem is applied to claims 10 and 18 accordingly.
Per claim 2, Kumar combined with Barr and Sulem discloses claim 1. Barr further teaches wherein said each synthetic sample’s neighbor in the latent space is an immediate neighbor (Barr: Section 3.2…Barr samples successive interpolated points along the latent line such that each sample’s adjacent sampled point is its immediate neighbor, which constitutes an immediate neighbor under BRI, "The interpolated points are then decoded by the UVAE and passed to the classifier to obtain a classification score"). The rationale to combine Barr with Kumar is the same as the parent claim.
Per claim 3, Kumar combined with Barr and Sulem discloses claim 1. Sulem further teaches wherein said method further includes generating counterfactual ensembles (Sulem: Abstract…Sulem generates counterfactual ensemble explanations, i.e., a set of diverse counterfactual examples, which constitutes generating counterfactual ensembles under BRI, "we propose a model-agnostic algorithm that generates counterfactual ensemble explanations for time series anomaly detection models"). The rationale to combine Sulem with Kumar is the same as the parent claim.
Per claim 4, Kumar combined with Barr and Sulem discloses claim 3. Sulem further teaches wherein the counterfactual ensembles highlight possible actionable changes that can be made to turn the anomalous sample into the non-anomalous sample (Sulem: Abstract…Sulem’s counterfactual ensemble comprises perturbed versions of the original anomalous series that the model deems normal, thereby highlighting the actionable changes that turn an anomalous sample into a non-anomalous one, which constitutes the claimed highlighting under BRI, "Our method generates a set of diverse counterfactual examples, i.e, multiple perturbed versions of the original time series that are not considered anomalous by the detection model"; Section 1…the explanation conveys what can be changed in the input to avoid the anomaly, "a user of an anomaly detection model is interested in (a) knowing what can be changed in the input data to avoid encountering the anomaly again in the future (preferentially with minimal cost)"). The rationale to combine Sulem with Kumar is the same as the parent claim.
Per claim 5, Kumar combined with Barr and Sulem discloses claim 1, Kumar further teaching wherein the previously trained anomaly detection VAE model is trained for time series reconstruction (Kumar: ¶[0047]…Kumar’s VAE is trained to process and reconstruct sequences of performance metrics collected at successive time intervals, i.e., time series, which constitutes training the VAE for time series reconstruction under BRI, "the machine learning model 115 may be trained to process one or more performance metrics that are received and/or collected from the information technology system 140 at successive time intervals").
Per claim 6, Kumar combined with Barr and Sulem discloses claim 1, Kumar further teaching wherein the test set includes a set of features and timestamps (Kumar: ¶[0047]…Kumar’s sequences comprise a set of performance metrics (features) collected at successive time intervals (timestamps), which constitutes a test set including a set of features and timestamps under BRI, "Table 2 below depicts a sequence of six different performance metrics (e.g., M1, M2, M3, M4, M5, and M6) that have been received and/or collected from the information technology system 140 at six successive time intervals (e.g., T0, T1, T2, T3, T4, and T5)").
Per claim 7, Kumar combined with Barr and Sulem discloses claim 1. Barr further teaches wherein computing the reconstruction errors is performed using a mean square error technique (Barr: Section Appendix A.4…Barr measures reconstruction using a mean squared error term in the autoencoder loss function, which constitutes computing reconstruction error using a mean square error technique under BRI, "This weight was applied in the loss function, a combination of mean squared error and categorical or binary cross entropy"). The rationale to combine Barr with Kumar is the same as the parent claim.
Per claim 8, Kumar combined with Barr and Sulem discloses claim 1, Kumar further teaching wherein, when the reconstruction error is above the threshold, data corresponding to the reconstruction error is considered to be anomalous (Kumar: ¶[0011]…Kumar detects the anomaly when the reconstruction error exceeds the threshold value, which constitutes considering data anomalous when the reconstruction error is above the threshold under BRI, "the occurrence of the anomaly at the information technology system may be detected based at least on the reconstruction error exceeding a threshold value").
Per claim 9, Kumar combined with Barr and Sulem discloses claim 1, Kumar further teaching wherein the previously trained anomaly detection VAE model is an unsupervised model (Kumar: ¶[0038]…Kumar trains the VAE using only sequences of non-anomalous performance metrics and without anomaly labels, which a PHOSITA would understand to be unsupervised training, and which constitutes an unsupervised model under BRI, "the autoencoder may be trained based on training data that includes one or more sequences of non-anomalous performance metrics but excludes sequences of anomalous performance metrics. Trained in this manner, the autoencoder may be able to encode sequences of non-anomalous performance metrics with minimal reconstruction error").
Per claim 15, Kumar combined with Barr and Sulem discloses claim 10. Barr further teaches wherein said each synthetic sample’s neighbor in the latent space is an immediate neighbor (Barr: Section 3.2…Barr samples successive interpolated points along the latent line such that each sample’s adjacent sampled point is its immediate neighbor, which constitutes an immediate neighbor under BRI, "The interpolated points are then decoded by the UVAE and passed to the classifier to obtain a classification score"). The rationale to combine Barr with Kumar is the same as the parent claim.
Per claim 16, Kumar combined with Barr and Sulem discloses claim 10, Kumar further teaching wherein the previously trained anomaly detection VAE model is trained for time series reconstruction (Kumar: ¶[0047]…Kumar’s VAE reconstructs sequences of performance metrics collected at successive time intervals, i.e., time series, which constitutes training for time series reconstruction under BRI, "the machine learning model 115 may be trained to process one or more performance metrics that are received and/or collected from the information technology system 140 at successive time intervals").
