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
Applicant's arguments filed 1/14/2026 have been fully considered but they are not persuasive.
Regarding applicants arguments for 35 U.S.C. 101 on page 7 of remarks, applicant argues “In the same Office Action, the Examiner has rejected claims 1-14 under 35 USC 101 "because the claimed invention is directed to abstract idea without significantly more." Specifically, the Examiner states that the claim limitations "recite a mental process" for some of the claim limitations and "a mathematical process" for other limitations …
Claims 1 and 8 are believed to now recite limitations that are inexorably tied to a specific
computational environment, and performing activities that are not merely done on a generic
computer system. As such, claims 1 and 8 are believed to now recite eligible subject matter, thus
rendering the rejections under 35 USC 101 moot.” However the applicant argues amended limitations and the amended limitation have not been examined. See updated 35 U.S.C. 101 rejection.
Regarding applicants arguments for 35 U.S.C. 103 on page 8-9 of remarks, applicant argues “Applicants respectfully traverse the rejection, in Calicut, one distribution is provided along with one range. There is no indication that data distributions and/or ranges are responsive to the domain involved. As such, claims 1 and 8 would be allowable over the cited art. However, in order to make this crystal clear, and to further prosecution, claim 1 and 8 have been amended to recite" validating the set of input data by: comparing range of the input data against ranges allowed by the domain; comparing distribution of the input data against a distribution curve allowed by the domain; correlating data fields within the input data; and identifying data within the data fields that are values outside of boundaries set by the correlation… Further, Claim 1 has been amended to recite that the domain is identified by a clustering model, in order to reflect the scope of original claim 8. None of the cited art discloses determining domain via clustering models. As none of the cited art does validations against ranges allowed by a domain, distribution curves as allowed by domain, or data field correlations to identify data outside of allowed boundaries, or domain identification via clustering models, Applicants believe claims 1 and 8, as amended are believed to be allowable.” The applicant relies upon amended limitations to overcome the prior art. However the amended limitations have not been examined rendering the argument moot and not persuasive.
Specification
The use of the term Intel SGX, AWS, and Azure which are a trade name or a mark used in commerce, has been noted in this application. The term should be accompanied by the generic terminology; furthermore the term should be capitalized wherever it appears or, where appropriate, include a proper symbol indicating use in commerce such as ™, SM , or ® following the term.
Although the use of trade names and marks used in commerce (i.e., trademarks, service marks, certification marks, and collective marks) are permissible in patent applications, the proprietary nature of the marks should be respected and every effort made to prevent their use in any manner which might adversely affect their validity as commercial marks.
Claim Rejections - 35 USC § 112 (b)
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.
Claim 8 contains the trademark/trade name Intel SGX in line 2. Where a trademark or trade name is used in a claim as a limitation to identify or describe a particular material or product, the claim does not comply with the requirements of 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph. See Ex parte Simpson, 218 USPQ 1020 (Bd. App. 1982). The claim scope is uncertain since the trademark or trade name cannot be used properly to identify any particular material or product. A trademark or trade name is used to identify a source of goods, and not the goods themselves. Thus, a trademark or trade name does not identify or describe the goods associated with the trademark or trade name. In the present case, the trademark/trade name is used to identify/describe “sequestered computing node” and, accordingly, the identification/description is indefinite. All dependent claims inherit the issue.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 rejected under 35 U.S.C. 101 because the claimed invention is directed to abstract idea without significantly more. The claim(s) recite(s) significantly more. The subject matter eligibility test for products and process is describe below for claim 1 in view of dependent claims.
Regarding claim 1:
Step 1: Is the claim to a process machine manufacture or composition of matter?
Yes – Claim 1 recites a method, which is a process that falls under the statutory categories.
Step 2A Prong 1: Does the claim recite an abstract idea, law of nature, or natural phenomenon?
Yes – The claim recites the following:
“validating the set of input data by: comparing range of the input data against ranges allowed by the domain;”- The limitations recites a mental process of validating the set of inputs by comparing (see MPEP 2106.04(a)(2)III).
“comparing and distribution of the input data against a distribution curve allowed by responsive to the domain;” – The limitation recites a mathematical process of comparing a distribution of input data against the allowed domain (See MPEP 2106.04(a)(2)I).
“correlating data fields within the input data; and identifying data within the data fields that are values outside of boundaries set by the correlation;” The limitation recites a mathematical process of correlating data and the mental process of identifying data within the data fields outside of the correlation (See MPEP 2106.04(a)(2)I) and See MPEP 2106.04(a)(2)III).
“when the validating fails, transforming the input data by at least one of a range transformation, a distribution transformation and a machine learning (ML) transformation;” – The limitation recites a mathematical process of transforming data by distribution transformation (See MPEP 2106.04(a)(2)I).
