Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined
under the first inventor to file provisions of the AIA .
Response to Arguments
The applicant argument filed 8/10/26 regarding the 35 U.S.C. § 101 is persuasive
and therefore, the examiner withdraws the 35 U.S.C. § 101 rejection.
The Applicant's arguments filed 8/10/26, regarding the prior art rejections have
been fully considered but they are not persuasive.
Applicant’s arguments regarding the limitation “generating a reduced seismic dataset by applying an unsupervised machine learning model to cluster the seismic data, the reduced seismic dataset having less data than the seismic data, the reduced seismic dataset including the amount of information about the subsurface formation” have been considered but nor persuasive. Applicant argues that Zhu merely uses SOM-K-means clustering to classify seismic facies and does not disclose generating the claimed reduced seismic dataset. Under the broadest reasonable interpretation consistent with the specification, claim 1 does not require any particular data reduction technique, reduction ratio, compression algorithm, representative sample selection procedure, or requirement that every item of information contained in the original seismic dataset be retained. The claim broadly requires applying an unsupervised machine-learning model to cluster seismic data and generating, as a result of the claimed processing, a data set containing less data than the original seismic data while still including information about the subsurface formation.
Zhu explicitly teaches applying an unsupervised SOM-K-means clustering algorithm to seismic information for seismic facies analysis. Zhu explains that its multi-attribute SOM-K-means algorithm performs two-stage clustering using multiple geological attributes and that the algorithm “can effectively extract complementary features from the multiple attribute volumes and comprehensively use different attributes to improve the recognition ability of seismic facies” (see Abstract). Thus, Zhu is not merely unrelated classification art; Zhu applies unsupervised clustering directly to seismic information containing geological information concerning the subsurface and recognizes that information according to similarities represented by the resulting clusters. Zhu therefore supplies the claimed use of an unsupervised machine-learning clustering technique in the seismic data context and establishes that clustered seismic information continues to represent information concerning the subsurface formation.
The examiner does not rely on Zhu as explicitly stating that its clustering output is a “reduced seismic dataset having less data than the seismic data”. Rather, the rejection is based on the combined teachings of the references and what those teachings would have suggested to a person having ordinary skill in the art. Hu teaches seismic data processing and machine-learning based prediction of subsurface velocity-model information, and Hu further recognizes full waveform inversion as known technique for constructing depth velocity models. Feng teaches a data driven full waveform inversion approach and addresses the substantial computational burden associated with conventional FWI. Zhu teaches that seismic data can be organized by unsupervised SOM-K-means clustering into groups characterized by seismic/geological information. In view of these teachings, one having ordinary skill in the art would have recognized that, after clustering the seismic data according to Zhu, fewer data associated with the resulting clusters could be used for the computationally expensive inversion/training operations rather than processing the entirety of the original seismic dataset.
It would therefore have been obvious to modify the seismic processing and machine learning workflow of Hu and Feng to apply Zhu’s unsupervised seismic clustering before the computationally intensive FWI processing and to perform the FWI using a reduced amount of the clustered seismic data because Feng recognizes the computational burden associated with FWI and Zhu provides a known technique for organizing seismic information into groups according to similar seismic/geological characteristics, thereby reducing the quantity of seismic data subjected to computationally intensive FWI while retaining seismic information representative of the subsurface formation for subsequent model training and velocity prediction. The claimed result does not require a particular manner of choosing the reduced data from the clusters, and the mere fact that Zhu employs clustering for seismic facies analysis does not teach away from using the same known clustering technique to organize seismic data before selecting a smaller amount of data for FWI. The proposed modification uses Zhu’s clustering technique according to its established function, grouping seismic information having related characteristics, and applies that organization to known computational problem identified by Feng.
Applicant additionally argues that Zhu is silent as to the requirement that “the reduced seismic dataset include the amount of information about the subsurface formation”.
This argument is not persuasive under the broadest reasonable interpretation of the claim. The claim does not quantify the recited “amount of information”, require information equivalence between the original and reduced datasets, or require retention of every datum in the original seismic dataset. Zhu’s clustered seismic information is generated from multiple geological/seismic attributes specifically for recognition of seismic facies and therefore continues to contain information concerning the subsurface formation. Accordingly, the fact that the proposed dataset contains fewer data items does not mean that it ceases to include the recited information about the subsurface formation.
