DETAILED ACTION
1. This communication is in request for continued examination filed on February 10, 2026 for Application No. 16/998,694 in which Claims 1-34 are presented for examination.
Notice of Pre-AIA or AIA Status
2. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Continued Examination Under 37 CFR 1.114
3. A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 02/10/2026 has been entered.
Information Disclosure Statement
4. The information disclosure statement submitted on 02/09/2026 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Response to Arguments
5. The amendments filed on February 10, 2026 have been considered. Claims 1-2, 4, 9, 17-18, and 25 have been amended. Claims 33 and 34 have been newly added. Thus, Claims 1-34 are pending and presented for examination.
6. Applicant’s arguments filed February 10, 2026 with respect to the 35 U.S.C. 101 rejection have been fully considered but they are not persuasive.
Applicant’s Arguments on Pgs. 10-11 of Arguments/Remarks state:
“Applicant submits that claim 1 recites features that cannot be performed in the human mind. For example, a human mind cannot "update one or more parameters of one or more first neural networks comprising the first plurality of neural network architectures to identify a first set of one or more accuracy metrics for at least one of the first plurality neural network architectures," "generate a second plurality of neural network architectures comprising at least one of the one or more components based, at least in part, on modifying one or more neural network architectures of the first subset of neural network architectures," "update one or more parameters of one or more second neural networks comprising the second plurality of neural network architectures to identify a second set of one or more accuracy metrics for each of the second plurality neural network architectures," "select a second subset of neural network architectures from a combination of the first plurality of neural network architectures and the second plurality of neural network architectures based, at least in part, on the first set and the second set of one or more accuracy metrics," and "cause one or more inferencing tasks to be performed on input data using the one or more second neural networks comprising the second subset of neural network architectures," as claimed.”
Examiner respectfully disagrees. At Step 2A Prong 1, the limitations "update one or more parameters of one or more first neural networks comprising the first plurality of neural network architectures to identify a first set of one or more accuracy metrics for at least one of the first plurality neural network architectures”, “select a first subset of neural network architectures from the first plurality of neural network architectures based, at least in part, on the first set of one or more accuracy metrics”, “generate a second plurality of neural network architectures based, at least in part, on modifying one or more neural network architectures of the first subset of neural network architectures”, “update one or more parameters of one or more second neural networks comprising the second plurality of neural network architectures to identify a second set of one or more accuracy metrics for each of the second plurality neural network architectures”, and “select a second subset of neural network architectures from a combination of the first plurality of neural network architectures and the second plurality of neural network architectures based, at least in part, on the first set and the second set of one or more accuracy metrics” may all be performed by mental process/manually by a user. For example, a user may update parameters of a first neural network (modifying weights, with the aid of pen and paper) to identify a first set of accuracy metrics (i.e., evaluating improvement/deterioration of performance or accuracy, mathematical formulas used to evaluate performance/accuracy, etc.) Further, the user may then select a first subset of the neural network architectures from the plurality, based on observing/analyzing the identified accuracy metrics and using judgement/evaluation to select networks with improved accuracy. Subsequently, the user may then generate a second plurality of neural network architectures based on modifying one or more of the first subset of neural network architectures, again with the aid of pen and paper. Similarly, the user may again update parameters of the second neural networks (modifying weights, with the aid of pen and paper) to identify a second set of accuracy metrics (i.e., evaluating improvement/deterioration of performance or accuracy, mathematical formulas used to evaluate performance/accuracy, etc.) Finally, the user may select a second subset of neural network architectures from a combination of the first and second plurality based, at least in part on, the identified accuracy metrics – hence, the user may choose to select the networks with maximal/optimal accuracy, as compared to the combination.
Examiner points out that the claims, as currently drafted, do not preclude the limitations from being feasibly performed by mental process. The instant limitations are very broad – for example, one or more broadly cited “parameters” are “updated” without significantly more, simple/generic neural networks are merely “modified” to produce a second plurality of generic networks without significantly more, and generic “accuracy metrics” are identified, without significantly more. Further, Applicant’s arguments merely state that these limitations cannot be performed by mental process, without specifically pointing out the technical limitations of the claim itself which may preclude such an interpretation.