Step 2 Prong 2: Does the claim recite additional elements that integrate the judicial exception into a particular application? No –
The claim includes the additional element(s):
“A computerized method of processing input data comprising: providing a set of input data into a sequestered computing node”
The additional elements fall under Insignificant Extra-Solution Activity as mere data gathering. See MPEP 2106.5(g). The judicial exemptions do not integrate into a practical application nor provide an improvement. The process does not provide an inventive concept nor provides a practical application.
“identifying a domain for [[a]] the set of input data using an AI clustering model;
The additional elements fall under “apply it” as using a generic computer to implement use an AI clustering model to identify a domain. See Mere Instructions to Apply an Exemption (see MPEP 2106.05(f)).
“wherein the sequestered computing node prevents access by any party to computing environment of the sequestered computing node, and generates an encrypted artifact containing results of all processing within the sequestered computing node;”
The additional elements fall under “apply it” as using a generic computer to implement a sequestered computing environment and generate an encrypted artifact. See Mere Instructions to Apply an Exemption (see MPEP 2106.05(f)).
“iteratively validating and transforming the input data until validation passes; and processing the validated data using at least one algorithm to generate the results.”
The additional elements fall under “apply it” as using a generic computer to iteratively validate and process the data using one algorithm to generate the results. See Mere Instructions to Apply an Exemption (see MPEP 2106.05(f)).
Step 2B: Does the claim recite additional elements that amount to significantly more than the judicial exception?
No - The claim does not include additional elements that are sufficient to amount to a significantly more than the judicial exemption. As an order whole, the claim is directed processing data and performing validations and transformations to validate. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of processing iteratively validating and transforming fall under using generic computer to apply an exemption. The method does not improve on the function of a computer, transforms an article into another article, nor is it applied by a particular machine, making the claim not patent eligible.
Regarding claim 2:
Step 2A Prong 2, Step 2B: The additional element(s):
“The method of claim 1, wherein the domain is at least one of pathology dependent and financial use case dependent.”
The additional elements fall under Insignificant Extra-Solution Activity. See MPEP 2106.5(g). The judicial exemptions do not integrate into a practical application nor provide an improvement. The process does not provide an inventive concept nor provides a practical application.
Regarding claim 3:
Step 2A Prong 2, Step 2B: The additional element(s):
“The additional elements fall under Insignificant Extra-Solution Activity. See MPEP 2106.5(g). The judicial exemptions do not integrate into a practical application nor provide an improvement. The process does not provide an inventive concept nor provides a practical application”
The additional elements fall under Insignificant Extra-Solution Activity. See MPEP 2106.5(g). The judicial exemptions do not integrate into a practical application nor provide an improvement. The process does not provide an inventive concept nor provides a practical application.
Regarding claim 4:
Step 2A Prong 2, Step 2B: The additional element(s):
“The method of claim 1, wherein the ML transform is trained on domain specific datasets.
The additional elements fall under Insignificant Extra-Solution Activity. See MPEP 2106.5(g). The judicial exemptions do not integrate into a practical application nor provide an improvement. The process does not provide an inventive concept nor provides a practical application.
Regarding claim 5:
Step 2A Prong 1:
“The method of claim 1, wherein the validation includes comparing the input data” – The limitation recites a mathematical calculation by determining an expected range and distribution curve (see MPEP 2106.04(a)(2).
Step 2A Prong 2, Step 2B: The additional element(s):
No additional elements. The judicial exemptions do not integrate into a practical application nor provide an improvement. The process does not provide an inventive concept nor provides a practical application.
Regarding claim 6:
Step 2A Prong 1:
“The method of claim 5, wherein the validation of expected distribution is a curve which fits within two standard deviations of the expected distribution.” - The limitation recites a mathematical process by validating the expected distribution by a curve within two standard deviations (see MPEP 2106.04(a)(2)I.
Step 2A Prong 2, Step 2B: The additional element(s):
No additional elements. The judicial exemptions do not integrate into a practical application nor provide an improvement. The process does not provide an inventive concept nor provides a practical application.
Regarding claim 7:
Step 2A Prong 1:
“The method of claim 5, wherein the validation of expected distribution is a curve which fits within a configurable threshold of standard deviations of the expected distribution.” - The limitation recites a mathematical process by validating the expected distribution by a curve which fits within a configurable threshold of standard deviations of the expected distribution (see MPEP 2106.04(a)(2)I.
Step 2A Prong 2, Step 2B: The additional element(s):
No additional elements. The judicial exemptions do not integrate into a practical application nor provide an improvement. The process does not provide an inventive concept nor provides a practical application.