Accordingly, Applicant’s argument that Zhu does not, by itself, explicitly disclose every word of the disputed limitation does not establish non-obviousness, because the rejection is based on the collective teachings of Hu, Feng, and Zhu and reasoned modification that those teachings would have suggested to one having ordinary skill in the art before the effective filling date of the claimed invention.
Under 35 USC § 101
Claims 1, 11, and 17 are directed to patent-eligible subject matter under 35 U.S.C.
§101. Although claims 1, 11, and 17 may recite or involve a mathematical concept, the claims integrate any mathematical concept into a practical application. The claims recite generating a reduced seismic dataset by applying an unsupervised machine learning model to cluster seismic data, performing full wave inversion (FWI) using the reduced seismic dataset, and training a supervised machine learning model using the resulting inversion models and forward modeling data for predicting seismic velocities. These limitations, considered as an ordered combination, provide a particular technological implementation in which computationally intensive FWI is performed on a reduced amount of seismic data and the resulting information is used to train the supervised machine learning model. Thus, the claims reflect an improvement to computerized seismic-data processing rather than merely using a computer as a tool to perform a mathematical concept. Accordingly, the recited judicial exception is integrated into a practical application and claims 1-20 are patent eligible under 35 U.S.C. §101.
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.
Claims 1-3, 7-13 and 17-20 are rejected under 35 U.S.C. 103 as being
unpatentable over Hu et al. (Pub. No. US 2024/0377546) (hereinafter Hu) in view of Feng et al. (NPL: “Multiscale Data-driven Seismic Full-waveform Inversion with Field Data Study” (IEEE (2023)) and further in view of Zhu et al. (NPL: “Seismic Facies Analysis Using the Multiattribute SOM-K-Means Clustering”, Hindawi (2022)) (hereinafter Zhu).
As per claims 1, 11 and 17, Hu teaches obtaining seismic data from a subsurface formation, the seismic data including an amount of information about the seismic formation (see ¶¶ [0003]-[0005] and [0051]-[0055], i.e., receiving seismic data from a subsurface formation including baseline seismic datasets and monitoring seismic datasets representing subsurface information). Hu further discloses generating velocity/property model predictions from seismic data using supervised machine learning workflows. In particular, Hu discloses extracting seismic signal changes between monitoring datasets and baseline datasets (see ¶ [0064]), generating labels associating seismic data changes with velocity model (see ¶ [0066]), training a machine learning model using the generated labels (¶ [0067]), and generating synthetic training data “by forward modeling” (¶ [0068]). Hu further discloses predicting velocity model changes by feeding seismic-derived data into a trained neural network (see ¶¶ [0069]-[0071]).
Although Hu fails to explicitly teach generating a reduced seismic dataset using unsupervised clustering or performing full waveform inversion (FWI) on such reduced seismic datasets, Hu recognizes FWI as a known seismic velocity model building technique, stating that “There are at least two main depth velocity model building technologies, velocity tomography and full waveform inversion” (see ¶ [0057]).
Feng, however, discloses a multiscale data-driven seismic full-waveform inversion framework in which supervised conventional neural networks are trained using seismic waveform data and corresponding velocity models generated from forward seismic modeling and inversion processes (see Abstract). Feng explicitly teaches that seismic full-waveform inversion estimates subsurface velocity models from seismic waveform data and supervised deep-learning models may be trained using forward modeled seismic responses paired with velocity models in order to predict seismic velocities directly from seismic data while reducing computational burden relative to conventional iterative FWI (see Abstract and page 5, section B. Multiscale InversionNet).
Zhu disclose “multiattribute SOM-K-means clustering algorithm” and “SOM-based K-means clustering” applied to seismic datasets for unsupervised seismic facies analysis (see pages 2-3, section 2. Related Work), where Zhu teaches clustering seismic attributes/traces into representative groups/classes using unsupervised k-means clustering.
It would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to modify the supervised seismic velocity prediction workflow of Hu with the FWI-based supervised inversion training techniques of Feng and further with the unsupervised seismic clustering techniques of Zhu because Feng teaches that supervised neural network velocity prediction may be trained using FWI-generated seismic/velocity model pairs, thereby enabling efficient seismic velocity prediction from seismic waveform data, and because Zhu teaches that unsupervised clustering may group similar seismic traces/attributes into representative clusters, thereby reducing computational burden associated with computationally expensive FWI and supervised neural network training. Such modification would merely involve combining known seismic inversion and machine learning preprocessing techniques according to known methods to obtain predictable improvements in processing efficiency and seismic velocity estimation accuracy. (See KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398 (2007)).