Applicant’s Arguments on Pgs. 11-12 of Arguments/Remarks state:
“Furthermore, under Step 2A Prong 2 of the Alice/Mayo analysis, the Office alleges that claim 1 does not recite a practical application of a judicial exception. The courts have found that a judicial exception is integrated into a practical application when a claim recites "an improvement in the functioning of a computer, or an improvement to other technology or technical field." M.P.E.P. 2106.04(d)(I). Applicant submits that even, assuming arguendo, that claim 1 were to recite a judicial exception, to which Applicant does not concede, the claim recites a practical application of "updating one or more parameters of one or more first neural networks comprising the first plurality of neural network architectures to identify a first set of one or more accuracy metrics for each of the first plurality neural network architectures," "generating a second plurality of neural network architectures comprising at least one of the one or more components based, at least in part, on modifying one or more neural network architectures of the first subset of neural network architectures," "updating one or more parameters of one or more second neural networks comprising the second plurality of neural network architectures to identify a second set of one or more accuracy metrics for at least one of the second plurality neural network architectures," "select a second subset of neural network architectures from a combination of the first plurality of neural network architectures and the second plurality of neural network architectures based, at least in part, on the first set and the second set of one or more accuracy metrics," and "cause one or more inferencing tasks to be performed on input data using the one or more second neural networks comprising the second subset of neural network architectures," which, for example, allows a neural network to be selected by a computer more efficiently with improved accuracy for performing a task. As such, this corresponds to improving the functioning of a computer, as well as improving artificial intelligence technology.”
Examiner respectfully disagrees. At Step 2A Prong 2 and Step 2B, the claim merely recites “one or more processors, comprising circuitry to […]” and “cause one or more inferencing tasks to be performed on input data using the one or more second neural networks comprising the second subset of neural network architectures”. At Step 2A Prong 2, the judicial exception is not integrated into a practical application, as the “one or more processors, comprising circuitry” is recited at a high-level of generality and amounts to no more than mere instructions to apply the exception using generic computer components. Further, the limitation “cause one or more inferencing tasks to be performed on input data using the one or more second neural networks comprising the second subset of neural network architectures” amounts to merely adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f). At Step 2B, the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Applicant further states that the claims “allow a neural network to be selected by a computer more efficiently with improved accuracy for performing a task”, however, the aforementioned limitations are not exclusive to a computer and are broad enough to be performed by mental process/manually by a user, as outlined by Examiner above and detailed further in the subsequent 35 U.S.C. 101 section below. Although Applicant recites that the claims correspond to improving the functioning of a computer, as well as AI technology, this improvement is not reflected in the currently drafted claim language.
Thus, the 35 U.S.C. 101 rejection is maintained.
7. Applicant’s arguments filed February 10, 2026 with respect to the 35 U.S.C. 101 rejection have been fully considered but they are not persuasive.
Applicant’s Arguments on Pg. 13 of Arguments/Remarks state:
“Without acquiescing to the rejection, the claims have been amended to further expedite prosecution. Support for all amended claims can be found in the specification, e.g., paragraphs [0056]-[0063], [0071]-[0074], [0118]-[0129], and [0135]-[0138], and no new matter has been added by these amendments. Applicant respectfully submits that McDonnell and Ganeson fail to teach or suggest claim 1. For example, McDonnell at col. 5, lines 51-58 describes that "[a]fter the relative fitness estimator is trained, the relative fitness estimator can be used (rather than using traditional fitness calculations) to estimate relative fitness values for neural networks in subsequent epochs. Additionally, in some implementations, a subset of the neural networks of one or more of the subsequent epochs can be used to generate updated training data to update or refine the relative fitness estimator." However, McDonnell at least fails to teach claim 1, such as, e.g., "update one or more parameters of one or more first neural networks comprising the first plurality of neural network architectures to identify a first set of one or more accuracy metrics for at least one of the first plurality neural network architectures," and "update one or more parameters of one or more second neural networks comprising the second plurality of neural network architectures to identify a second set of one or more accuracy metrics for at least one of the second plurality neural network architectures," as claimed.
Ganeson fails to remedy the deficiencies of McDonnell. For example, Ganeson describes at col. 9, lines 45-50 that "FIG. 4 illustrates an example of a process 400 for progressively creating a plurality of candidate models using different configuration settings." However, Ganeson at least fails to teach the missing elements of McDonnell.
Therefore, for at least the foregoing reasons, Applicant respectfully submits that claim 1 is allowable over McDonnell and Ganeson. Withdrawal of the pending rejection under 35 U.S.C. § 103 is, therefore, respectfully requested.”
Examiner respectfully disagrees. McDonnell Figure 4 depicts an example of generating a neural network according to the methods presented. More specifically, Figure 4 label 404 describes how one or more parameters (hyperparameters) are updated through the generation of an input matrix, which is used to generate and represent the first neural networks. Subsequently, at label 406, the input matrices (with updated hyperparameters) are input to a relative fitness estimator to generate estimated fitness (accuracy) data for the population – hence, McDonnell clearly teaches the limitation “update one or more parameters of one or more first neural networks comprising a first plurality of neural network architectures to identify a first set of one or more accuracy metrics for at least one of the first plurality of neural network architectures”. Furthermore, Figure 4 label 412 similarly describes how the input matrices representative of each of the neural networks (generated by the modifications shown in label 408) are updated, hence the hyperparameters are also updated and the fitness (accuracy) is again subsequently identified – hence, McDonnell clearly teaches the limitation “update one or more parameters of one or more second neural networks comprising the second plurality of neural network architectures to identify a second set of one or more accuracy metrics for at least one of the second plurality of neural network architectures”. This applies to Independent Claims 1, 9, 17, 25 and their respective dependents 2-8, 10-16, 18-24, and 26-34.