Claims 8-14 recite a system and are analogous to the method of claims 1-7. Therefore, the rejections of claim 1-7 above applies to claims 8-14.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1, 2, 3, 4, 8, 9, 10 and 11 are rejected under 35 U.S.C. 103 as being unpatentable over Kunkel, Roland, et al. "Tensorscone: A secure tensorflow framework using intel sgx." arXiv preprint arXiv:1902.04413 (2019) (“Kunkel”) in view of Callcut et al. (US20200311300A1) (“Callcut”) and further in view of K. Stacke, G. Eilertsen, J. Unger and C. Lundström, "Measuring Domain Shift for Deep Learning in Histopathology," in IEEE Journal of Biomedical and Health Informatics, vol. 25, no. 2, pp. 325-336, Feb. 2021 (“Stacke”).
Regarding claim 1 and analogous claims 8, Kunkel A computerized method of processing input data comprising: providing a set of input data into a sequestered computing node, wherein the sequestered computing node prevents access by any party to computing environment of the sequestered computing node, and generates an encrypted artifact containing results of all processing within the sequestered computing node (Kunkel page 2,
PNG
media_image1.png
437
541
media_image1.png
Greyscale
2 BACKGROUND AND THREAT MODEL
2.1 Intel SGX and Shielded Execution
Intel Software Guard Extension (SGX) is a set of x86 ISA extensions for Trusted Execution Environment (TEE) [22]. SGX provides an abstraction of secure enclave—a hardware-protected memory region for which the CPU guarantees the confidentiality and integrity of the data and code residing in the enclave memory. The enclave memory is located in the Enclave Page Cache (EPC)—a dedicated memory region protected by an on-chip Memory Encryption Engine (MEE). The MEE encrypts and decrypts cache lines with writes and reads in the EPC, respectively. Intel SGX supports a call-gate mechanism to control entry and exit into the TEE.
Shielded execution based on Intel SGX aims to provide strong confidentiality and integrity guarantees for applications deployed on an untrusted computing infrastructure [15, 17, 39, 52, 60]. Our work builds on the SCONE [15] shielded execution framework. In the SCONE framework, the applications are statically compiled and linked against a modified standard C library (SCONE libc). In this model, application’s address space is confined to the enclave memory, and interaction with the untrusted memory is performed via the system call interface. In particular, SCONE runtime provides an asynchronous system call mechanism [55] in which threads outside the enclave asynchronously execute the system calls. Furthermore, it ensures memory safety [37] for the applications running inside the SGX enclaves [34]. Lastly, SCONE provides an integration to Docker for seamlessly deploying container images [A computerized method of processing input data comprising: providing a set of input data into a sequestered computing node, wherein the sequestered computing node prevents access by any party to computing environment of the sequestered computing node,].
Page 4 3.3 TensorSCONE Controller para 2 and 3,
File system shield. The file system shield protects confidentiality and integrity of data files. Whenever the application would write a file, the shield either encrypts and authenticates, simply authenticates or passes the file as is. The choice depends on user-defined path prefixes, which are part of the configuration of an enclave. The shield splits files into chunks that are then handled separately. Metadata for these chunks is kept inside the enclave, meaning it is protected from manipulation. The secrets used for these operations are different from the secrets used by the SGX implementation. They are instead configuration parameters at the startup time of the enclave.
Network shield. TensorFlow applications do not inherently include end-to-end encryption for network traffic. Users who want to add security must apply other means to secure the traffic, such as a proxy for the Transport Layer Security (TLS) protocol. According to the threat model however, data may not leave the enclave unprotected, because the system software is not trusted. Network communication must therefore always be end-to-end protected. Our network shield wraps sockets, and all data passed to a socket will be processed by the network shield instead of the system software. The shield then transparently wraps the communication channel in a TLS connection on behalf of the user application. The keys for TLS are saved in files and protected by the file system shield [, and generates an encrypted artifact containing results of all processing within the sequestered computing node;].));
identifying a domain for [[a]] the set of input data using an AI clustering model (Kunkel Page 2, 2.2 Machine Learning using TensorFlow para 1, Machine learning approaches aim to find solutions to problems by automatically deducing the required domain knowledge from example datasets [53] [identifying a domain]. Particularly, statistical models are leveraged to allow an information retrieval system to generalize and learn domain knowledge in order to solve a specific task. Broadly speaking, the machine learning approaches can be distinguished: supervised, unsupervised and reinforcement learning. All forms have in common that they require data sets, a defined objective function, a model and a way to update the model according to new inputs. In our work, we focus on supervised learning, but our approach is generalizable to the other two types. An overview of the process can be seen in Figure 1 [for [[a]] the set of input data using an AI clustering model].);
Kunkel does not explicitly teach validating the set of input data by: comparing range of the input data against ranges allowed by the domain;
comparing and distribution of the input data against a distribution curve allowed by responsive to the domain;
correlating data fields within the input data;
and identifying data within the data fields that are values outside of boundaries set by the correlation;
when the validating fails, transforming the input data by at least one of a range transformation, a distribution transformation and a machine learning (ML) transformation;
iteratively validating and transforming the input data until validation passes; and processing the validated data using at least one algorithm to generate the results.