As per claims 2, 12 and 18, the combination of Hu, Feng and Zhu teaches the system as stated above. Hu further teaches generating improved subsurface imaging and velocity models from the predicted velocity changes (see ¶¶ [0057] and [0069]). Specifically, Hu discloses that the resulting velocity model “is generated more efficiently and/or accurately than past velocity models” and may be employed for seismic imaging and monitoring (see ¶ [0069]). Feng further teaches that the deep-learning FWI workflow substantially reduces computational burden relative to conventional iterative FWI while maintaining high-resolution velocity reconstruction (see pages 7-8, section V. Results). It would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to employ the trained supervised FWI workflow of Feng within the imaging workflow of Hu because the supervised neural-network inversion reduces the computational cost associated with the conventional FWI, thereby enabling improved resolution seismic imaging with reduced computational cost.
As per claims 3 and 13, the combination of Hu, Feng and Zhu teaches the system as stated above. Zhu further discloses unsupervised k-means clustering applied to seismic datasets using “SOM-K-Means clustering algorithm” (see pages 2-3, section 2. Related Work), where Zhu teaches clustering seismic attributes/traces into representative groups/classes using unsupervised k-means clustering.
As per claim 7, the combination of Hu, Feng and Zhu teaches the system as stated above. Hu further teaches using amplitude-related and kinematic/phase-related seismic signal information for velocity model estimation. Specifically, Hu discloses that seismic signals may exhibit “time shift, squeezing, stretching” and “amplitude change” used for “velocity model change estimation” (see ¶ [0060]). Hu further teaches preprocessing including “spectral shaping” and signal balancing (see ¶ [0061]). Such kinematic/time-shift and spectral information corresponds to phase and amplitude information conventionally extracted from seismic data.
As per claim 8, the combination of Hu, Feng and Zhu teaches the system as stated above. Zhu further teaches clustering seismic data into representative groups/classes/facies for downstream processing (see pages 2-3, section 2. Related Work). Once seismic data are clustered into representative groups, selecting representative samples from each cluster for downstream model training constitutes a routine and well-known data science implementation technique for reducing training burden while maintaining representative coverage of clustered data. It would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to randomly select samples from each cluster because representative sampling from clustered data reduces the amount of training data processed while preserving representative subsurface information, thereby reducing computational expense associated with FWI and neural-network training.
As per claims 9 and 19, the combination of Hu, Feng and Zhu teaches the system as stated above. Hu further teaches iterative refinement of the machine learning model and predicted velocity models. Specifically, Hu discloses that prediction differences/residuals “may be back-propagated to the machine learning model…such that subsequent predictions result in a prediction…closer to the ground truth” (see ¶ [0067]). Hu additionally teaches that geologic interpretations and models “may be refined in an iterative fashion” using feedback loops (see ¶ [0079]).
As per claims 10 and 20, the combination of Hu, Feng and Zhu teaches the system as stated above. Hu further teaches forward modeling and differences/residuals determination during supervised training. Specifically, Hu discloses that “synthetic training data may be generated…by forward modeling” (see ¶ [0068]) and that “differences (residual)” between predicted and ground-truth models are determined and back-propagated during iterative training (see ¶ [0067]).
Claims 4 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Hu
in view of Feng and further in view of Zhu and Engquist et al. (NPL: “Optimal Transport for Seismic Full Waveform Inversion”, Texas Consortium for Computational Seismology, (2016)) hereinafter Engquist.
As per claims 4 and 14, the combination of Hu, Feng and Zhu teaches the system
as stated above. The combination fails to teach that generating the inversion models and the forward modeling data comprises a reinforcement machine learning model having a long-term reward comprising a misfit reduction of a current model as compared with a model from a prior iteration.