Examiner also notes that it is not clear how the instant claims differ from any standard/generic neural network training, as iterative improvements to a plurality of models and selecting models, from the plurality, with only the highest accuracy when handling inferencing tasks is seemingly known in the art (as proven by the citation of prior art, such as the McDonnell reference, below). Applicant is encouraged to highlight the novelty point of their invention and incorporate such supposed novelty into the instant claim language.
Further, Applicant's arguments fail to comply with 37 CFR 1.111(b) because they amount to a general allegation that the claims define a patentable invention without specifically pointing out how the language of the claims patentably distinguishes them from the references.
Thus, the 35 U.S.C. 103 rejection is maintained.
Claim Rejections - 35 USC § 101
8. 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.
9. Claims 1-34 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding Claim 1:
Step 1: Claim 1 is a system type claim. Therefore, Claims 1-8 and 33-34 are directed to either a process, machine, manufacture, or composition of matter.
2A Prong 1: If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation by mathematical calculation but for the recitation of generic computer components, then it falls within the “Mathematical Concepts” grouping of abstract ideas.
update one or more parameters of one or more first neural networks comprising a first plurality of neural network architectures to identify a first set of one or more accuracy metrics for at least one of the first plurality of neural network architectures (mental process – updating one or more parameters of one or more first neural networks comprising a first plurality of architectures to identify a first set of one or more accuracy metrics for at least one of the first plurality of neural network architectures may be performed manually by a user observing/analyzing the neural networks comprising a first plurality of architectures and accordingly using judgement/evaluation to update one or more parameters of the neural networks, with the aid of pen and paper, to correspondingly identify a first set of accuracy metrics (i.e., evaluating improvement in performance/accuracy after the update as compared to before, mathematical formulas used to evaluate performance/accuracy, etc.) based on said update)
select a first subset of neural network architectures from a first plurality of neural network architectures, based at least in part, on the first set of one or more accuracy metrics (mental process – selecting a first subset of neural network architectures may be performed manually by a user observing/analyzing one or more accuracy metrics and accordingly using judgement/evaluation to select a first subset of neural network architectures based on said analysis)
generate a second plurality of neural network architectures based, at least in part, on modifying one or more neural network architectures of the first subset of neural network architectures (mental process – generating a second plurality of neural network architectures may be performed manually by a user observing/analyzing the first subset of architectures and accordingly using judgement/evaluation to modify the architectures of the first subset to generate a second plurality, with the aid of pen and paper)
update one or more parameters of one or more second neural networks comprising the second plurality of neural network architectures to identify a second set of one or more accuracy metrics for at least one of the second plurality of neural network architectures (mental process – updating one or more parameters of one or more second neural networks comprising the second plurality of architectures to identify a second set of one or more accuracy metrics for at least one of the second plurality of neural network architectures may be performed manually by a user observing/analyzing the neural networks comprising a second plurality of architectures (based on modifying the first set of neural network architectures) and accordingly using judgement/evaluation to update one or more parameters of the neural networks, with the aid of pen and paper, to correspondingly identify a second set of accuracy metrics (i.e., evaluating improvement in performance/accuracy after the update as compared to before, mathematical formulas used to evaluate performance/accuracy, etc.) based on said update)
select a second subset of neural network architectures from a combination of the first plurality of neural network architectures and the second plurality of neural network architectures based, at least in part, on the first set and the second set of one or more accuracy metrics (mental process – selecting a second subset of neural network architectures may be performed manually by a user observing/analyzing the combination of the first plurality and second plurality of architectures and accordingly using judgement/evaluation to select a second subset based on accuracy metrics (i.e., the user may choose to select a second subset with improved accuracy when reviewing the accuracy metrics))
2A Prong 2: This judicial exception is not integrated into a practical application.
Additional elements:
one or more processors, comprising circuitry to […] (recited at a high-level of generality (i.e., as generic processors comprising generic circuitry configured to perform the specific operations of claim 1) such that it amounts to no more than mere instructions to apply the exception using generic computer components)
cause one or more inferencing tasks to be performed on input data using the one or more second neural networks comprising the second subset of neural network architectures (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner's note: high level recitation of applying already trained/configured machine learning models without significantly more)
2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Additional elements:
one or more processors, comprising circuitry to […] (mere instructions to apply the exception using generic computer components cannot provide an inventive concept)
cause one or more inferencing tasks to be performed on input data using the one or more second neural networks comprising the second subset of neural network architectures (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner's note: high level recitation of applying already trained/configured machine learning models without significantly more. This cannot provide an inventive concept)
For the reasons above, Claim 1 is rejected as being directed to an abstract idea without significantly more. This rejection applies equally to dependent claims 2-8 and 33-34. The additional limitations of the dependent claims are addressed below.