Per claim 17, Kumar combined with Barr and Sulem discloses claim 10, Kumar further teaching wherein the test set includes a set of features and timestamps (Kumar: ¶[0047]…Kumar’s sequences comprise performance metrics (features) collected at successive time intervals (timestamps), which constitutes a test set including features and timestamps under BRI, "Table 2 below depicts a sequence of six different performance metrics (e.g., M1, M2, M3, M4, M5, and M6) that have been received and/or collected from the information technology system 140 at six successive time intervals (e.g., T0, T1, T2, T3, T4, and T5)").
Per claim 19, Kumar combined with Barr and Sulem discloses claim 18, Kumar further teaching wherein the previously trained anomaly detection VAE model is trained for time series reconstruction (Kumar: ¶[0047]…Kumar’s VAE reconstructs sequences of performance metrics collected at successive time intervals, which constitutes training for time series reconstruction under BRI, "the machine learning model 115 may be trained to process one or more performance metrics that are received and/or collected from the information technology system 140 at successive time intervals").
Per claim 20, Kumar combined with Barr and Sulem discloses claim 18, Kumar further teaching wherein the previously trained anomaly detection VAE model is an unsupervised model (Kumar: ¶[0038]…Kumar trains the VAE using only non-anomalous sequences and without anomaly labels, which constitutes an unsupervised model under BRI, "the autoencoder may be trained based on training data that includes one or more sequences of non-anomalous performance metrics but excludes sequences of anomalous performance metrics. Trained in this manner, the autoencoder may be able to encode sequences of non-anomalous performance metrics with minimal reconstruction error").
Claims 11-14 are rejected under 35 USC 103 as being unpatentable over Kumar in view of Barr and Sulem, and further in view of Anomaly Detection of Time Series with Smoothness-Inducing Sequential Variational Auto-Encoder to Li et al. (hereinafter Li).
Per claim 11, Kumar combined with Barr and Sulem discloses claim 10. Kumar combined with Barr and Sulem does not expressly disclose, but Li does teach: wherein weights of the previously trained anomaly detection VAE model are randomly initialized (Li: Section IV, Algorithm 1…Li’s variational-autoencoder training procedure begins by randomly initializing the network parameters, which constitutes randomly initializing the weights of the anomaly-detection VAE under BRI, "Randomly initialize θ, Φ; Initialize h0 = 0").
Per claim 12, Kumar combined with Barr and Sulem discloses claim 10. Kumar combined with Barr and Sulem does not expressly disclose, but Li does teach: wherein the reconstruction errors are backpropagated through a network of the previously trained anomaly detection VAE model (Li: Section II…Li trains the variational-autoencoder network by standard backpropagation of the reconstruction objective through the network, which constitutes backpropagating the reconstruction errors through the network of the VAE under BRI, "the network can be learned by standard backpropagation").
Per claim 13, Kumar combined with Barr and Sulem discloses claim 10. Kumar combined with Barr and Sulem does not expressly disclose, but Li does teach: wherein the threshold operates as a trade-off between a decrease to a false alarm rate and an increase to a true positive rate (Li: Section V…Li computes true positive rate and false positive rate as a function of the detection threshold and explains that a trade-off is needed because the detection characteristics vary with the threshold, which constitutes the threshold operating as a trade-off between the false alarm (false positive) rate and the true positive rate under BRI, "we compute precision, recall, true positive rate (TPR), false positive rate (FPR) at time series point level. Trade-off is needed as the detection characteristics vary by different threshold α").
Per claim 14, Kumar combined with Barr and Sulem discloses claim 10. Kumar combined with Barr and Sulem does not expressly disclose, but Li does teach: wherein the threshold is defined using a z-score in a normal distribution (Li: Section II…Li defines the detection threshold by thresholding the reconstruction-probability scores produced by its Gaussian (normally distributed) generative model, "the anomalies are detected by thresholding reconstruction probability"; and a PHOSITA would have found it obvious to define such a threshold using a z-score of the normal distribution of scores, as z-score thresholding is a well-known statistical technique for flagging outliers that yields predictable results, per KSR rationale (A)).
As it pertains to claims 11-14, Kumar, Barr, Sulem and Li are analogous art because they are from the same field of endeavor, specifically variational-autoencoder-based anomaly detection of time-series data and the explanation of such models. They address the same problem of detecting anomalies with a VAE reconstruction-error model and configuring its training and detection threshold.
Before the effective filing date of the claimed invention, it would have been obvious to a PHOSITA to implement the VAE anomaly detector of Kumar (as modified by Barr and Sulem) using the conventional VAE training and thresholding details taught by Li — randomly initializing the network weights, learning the network by backpropagation, and setting the detection threshold as a trade-off between the true-positive and false-alarm rates.
The suggestion/motivation for doing so is provided by Li itself, which teaches that its variational-autoencoder anomaly detector “can be learned by standard backpropagation” (Li: Section II…"the network can be learned by standard backpropagation") and that a threshold “trade-off is needed as the detection characteristics vary by different threshold” (Li: Section V…"we compute precision, recall, true positive rate (TPR), false positive rate (FPR) at time series point level. Trade-off is needed as the detection characteristics vary by different threshold α"); applying these conventional training and thresholding techniques to Kumar’s VAE detector yields the predictable result of a properly trained and calibrated anomaly detector, per KSR rationale (A) combining prior-art elements according to known methods to yield predictable results.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ALAN CHEN whose telephone number is (571)272-4143. The examiner can normally be reached M-F 10-7.
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, Kamran Afshar can be reached at (571) 272-7796. 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.
/ALAN CHEN/Primary Examiner, Art Unit 2125
Prosecution Insights