However Callcut teaches validating the set of input data by: comparing range of the input data against ranges allowed by the domain (Callcut para 0069, In some embodiments , the data preparation constraints include but are not limited to one or more of following : input data requirements and annotation protocol requirements . The input data requirements may include optimization and / or validation selection criteria for data assets to be run on the algorithm or model [validating the set of input data by] . The optimization and / or validation selection criteria define characteristics , data formats , and requirements for input data ( e.g. , external data ) to be usable in the models . The characteristics and requirements for the input data refer to the characteristics and requirements of data such that the data is usable to optimize and / validate the model . For example, an algorithm implemented by the model may need training data that accurately represents the environment that the model will operate in such as different ethnic or geographical data etc. , to create more a more generalizable algorithm . In some instances , the characteristics and requirements of the input data are defined based on : (i) the environment of the model , ( ii ) distribution of examples such as 50 % male and 50 % female , (iii) parameters and types of devices generating data ( e.g. , image data ) and/or measurements , (iv) variance versus bias models with high variance can easily fit into training data and welcome complexity but are sensitive to noise ; whereas models with high bias are more rigid , less sensitive to variations in data and noise , and prone to missing com plexities, (v) the task ( s ) implemented by the models such as classification , clustering , regression , ranking , and the like , or ( vi ) any combination thereof . For example , the characteristics and requirements of the data for models developed to predict the presence of a tumor using classification and predict the presence of a tumor using classification and clustering techniques may include a requirement for three dimensional imaging data and / or biomarker testing data from an equal mix of females and males between the ages of 30 and 60 used in the identification of such a tumor . The formatting refers to not only the file format of the data ( e.g. , all image data should be in a .jpeg file format ) , but also the consistency of the records themselves . For example , the constraints for the data format may define a standard system of nomenclature ( e.g. , provide standardized codes , terms , synonyms and definitions which cover anatomy , diseases , findings , procedures , microorganisms , substances , etc. ) for the data sets recognized by the models [comparing range of the input data against ranges allowed by the domain;]);
when the validating fails, transforming the input data [by at least one of a range transformation, a distribution transformation and a machine learning (ML) transformation] ((Callcut Para 0113 line 13-19, At block 1110 , data assets are identified as being available from a data host based on the input data requirements ( e.g. , the optimization and / or validation selection criteria ) for the data assets . The data assets may be identified by running one or more queries on data storage structures of one or more hosts based on the optimization and / or validation selection criteria [when the validating fails,].
Para 0114, At block 1115 , the data host is brought onboard ( if not previously brought onboard). The onboarding comprises confirming that the use of the data assets with the model is in compliance with data privacy requirements . At block 1120 , the data assets are curated within a data storage structure that is within infrastructure of the data host . The curating may comprises selecting the data storage structure from multiple data storage structures and provisioning the data storage structure within the infrastructure of the data host . The selection of the data storage structure may be based on a type of algorithm within the model , a type of data within the data assets , system requirements of the computing device , or a combination thereof . At block 1125 , the data assets are prepared within the data storage structure for processing by the model . The preparing the data assets may comprise applying one or more transforms to the data assets , annotating the data assets , harmonizing the data assets , or a combination thereof [transforming the input data]);
iteratively validating and transforming the input data until validation passes (Callcut
PNG
media_image2.png
571
759
media_image2.png
Greyscale
[iteratively],
Para 0077, At block 340 , a determination is made as to whether the annotated data assets are harmonized in accordance with an algorithm protocol ( e.g. , the training and / or validation constraints defined in block 310) [validating]. Data harmonization is the process of bringing together data sets of varying file formats , naming conventions , and columns , and transforming it into one cohesive data set . In some instances , the determination may be made by comparing the training and / or validation constraints to the harmonization of the annotated data assets . When the annotated data assets are harmonized in accordance with an algorithm protocol , then further harmonization is not required and the process continues at block 360. When the annotated data assets are not harmonized in accordance with an algorithm protocol , then harmonization is required and the process continues at block 345. At block 345 , the annotated data assets are harmonization as described in detail with respect to FIG . 8. In some instances , the algorithm protocol requires the data assets to be transformed into a specific format or modified in a specific manner for computation. Harmonization of the data may be performed to transform the data assets [and transforming the input data until validation passes]);
and processing the validated data using at least one algorithm to generate the results (Callcut Para 0081, At block 360 , the optimization and / or validation of the models is performed using the data assets . In some instances , the optimization comprises initializing the algorithms with predefined values or random values for weights and biases and attempting to predict an output with those values . In certain instances , the models are pre - trained by the algorithm developer , and thus the algorithms may already be initialized with weights and biases. In other instances , predefined values or random values for weights and biases may be defined by the algorithm developer in block 310 and populated into the algorithms at block 360. Thereafter , data assets and hyperparameters may be input into the algorithms , inferences or predictions may be computed , and testing or comparisons may be made to determine how accurately the trained models predicted the output [processing the validated data]. The optimization may include running one or more instances of training and / or testing with the data assets in an attempt to optimize performance of the models ( e.g. , optimize the weights and biases ) as described in detail with respect to FIG . 9 [using at least one algorithm to generate the results.]).