Regarding Claim 2:
Step 2A Prong 1: See the rejection of Claim 1 above, which Claim 2 depends on.
the second plurality of neural network architectures is to be generated by adjusting one or more configuration settings for each of the neural network architectures of the first subset of neural network architectures (mental process – generating a second plurality of networks may be performed manually by a user observing/analyzing the first neural networks and using judgement/evaluation to accordingly adjust configuration settings and generate a second plurality of networks, with the aid of pen and paper)
the second subset of neural network architectures is to be selected based, at least in part, on each of the neural network architectures of the second subset having an accuracy at least greater than one or more neural network architectures of the first subset (mental process – selecting the second neural networks based on accuracy may be performed manually by a user observing/analyzing the accuracy of the first/second neural networks and using judgement/evaluation to select second neural networks with an accuracy greater than the accuracy of the one or more first neural networks)
Step 2A Prong 2 & Step 2B:
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 3:
Step 2A Prong 1:
See the rejection of Claim 2 above, which Claim 3 depends on.
Step 2A Prong 2 & Step 2B:
wherein the first subset of neural network architectures is to be trained in parallel using one or more parallel processing units (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying that the neural networks are trained in parallel does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 4:
Step 2A Prong 1: See the rejection of Claim 2 above, which Claim 4 depends on.
[…] select the second subset based, at least in part, on the first set of one or more accuracy metrics from varying neural network architectures of the first subset (mental process – selecting the second subset may be performed manually by a user observing/analyzing the varied neural network architectures of the first subset and accordingly using judgement/evaluation to select the second subset based on analyzing the corresponding accuracy metrics)
Step 2A Prong 2 & Step 2B:
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 5:
Step 2A Prong 1:
See the rejection of Claim 2 above, which Claim 5 depends on.
Step 2A Prong 2 & Step 2B:
wherein the first subset of neural network architectures are further randomly selected from the first plurality of neural network architectures (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying that the neural networks are randomly selected does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 6:
Step 2A Prong 1: See the rejection of Claim 2 above, which Claim 6 depends on.
wherein the one or more configuration settings for each of the neural network architectures of the first subset of neural network architectures are adjusted based, at least in part, on configuration settings associated with the first plurality of neural network architectures (mental process – adjusting configuration settings may be performed manually by a user observing/analyzing the configuration settings associated with each network and accordingly using judgement/evaluation to adjust the according settings of each first neural network, with the aid of pen and paper)
Step 2A Prong 2 & Step 2B:
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 7:
Step 2A Prong 1: See the rejection of Claim 1 above, which Claim 7 depends on.
wherein the first subset is to be selected based, at least in part, on an accuracy associated with a remainder of the neural network architectures in the first plurality being less than the accuracy associated with the first subset (mental process – selecting a first subset may be performed manually by a user observing/analyzing the accuracy associated with the remainder of network architectures and accordingly using judgement/evaluation to select the subset based on the accuracy being less than the accuracy associated with the first subset)
Step 2A Prong 2 & Step 2B:
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 8:
Step 2A Prong 1: See the rejection of Claim 1 above, which Claim 8 depends on.
[…] generate the second plurality based, at least in part, on one or more accuracy metrics resulting from varying neural network architectures of the first plurality (mental process – generating the second plurality may be performed manually by a user observing/analyzing the accuracy metrics and accordingly using judgement/evaluation to generate the second plurality of networks, with the aid of pen and paper, based on said accuracy metrics)
Step 2A Prong 2 & Step 2B:
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Independent Claim 9 recites substantially the same limitations as Claim 1, in the form of a system comprising one or more processors, including generic computer components. The claim is also directed to performing mental processes without significantly more, therefore it is rejected under the same rationale.
For the reasons above, Claim 9 is rejected as being directed to an abstract idea without significantly more. This rejection applies equally to dependent claims 10-16. The additional limitations of the dependent claims are addressed below.
Regarding Claim 10:
Step 2A Prong 1: See the rejection of Claim 9 above, which Claim 10 depends on.
selecting the first subset (mental process – selecting the first subset may be performed manually by a user observing/analyzing the networks and accordingly using judgement/evaluation to select a first subset of networks)
determining the second subset based, at least in part, on accuracy of the second subset having an accuracy at least greater than one or more neural network architectures of the first subset (mental process – determining the second subset may be performed manually by a user observing/analyzing the accuracy of the plurality of networks and accordingly using judgement/evaluation to determine the second subset based on the accuracy being greater than one or more architectures of the first subset)
Step 2A Prong 2 & Step 2B:
the one or more processors further determine the second subset by […] (mere instructions to apply the exception using generic computer components cannot provide an inventive concept)
performing a first training for the first plurality according to one or more first neural network architecture settings (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner's note: high level recitation of training a machine learning model with previously determined data. This cannot provide an inventive concept)
performing a second training for the second plurality according to one or more second neural network architecture settings (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner's note: high level recitation of training a machine learning model with previously determined data. This cannot provide an inventive concept)
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 9.