Engquist, however, teaches iterative full waveform inversion optimization in which
model parameters are iteratively updated according to reduction of mismatch between modeled seismic responses and measured seismic data. In particular, Engquist teaches that “unknown coefficients…are adjusted to minimize the mismatch with the measured data” (see page 1, Abstract and Introduction). Engquist therefore teaches iterative model optimization based on reduction of seismic inversion misfit between successive model iterations, where the mismatch-reduction objective functions as an optimization/reward criterion for improving the inversion model over prior iterations (see also pages 18-20). It would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to incorporate the iterative misfit-reduction optimization techniques of Engquist into the FWI-based seismic inversion workflow of Hu and Feng because iterative optimization of seismic responses and measured seismic data improves convergence and inversion
accuracy, thereby enabling improved seismic velocity prediction through iterative refinement of inversion models relative to prior iterations. Such modification merely applies a known optimization criterion to a known seismic inversion workflow in order to obtain predictable improvements in seismic inversion performance and model convergence. (See KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398 (2007)).
Claims 5 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over Hu
in view of Feng and further in view of Zhu and Bod
o
´
et al. (NPL: “Active Learning with Clustering”, JMLR: Workshop and Conference Proceedings 16 (2011) 127–139 Workshop on Active Learning and Experimental Design) hereinafter Bod
o
´
.
As per claims 5 and 15, the combination of Hu, Feng and Zhu teaches the system
as stated above. The combination fails to teach that generating a reduced seismic dataset comprises applying an active learning model to determine a number of clusters to be formed by the unsupervised machine learning model.
Bod
o
´
, however, integrating active learning with k-means clustering workflows,
including determining clustering parameters such as the number of clusters and selecting samples from clusters for learning operations (see page 134, i.e., “the K-means clustering has two parameters: “the first k denotes the number of clusters formed from the unlabeled and test data, while klab is the number of clusters containing labeled points”. It would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to employ the active-learning cluster-selection techniques of Bod
o
´
within the seismic clustering workflow of Zhu and seismic inversion/training workflow of Hu and Feng because active-learning guided determination of clustering parameters improves representative sample selection and training efficiency, thereby reducing computational burden associated with computationally expensive FWI and supervised seismic model training while maintaining representative seismic data coverage. Such modification merely combines known clustering optimization techniques with known seismic clustering and inversion workflows to achieve predictable computational efficiency improvements. (See KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398 (2007)).
Claims 6 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Hu
in view of Feng and further in view of Zhu and Bod
o
´
and Umargono et al. (NPL: “K-Means Clustering Optimization Using the Elbow Method and Early Centroid Determination Based on Mean and Median Formula”, Advances in Social Science, Education and Humanities Research, volume 474 Proceedings of the 2nd International Seminar on Science and Technology (ISSTEC 2019) (hereinafter Umargono).
As per claims 6 and 16, the combination of Hu, Feng, Zhu, and Bod
o
´
teaches the system as stated above. The combination fails to teach that the active learning model determines the number of clusters based on a root mean square of each cluster formed by the unsupervised machine learning model.
Umargono, however, teaches evaluating clustering quality and determining an appropriate number of clusters using per-cluster sum squared error (SSE) metrics. Specifically, Umargono teaches “calculate the sum squared error (SSE) per cluster” and determining “the best number of clusters by comparing the difference of the sum of square error (SSE) of each cluster” (see page 124, section Data Allocation). It would have been obvious to one having ordinary skill in the art before the effective filling date of the claimed invention to employ RMS/RMSE-based clusters evaluation metrics when determining the number of seismic-data clusters within the active-learning clustering workflow because RMS/RMSE metrics were conventional derivative forms of SSE/MSE cluster-quality metrics used to evaluate cluster compactness and determine suitable cluster configurations in k-means clustering systems, thereby enabling improved representativeness and stability of clustered seismic datasets while reducing computational burden associated with computationally expensive FWI and supervised seismic model training. Such modification merely applies a known cluster-quality optimization technique to a known seismic clustering workflow in order to obtain predictable improvement in cluster data selection and inversion training efficiency. (See KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398 (2007)).
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time
policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Contact information
Any inquiry concerning this communication or earlier communications from the
examiner should be directed to MOHAMED CHARIOUI whose telephone number is (571)272-2213. The examiner can normally be reached Monday through Friday, from 9 am to 6 pm.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Andrew Schechter can be reached on (571) 272-2302. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
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.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
Mohamed Charioui
/MOHAMED CHARIOUI/Primary Examiner, Art Unit 2857
Prosecution Insights