Kunkel and Callcut are considered to be analogous to the claim invention because they are in the same field of machine learning. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Kunkel to incorporate the teachings of Callcut to validate date use on the algorithm or model. Doing so to validate the machine learning model is able to achieved sufficient validation (Callcut para 0111, At block 1015 , performance or accuracy of the model is computed based on gold standard labels ( i.e. , ground truths ) and determination is made as to whether the model has been validated . For example , an algorithm designed to detect breast cancer lesions in mammograms can be validated on a set of mammograms that have been labeled by medical experts as either containing or not containing a lesion . The performance of the algorithm relative to this expertly labeled set of mammograms is the validation report . In some instances , feature selection , classification , and parameterization of the model are visualized ( e.g. , using area under curve analysis ) and ranked , according to validation criteria defined ( i.e. , criteria upon which validation of the model is determined ) by algorithm developer in block 310 of FIG . 3. In some instances , determining whether the model has been validated includes determining whether the model has satisfied validation termination criteria ( i.e. , criteria that defines whether a model has achieved sufficient validation).
Stacke teaches comparing and distribution of the input data against a distribution curve allowed by responsive to the domain (Stacke Page 327, B. Related Work para 7, Similar to previous work in domain adaptation and generalization, we look at the shift between different image domains. However, we are interested in studying its properties and how it can be measured reliably in histopathology applications. We do this in a statistical comparison between distributions of the representations within a neural network. Thus, the focus is on finding a suitable deep representation and comparing the statistics of this between datasets from different domains and for different training data transformations. This is a significant difference as compared to previous work in domain adaptation/generalization;
IV. Quantifying Domain Shift para 3, Building on the idea of [51], we propose the representation shift metric, which measures the differences in distributions of the learned feature representation [comparing and distribution], comparing the training set (source) to a dataset from a different domain (target). This way, we can quantify the domain shift, not in image space, but in the model specific latent representation space. In a “well-trained” model, a deeper convolutional layer spans a space which corresponds only to image features which are relevant for the specific task at hand, irrelevant features should have been discarded. Looking at the statistical distributions of activation values for each filter in one of the layers, we can see how the model has described a particular dataset in this latent representation. By comparing this description with the description of a second dataset we can measure statistical discrepancies D between them. If the model has succeeded in learning only relevant features, the distributions for each filter should be similar (i.e., small distances). If this is not the case, then the representation of the first dataset depends on features not present in the second dataset (or vice versa), likely caused by domain shift in the image domain. [of the input data against a distribution curve allowed by responsive to the domain]);
correlating data fields within the input data; and identifying data within the data fields that are values outside of boundaries set by the correlation (Stacke Page 331 D. Experiments para 1 The domain shift analysis experiments were designed as follows. Cross-dataset generalization for tumor classification was evaluated by training CNN models as tumor classifiers, detecting tumor vs. non-tumor patches. The training data was separated in two sets: the first consisted of slides scanned … Secondly, using the results from the first part, we heuristically investigate the correlation of the model accuracy with the presented metric for domain shift quantification, the representation shift, described in Section IV. Three different discrepancy metrics were evaluated: Wasserstein distance, Kullback-Leibler divergence, and Kolmogorov-Smirnov statistic [correlating data fields within the input data; ].
Page 331 D. Experiments para 3 Kullback-Leibler (KL) divergence is defined as
PNG
media_image3.png
46
323
media_image3.png
Greyscale
where P and Q are discrete probability distributions. Here, we define it as the divergence from the target domain to the source domain, where the discrete probability density functions are approximated from the samples with 100 values (range determined by the minimum and maximum values of the samples of the two distributions which are compared). A small constant (10−3) was added where the discrete distributions are zero, for numerical stability [and identifying data within the data fields that are values outside of boundaries set by the correlation;]);
[when the validating fails, transforming the input data] by at least one of a range transformation, a distribution transformation and a machine learning (ML) transformation (Stacke Page 330 C. Data Transformations, Three types of data transformations were included in the study (see Fig. 5 for example images). Color and intensity data augmentation is a standard component in the DNN training pipeline, which can prevent over-fitting and increase generalization. Staining normalization is a special-purpose technique that aim at normalizing differences between stained slides, by using a reference image from the source data to transform all images. Both these transformations can be seen as domain generalization strategies, since no target data is needed to define the augmentation/normalization. Style transfer using Cycle-GAN aim to transfer the “style” of one medical center to another. Style transfer is a domain adaptation technique, as sufficient data from both domains is needed in order to train the models to do the transformation [a machine learning (ML) transformation].