Regarding Claim 11:
Step 2A Prong 1:
See the rejection of Claim 10 above, which Claim 11 depends on.
Step 2A Prong 2 & Step 2B:
wherein one or more parallel processing units perform the first training and the second training (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying that one or more parallel processing units perform the first/second training does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 9.
Regarding Claim 12:
Step 2A Prong 1:
See the rejection of Claim 10 above, which Claim 12 depends on.
Step 2A Prong 2 & Step 2B:
wherein the one or more first neural network architecture settings comprise one or more data values used to initialize each of the neural network architectures of the first plurality (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying that the one or more first neural network settings comprises one or more data values used to initialize the networks does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 9.
Regarding Claim 13:
Step 2A Prong 1:
See the rejection of Claim 12 above, which Claim 13 depends on.
Step 2A Prong 2 & Step 2B:
wherein the one or more second neural network architecture settings comprise one or more adjusted data values from the one or more first neural network architecture settings (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying that the one or more second neural network settings comprises one or more adjusted data values does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 9.
Regarding Claim 14:
Step 2A Prong 1:
See the rejection of Claim 10 above, which Claim 14 depends on.
Step 2A Prong 2 & Step 2B:
wherein different neural network architectures for each of the first plurality is determined based, at least in part, on an activation key (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying that the different neural network architectures is determined based on a generic activation key does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 9.
Regarding Claim 15:
Step 2A Prong 1:
See the rejection of Claim 14 above, which Claim 15 depends on.
Step 2A Prong 2 & Step 2B:
wherein different neural network architectures for each of the first plurality comprises one or more neural network layers indicated by the activation key (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying that the different neural network architectures comprises one or more neural network layers indicated by a generic activation key does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 9.
Regarding Claim 16:
Step 2A Prong 1:
See the rejection of Claim 14 above, which Claim 16 depends on.
Step 2A Prong 2 & Step 2B:
wherein different neural network architectures for each of the first plurality of neural networks comprises one or more neural network blocks indicated by the activation key (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying that the different neural network architectures comprises one or more neural network blocks indicated by a generic activation key does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 9.
Independent Claim 17 recites substantially the same limitations as Claim 1, in the form of a non-transitory machine-readable medium having stored thereon a set of instructions to be performed by one or more processors, including generic computer components. The claim is also directed to performing mental processes without significantly more, therefore it is rejected under the same rationale.
For the reasons above, Claim 17 is rejected as being directed to an abstract idea without significantly more. This rejection applies equally to dependent claims 18-24. The additional limitations of the dependent claims are addressed below.
Regarding Claim 18:
Step 2A Prong 1: See the rejection of Claim 17 above, which Claim 18 depends on.
[…] determine an accuracy of the neural network architectures of the first plurality (mental process – determining an accuracy of the plurality of neural networks may be performed manually by a user observing/analyzing the first neural network architectures and accordingly using judgement/evaluation to determine an accuracy of said networks)
selecting the first subset from the first plurality of neural network architectures (mental process – selecting one or more of the plurality of neural networks may be performed manually by a user observing/analyzing the plurality of neural network architectures and accordingly using judgement/evaluation to select a first subset of the networks from the plurality)
[…] determine an accuracy of the one or more architectures of the first subset (mental process – determining an accuracy of the first neural networks may be performed manually by a user observing/analyzing the first subset of architectures and accordingly using judgement/evaluation to determine an accuracy of said networks)
Step 2A Prong 2 & Step 2B:
training, based on one or more settings, the first plurality to […] (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner's note: high level recitation of training a machine learning model with previously determined data. This cannot provide an inventive concept)
training, based on one or more adjusted settings, one or more architectures of the first subset to […] (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner's note: high level recitation of training a machine learning model with previously determined data. This cannot provide an inventive concept)
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 17.
Regarding Claim 19:
Step 2A Prong 1:
See the rejection of Claim 18 above, which Claim 19 depends on.
Step 2A Prong 2 & Step 2B:
wherein one or more graphics processing units perform training, as a first parallel operation, of the first plurality and perform training, as a second parallel operation, of the one or more first subset (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner's note: high level recitation of training a machine learning model with previously determined data. This cannot provide an inventive concept)
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 17.
Regarding Claim 20:
Step 2A Prong 1:
See the rejection of Claim 18 above, which Claim 20 depends on.
Step 2A Prong 2 & Step 2B:
wherein the one or more settings are determined based, at least in part, on a visualization (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying that the one or more settings are determined based, at least in part, on a visualization does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 17.
Regarding Claim 21:
Step 2A Prong 1:
See the rejection of Claim 20 above, which Claim 21 depends on.
Step 2A Prong 2 & Step 2B:
wherein the visualization comprises one or more images comprising information about one or more previously selected neural network architectures (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying that the visualization comprises one or more images comprising information about one or more previously selected networks does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 17.
Regarding Claim 22:
Step 2A Prong 1:
See the rejection of Claim 18 above, which Claim 22 depends on.
Step 2A Prong 2 & Step 2B:
wherein the one or more settings comprise data values to initialize each neural network architecture of the first plurality (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying that the one or more settings comprises data values to initialize each neural network does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 17.
Regarding Claim 23:
Step 2A Prong 1:
See the rejection of Claim 22 above, which Claim 23 depends on.
Step 2A Prong 2 & Step 2B:
wherein the one or more adjusted settings comprise data values from the one or more settings modified to change the accuracy of the one or more architectures of the first subset (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying that the one or more settings comprise data values from the one or more settings modified to change accuracy of the one or more networks does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 17.
Regarding Claim 24:
Step 2A Prong 1: See the rejection of Claim 17 above, which Claim 24 depends on.
select the one or more of the neural network architectures based, at least in part, on a time to perform segmentation of one or more medical images (mental process – selecting on or more neural networks based on a time to perform segmentation of one or more medical images may be performed manually by a user observing/analyzing the set of one or more neural network architectures and one or more medical images and accordingly using judgement/evaluation to select the neural networks based on an observed time to perform segmentation)
Step 2A Prong 2 & Step 2B:
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 17.
Independent Claim 25 recites substantially the same limitations as Claim 1, in the form of a method. The claim is also directed to performing mental processes without significantly more, therefore it is rejected under the same rationale.
For the reasons above, Claim 25 is rejected as being directed to an abstract idea without significantly more. This rejection applies equally to dependent claims 26-32. The additional limitations of the dependent claims are addressed below.
Regarding Claim 26:
Step 2A Prong 1: See the rejection of Claim 25 above, which Claim 26 depends on.
modifying one or more settings associated with one or more neural network architectures of the first plurality (mental process – modifying one or more settings may be performed manually by a user observing/analyzing the neural network architectures and accordingly using judgement/evaluation to modify one or more settings associated with the architectures, with the aid of pen and paper)
selecting the first subset based, at least in part, on one or more architectures of the first plurality having an accuracy less than the neural network architectures of the first subset (mental process – selecting one or more second neural networks may be performed manually by a user comparing the accuracy of said networks and selecting the second neural networks with an accuracy less than that of the first neural networks)
Step 2A Prong 2 & Step 2B:
training the one or more neural network architectures of the first plurality (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner' s note: high level recitation of training a machine learning model with previously determined data. This cannot provide an inventive concept)
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 25.
Regarding Claim 27:
Step 2A Prong 1:
See the rejection of Claim 26 above, which Claim 27 depends on.
Step 2A Prong 2 & Step 2B:
wherein the one or more neural network architectures of the first plurality are trained in parallel using one or more parallel processing units (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying that the one or more first neural networks are trained in parallel does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 25.
Regarding Claim 28:
Step 2A Prong 1:
See the rejection of Claim 26 above, which Claim 28 depends on.
Step 2A Prong 2 & Step 2B:
wherein the one or more settings comprise one or more data values usable to initialize one or more components in each neural network architectures of the first plurality (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying that the one or more settings comprise one or more data values usable to initialize one or more components in each neural network does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 25.
Regarding Claim 29:
Step 2A Prong 1:
See the rejection of Claim 26 above, which Claim 29 depends on.
Step 2A Prong 2 & Step 2B:
wherein each of the one or more neural network architectures of the first plurality comprises an architecture determined based, at least in part, on an activation key (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying that each neural network comprises an architecture based, at least in part, on a generic activation key does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 25.
Regarding Claim 30:
Step 2A Prong 1:
See the rejection of Claim 29 above, which Claim 30 depends on.
Step 2A Prong 2 & Step 2B:
wherein the activation key comprises one or more numerical values to indicate a number of layers for each neural network architecture in the first plurality (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying that the activation key comprises one or more numerical values to indicate a number of layers for each network does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 25.
Regarding Claim 31:
Step 2A Prong 1:
See the rejection of Claim 29 above, which Claim 31 depends on.
Step 2A Prong 2 & Step 2B:
wherein the activation key comprises one or more numerical values to indicate one or more neural network blocks to be used in one or more layers for each neural network architecture in the first plurality (Field of Use – limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception does not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application; in this case specifying that the activation key comprises one or more numerical values to indicate one or more neural network blocks to be used in one or more layers for each network does not integrate the exception into a practical application nor amount to significantly more – See MPEP 2106.05(h))
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 25.
Regarding Claim 32:
Step 2A Prong 1: See the rejection of Claim 26 above, which Claim 32 depends on.