Page 330 Fig. 5.,
PNG
media_image4.png
303
592
media_image4.png
Greyscale
);
Kunkel and Callcut and Stacke are considered to be analogous to the claim invention because they are in the same field of machine learning. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Kunkel in view of Callcut to incorporate the teachings of Stacke to use a machine learning algorithm to transform data. Doing so to validate the machine learning model based on the data used during training to affirm the model performance(Stacke page 334 VII. Discussion Para 4, Before a CNN can be used in a clinical setting, it needs to be properly validated. This validation should include evaluation on data with more variance than included in the training set, to affirm model performance. However, it is not possible to cover all data variations that may occur. Using the representation shift it is possible to do both an initial control on a large amount of unlabelled data, and continuous monitoring of a deployed model. By monitoring a model’s data representation, differences in data statistics and model performance degradation can be detected.)
Regarding claim 2 and analogous claim 9, Kunkel in view of Callcut and Stacke teach the method of claim 1
Kunkel, Callcut and Stacke are combined with the same rationale as set forth above with respect to claim 1 and analogous 8.
Callcut further teaches wherein the domain is at least one of pathology dependent and financial use case dependent (Callcut Para. 0071, In some embodiments , the validation constraints include but are not limited to one or more of following : validation data selection criteria , validation termination criteria , and a validation report requirements . The validation data selection criteria may include selection criteria for the validation data set that can include any factors required to select an appropriate subset of the data for the application being developed . For example , in healthcare applications , cohort selection includes , but is not limited to clinical cohort criteria , demographic criteria , and data set class balance . In healthcare algorithm development , cohort studies are a type of medical research used to investigate the causes of disease and to establish links between risk factors and health outcomes in groups of people , known as a cohort . Retrospective cohort studies look at data that already exists and try to identify risk factors for particular conditions . In a prospective cohort study , researchers raise a question and form a hypothesis about what might cause a disease . Then the researchers observe a cohort over a period of time , to prove or disprove the hypothesis . Thus , the clinical cohort criteria may define a group of people that the data is to be obtained from for the study , the type of study ( e.g. , retrospective or prospective ) , risk factors that the group may have exposure to over a period of time , question / hypothesis to be solved and associated disease or condition , and / or other parameters that define criteria for the cohort study [wherein the domain is at least one of pathology dependent]).
Regarding claim 3 and analogous claim 10, Kunkel in view of Callcut and Stacke teach the method of claim 1
Kunkel, Callcut and Stacke are combined with the same rationale as set forth above with respect to claim 1 and analogous 8.
Callcut further teaches further comprising cleaning the input data (Callcut Para 0076, At block 335 , once the data assets are prepare for annotation , the data assets are annotated as described in detail with respect to FIG . 7. Each algorithm of the models may require data to be labeled in a specific way . For example , a breast cancer detection / screening system may require specific lesions to be ized and identified . Another example , would be gastro - intestinal cancer digital pathology , in which each image may need to be segmented and labeled by the type of tissue present (normal ,necrotic ,malignant, etc. ) . In some instances involving text or clinical data , annotation may include applying a labeling ontology to selected subsets of text and structured data . The annotation is performed ly to the data host in the secure capsule computing service . A key principle to the transformation and annotation processes is that the platform facilitates a variety of processes to apply and refine data cleaning and transformation algorithms [further comprising cleaning the input data], while preserving the privacy of the data assets, all without requiring data to be moved outside of then technical purview of the data host).
Regarding claim 4 and analogous claim 11, Kunkel in view of Callcut and Stacke teach the method of claim 1
Kunkel, Callcut and Stacke are combined with the same rationale as set forth above with respect to claim 1 and analogous 8.
Stacke further teaches wherein the ML transform is trained on domain specific datasets (Stacke page 330, Fig. 6
PNG
media_image5.png
628
580
media_image5.png
Greyscale
[on domain specific datasets]
Page 331 C. Data Transformations 1) Color and Intensity, line 13-18, Tumor classification on H&E stained images are less dependent on color information and more on pattern, making it possible to heavily augment the colors, while retaining high model accuracy. The augmentations were done online during training [wherein the ML transform is trained]).
Claim(s) 5 and 12 are rejected under 35 U.S.C. 103 as being unpatentable over Kunkel in view of Callcut and Stacke and further in view of Li (US20190236114A1) (“Li”).
Regarding claim 5 and analogous claim 12, Kunkel in view of Callcut and Stacke teach the method of claim 1.
Kunkel, Callcut and Stacke are combined with the same rationale as set forth above with respect to claim 1 and analogous 8.
Kunkel does not explicitly teach wherein the validation includes comparing the input data
However Li teaches wherein the validation includes comparing the input data(Li Para. 45, FIG. 5 is a schematic diagram of a curve corresponding to t distribution shown in FIG. 4. A shape of the t distribution curve is related to a value of the degree of freedom v. If the value of the degree of freedom v is smaller, the t distribution curve is flatter, a middle part of the curve is lower, and the two tails of the curve are higher. If the value of the degree of freedom v is larger, the t distribution curve is closer to a normal distribution curve. When the value of the degree of freedom v is infinite, the t distribution curve is a standard normal distribution curve. FIG. 5 is a schematic diagram of a curve corresponding to t distribution when a degree of freedom v is equal to 34 in FIG. 4. When t=2.032, as shown in FIG. 5, the corresponding P=0.05 is a sum of shadow areas on two sides of the t distribution curve. To be specific, a total area under the t distribution curve is 1, and a total area of the shadow areas is 0.05. When t>2.032, the corresponding shadow areas on two sides of the t distribution curve is smaller, in other words, the probability P is smaller.
Para 0060, In an implementation, the data in the data group that is to be validated and the data in the comparison data group are severely positively skewed. In this case, a reciprocal transformation can be performed on the data in the data group that is to be validated and the data in the comparison data group. In another implementation, population distribution of the data in the data group that is to be validated and population distribution of the data in the comparison data group are binomial distribution whose population rate is relatively small or whose population rate is relatively large [distribution curve for the ]. In this case, an arcsine square root transformation can be performed on the data in the data group that is to be validated and the data in the comparison data group [the validation includes comparing the input data].
Para 0077, At 902, by a data processing platform, a to-be-validated data group including to-be-validated data corresponding to a predetermined feature is obtained In some implementations, the predetermined feature defines a type of numerical values within a predetermined range. From 902, method 900 proceeds to 904 [expected range]).
Kunkel and Li are considered to be analogous to the claim invention because they are in the same field of processing data. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Kunkel to incorporate the teachings of Li to validate data based on a distribution and range. Doing so to detect abnormal data and continue to use the data (Li Para 0084, Implementation of the present application provide methods and apparatuses for improving abnormal data detection in data processing. In some implementations, a processing platform (e.g., an payment processing server) obtains data that is to be validated and that corresponds to a predetermined feature from a data providing platform as a data group that is to be validated ( e.g., a data group that corresponds to user transaction amounts). In addition, the processing platform can further obtains historical data of the data that is to be validated as a comparison data group. The historical data may also corresponds to the same predetermined feature, and the comparison data group can be provided by the data providing platform in advance. Then, the processing platform performs a two-group significance test on the data group that is to be validated and the comparison data group, and determines whether there is abnormal data based on a test result. If there is no abnormal data, the processing platform can continue to process the data or send the data to a next service step. If the processing platform determines that there is abnormal data, the processing platform can start alerting, instruct related persons to analyze the cause of the data exception, and trigger related solutions.).
Claim(s) 6 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Kunkel in view of Callcut and Stacke and further in view of Li and T. -C. Hsu and C. Lin, "Generative Adversarial Networks for Robust Breast Cancer Prognosis Prediction with Limited Data Size," 2020 42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), Montreal, QC, Canada, 2020, pp. 5669-5672, (“Hsu”).
Regarding claim 6 and analogous claim 13, Kunkel in view of Callcut, Stacke and Li teach the method of claim 5.
Kunkel, Callcut and Stacke are combined with the same rationale as set forth above with respect to claim 1 and analogous 8.
Kunkel and Li are combined with the same rationale as set forth above with respect to claim 5 and analogous 12.
Callcut does not explicitly teach wherein the validation of expected distribution is a curve which fits within two standard deviations of the expected distribution.
However Hsu teaches wherein the validation of expected distribution is a curve which fits within two standard deviations of the expected distribution (Hsu Page 5672,
PNG
media_image6.png
1023
765
media_image6.png
Greyscale
[a curve which fits within two standard deviations of the expected distribution.]
Page 5672 C. Results Para 3, To further analyze how well each model performed in terms of patient stratification, we also plotted KM-plots, as shown in Fig. 2. The curves shown on the figure were the mean of the curves of 30 realizations, and the light/dark shades were the curves one/two standard deviations away from the mean curve, respectively. It showed that both DADA and LR provided insignificant stratification since the two curves had a large potential of overlapping (Fig. 2a and Fig. 2b). On the other hand, wDADA and Bimodal both achieved remarkable stratification (Fig. 2c and Fig. 2d). To quantify the separation between the curves, we also conducted the log-rank test. We counted the number of realizations with p-values greater than 0.005, which represents insignificant stratification (the bad in subfigure captions). We observed that both DADA and LR had approximately 50% of the chance of producing poor stratification. At the same time, wDADA and Bimodal achieved a much lower probability as well as having smaller standard deviations, which indicates the robustness of both models [wherein the validation of expected distribution]).