[…] performs segmentation of medical images (mental process – performing segmentation of medical images may be performed manually by a user observing/analyzing the medical images and accordingly using judgement/evaluation to segment said images)
Step 2A Prong 2 & Step 2B:
wherein one or more architectures of the second subset performs segmentation of medical images with maximal accuracy (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner' s note: high level recitation of applying a trained machine learning model without significantly more. This cannot provide an inventive concept)
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 25.
Regarding Claim 33:
Step 2A Prong 1: See the rejection of Claim 1 above, which Claim 33 depends on.
update the one or more parameters of the one or more first neural networks […] (mental process – updating one or more parameters of the one or more first neural networks may be performed manually by a user observing/analyzing the one or more first neural networks and accordingly using judgement/evaluation to update the one or more parameters of the one or more first neural networks, with the aid of pen and paper)
Step 2A Prong 2 & Step 2B:
[…] training the one or more first neural networks to perform a task and identify the first set of one or more accuracy metrics associated with performing the task (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner' s note: high level recitation of training a machine learning model with previously determined data without significantly more. This cannot provide an inventive concept)
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Regarding Claim 34:
Step 2A Prong 1:
See the rejection of Claim 1 above, which Claim 34 depends on.
Step 2A Prong 2 & Step 2B:
wherein the one or more parameters of the one or more first neural networks comprise one or more neural network weights that are to be modified by training the one or more first neural networks (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner's note: high level recitation of training a machine learning model with previously determined data without significantly more. This cannot provide an inventive concept)
Accordingly, under Step 2A Prong 2 and Step 2B, these additional elements do not integrate the abstract idea into practical application because they do not impose any meaningful limits on practicing the abstract idea, as discussed above in the rejection of claim 1.
Claim Rejections - 35 USC § 103
10. 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.
11. Claims 1-23, 25-31, and 33-34 are rejected under 35 U.S.C. 103 as being unpatentable over McDonnell et al. (hereinafter McDonnell) (US Patent 10685286), in view of Ganesan et al. (hereinafter Ganesan) (US Patent 11494587).
Regarding Claim 1, McDonnell teaches one or more processors (McDonnell, Figure 6, label 602, which depicts one or more processors), comprising: circuitry (McDonnell, Figure 6, label 604, which depicts one or more logic circuits within the one or more processors) to:
update one or more parameters of one or more first neural networks comprising a first plurality of neural network architectures (McDonnell, Col. 20 lines 49-59, “In the example illustrated in FIG. 4, the operations performed iteratively include, at 404, generating a matrix representation for each neural network of a population of neural networks. For example, the matrix mapper 128 can generate the matrix representations 304 of FIG. 3 based on the neural networks of the population 302 and based on the grammar(s) 116. The matrix representation of a particular neural network includes two or more rows of values. Each row corresponds to a set of layers (e.g., one or more layers) of the particular neural network, and each value specifies a hyperparameter of the set of layers”, therefore, one or more parameters (including matrix representations specifying hyperparameters) of one or more first neural networks comprising a first plurality of architectures (based on different grammars which specify different combinations of features, hence different architectures – see McDonnell supporting citation Col. 4 lines 34-42) are updated. See Figure 4 label 404) to identify a first set of one or more accuracy metrics for at least one of the first plurality of neural network architectures (McDonnell, Col. 20 lines 65-67 & Col. 21 lines 1-6, “In the example illustrated in FIG. 4, the operations performed iteratively also include, at 406, providing the matrix representations as input to a relative fitness estimator to generate estimated fitness data for neural networks of the population. For example, the matrix representations 304 of FIG. 3 can be provided as input to the relative fitness estimator 134. The estimated fitness data are based on expected fitness of neural networks predicted by the relative fitness estimator.”, thus, a first set of one or more accuracy metrics (fitness data – see supporting McDonnell Col. 20 lines 2-11) are identified for the at least one of the first plurality of neural network architectures. See Figure 4 label 406);
select a first subset of neural network architectures from the first plurality of neural network architectures based, at least in part, on the first set of one or more accuracy metrics (McDonnell, Col. 10 lines 41-46, “To illustrate, in some implementations, the model building engine 126 uses the estimated fitness data to select neural networks to undergo mutation, cross-over, or both. The model building engine 126 can also, or in the alternative, use the estimated fitness data to select neural networks for extinction.”, therefore, a subset of neural networks may be selected based, at least in part, on the first set of accuracy metrics (fitness data generated));
generate a second plurality of neural network architectures (McDonnell, Col. 21 lines 7-17, “In the example illustrated in FIG. 4, the operations performed iteratively further include, at 408, generating, based on the population of neural networks and based on the estimated fitness data, a subsequent population of neural networks. For example, the evolutionary processor 316 of the model building engine 126 can perform various evolutionary operations using members of the population 302 to generate the population 320. In this example, the evolutionary operations performed and the neural networks used for the evolutionary operations are based on the estimated fitness data 310 and the parameters 118.”, thus, a second plurality of neural network architectures is generated based on modifying the first set of neural networks through evolutionary operations. See Figure 4 label 408) based, at least in part, on modifying one or more neural network architectures of the first subset of neural network architectures (McDonnell indeed teaches generating a second plurality of neural network architectures based, at least in part, on modifying one or more neural network architectures of the first subset, as cited above. However, McDonnell’s recitation of evolutionary operations, such as the use of crossover and mutation operations, is brief and not further expanded upon. Thus, Examiner introduces the Ganesan reference below to teach the specifics of such modifications);