Kunkel and HSU are considered to be analogous to the claim invention because they are in the same field of machine learning. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Kunkel to incorporate the teachings of Hsu to training a model that produces data within 2 standard deviations of the expected data. Doing so to generate accurate results and have a flexible model that can be incorporated into ensemble learning and simi-supervised learning (Hsu Page 5669 Abstract line 11-23, We found that wDADA achieved 0.6726_0.0278, 0.7538_0.0328, and 0.6507_0.0248 in terms of accuracy, AUC, and concordance index in predicting 5-year DSS, respectively, which is comparable to our previously proposed Bimodal model (accuracy: 0.6889_0.0159; AUC: 0.7546_0.0183; concordance index: 0.6542_0.0120), which needs careful calibration and extensive search on pre-trained network architectures. The flexibility of the proposed wDADA allows us to incorporate it with ensemble learning and semi-supervised learning to further improve performance. Our results indicate that it is possible to utilize generative adversarial networks to train deep models in medical applications, wherein only limited data are available.).
Claim(s) 7 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Kunkel in view of Callcut and Stacke and further in view of Li and Gurumurthy, Swaminathan, Ravi Kiran Sarvadevabhatla, and R. Venkatesh Babu. "Deligan: Generative adversarial networks for diverse and limited data." Proceedings of the IEEE conference on computer vision and pattern recognition. 2017 (“Swaminathan”).
Regarding claim 7 and analogous claim 14, Kunkel in view of Callcut, Stacke and Li teach the method of claim 5.
Kunkel, Callcut and Stacke are combined with the same rationale as set forth above with respect to claim 1 and analogous 8.
Kunkel and Li are combined with the same rationale as set forth above with respect to claim 5 and analogous 12.
Kunkel does not explicitly teach wherein the validation of expected distribution is a curve which fits within a configurable threshold of standard deviations of the expected distribution.
However Swaminathan teaches wherein the validation of expected distribution is a curve which fits within a configurable threshold of standard deviations of the expected distribution (Gurumurthy Page 167 3. Generative Adversarial Networks (GANs) Para 2-3, The generator G is modelled so that it transforms a random vector z into an image
x
G
, i.e.
x
G
= G(z). z typically arises from an easy-to-sample distribution, for e.g. z
~
U(−1, 1) where U denotes a uniform distribution. G is trained to generate images which are indistinguishable from a sampling of the true distribution. In other words, while training G, we try to maximise
p
d
a
t
a
(
x
G
), the probability that the generated samples belong to the data distribution.
PNG
media_image7.png
126
477
media_image7.png
Greyscale
The above equations make explicit the fact that GANs assume a fixed, easy to sample, prior distribution pz(z) and then maximize pdata(xG|z) by training the generator network to produce samples from the data distribution. [wherein the validation of expected distribution is a curve].
Page 169 4.1. Learning μ and σ para 1-2,
For each Gaussian component, we first need to initialize its parameters. For
μ
i
,
≤
i
≤
N
, we sample from a simple prior – in our case, a uniform distribution U(−1, 1). For σi, we assign a small, fixed non-zero initial value (0.2 in our case). Normally, the number of samples we generate from each Gaussian relative to the other Gaussians during training gives us a measure of the ‘weight’ _ for that component. However, _ is not a trainable parameter in our model since we cannot obtain gradients for πis. Therefore, as mentioned before, we consider all components to be equally important.
To generate data, we randomly choose one of the N Gaussian components and sample a latent vector z from the chosen Gaussian (Equation 8). z is passed to G to obtain the output data (image). The generated sample z can now be used to train parameters of D or G using the standard GAN training procedure (Equation 5). In addition, μ and σ are also trained simultaneously along with G’s parameters, using gradients arising from G’s loss function
Page 169 5.1. Modified Inception Score Passing a generated image x = G(z) through a trained classifier with an “inception” architecture [22] results in a conditional label distribution p(y|x). If x is realistic enough, it should result in a “peaky” label distribution i.e. p(y|x) should have low entropy. [is a curve which fits within a configurable threshold of standard deviations of the expected distribution]).
Kunkel and Gurumurthy are considered to be analogous to the claim invention because they are in the same field of machine learning. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to have modified Kunkel to incorporate the teachings of Gurumurthy to training a model based on the sample distribution by decreasing the sample distribution. Doing so to stabilize the model and produce diverse sample even in low data scenarios (Gurumurthy Page 173 7. Conclusions and Future Work line 1-7, In this work, we have shown that reparameterizing the latent space in GANs as a mixture model can lead to a powerful generative model. Via experiments across a diverse set of modalities (digits, hand-drawn object sketches and color photos of objects), we have observed that this seemingly simple modification helps stabilize the model and produce diverse samples even in low data scenarios.).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). 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.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ALFREDO CAMPOS whose telephone number is (571)272-4504. The examiner can normally be reached 7:00 - 4:00 pm M - F.
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, Michael J. Huntley can be reached at (303) 297-4307. 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.
/ALFREDO CAMPOS/Examiner, Art Unit 2129
/MICHAEL J HUNTLEY/Supervisory Patent Examiner, Art Unit 2129