update one or more parameters of one or more second neural networks comprising the second plurality of neural network architectures to identify a second set of one or more accuracy metrics for at least one of the second plurality of neural network architectures (McDonnell, Col. 21 lines 38-47, “In the example illustrated in FIG. 4, the operations performed iteratively include, at 412, generating updated training data indicating the fitness values and, for each fitness value, a matrix representation of a neural network associated with the fitness value. For example, the training data generator 210 of FIG. 2 can combine matrix representations 206 and the calculated fitness values 208 for the one or more sample neural networks from the subsequent population (e.g., the population 320 of FIG. 3) to generate updated training data 212.”, therefore, one or more parameters (including matrix representations specifying hyperparameters) of one or more second neural networks comprising the second plurality of neural network architectures is updated, in order to identify a second set of one or more accuracy metrics (fitness data – see supporting McDonnell Col. 20 lines 2-11) for the at least one of the second plurality of neural network architectures. See Figure 4 label 410 & 412);
select a second subset of neural network architectures from a combination of the first plurality of neural network architectures and the second plurality of neural network architectures based, at least in part, on the first set and the second set of one or more accuracy metrics (McDonnell, Col. 19 lines 45-57, “As another example, the neural network selector 324 can prompt the relative fitness estimator 134 to generate estimated fitness data 310 for the final population. In this example, the neural network selector 324 can use the estimated fitness data 310 to select the neural network 136. In some implementations, the neural network selector 324 can use the estimated fitness data 310 to select a subset of the final population for which fitness values 208 are to be calculated. For example, the ten (10) highest ranked neural networks of the final population based on the estimated fitness data 310 can be provided to the fitness calculator 130 to calculate fitness values 208, which are used to select the neural network 136.”, therefore, a second subset of neural network architectures may be selected, by the network selector, from the final population (comprising the first/second plurality of neural networks) based on the fitness values/accuracy metrics. See supporting Figure 4 label 416); and
cause one or more inferencing tasks to be performed on input data using the one or more second neural networks comprising the second subset of neural network architectures (McDonnell, Col. 6 lines 44-54, “For example, if the data set used to generate the neural network is time series data captured by a particular sensor coupled to a particular system, the neural network can be used to predict future sensor data values or future states of the particular system based on real-time sensor data from the particular sensor (or from similar sensors associated with other systems). Thus, the neuroevolutionary process automates creation of a software tool (e.g., a neural network) that can be used for a variety of purposes depending on the input data set used.”, thus, one or more inferencing tasks (such as predicting future sensor data values or future states of a system) may be performed on input data (real-time sensor data) using the one or more second neural networks comprising the second subset of neural network architectures (which are generated through the neuroevolutionary process, as outlined in the rejection of the preceding limitations)).
While McDonnell teaches generating a second plurality of neural networks through the use of various evolutionary operations, such as crossover and mutation, McDonnell only briefly discloses this concept and does not further refine the modifications performed. Thus, Ganesan is introduced below for the purpose of teaching the specifics of such modifications.
Ganesan teaches generating a second plurality of neural network architectures based, at least in part, on modifying one or more neural network architectures of the first subset of neural network architectures (Ganesan, Figure 4, which depicts a process for progressively creating a plurality of candidate models using different configuration settings, those of which are illustrated by Figure 5. Ganesan Figure 4 label 404 shows that a candidate model is created using initial settings and then based on accuracy and adjustments, a new candidate model is created using the adjusted target settings in label 412. Thus, Ganesan clearly teaches generating a second plurality of neural network architectures based on modifying one or more neural network architectures of a first subset).
It would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the one or more processors of claim 1, as disclosed by McDonnell to include generating a second plurality of neural network architectures based, at least in part, on modifying one or more neural network architectures of the first subset of neural network architectures, as disclosed by Ganesan. One of ordinary skill in the art would have been motivated to make this modification to optimize the plurality of neural networks generated, hence improving accuracy and effectiveness over a variety of heterogeneous frameworks and data sources (Ganesan, Col. 3 lines 61-67 & Col. 4 lines 1-3, “Furthermore, in various embodiments. ML models can be generated via heterogeneous ML frameworks and selected and adapted to particular data sources. In this way, data processing of incoming data streams can be demonstrably improved in accuracy and effectiveness.”).
Regarding Claim 2, McDonnell in view of Ganesan teaches the one or more processors of claim 1, wherein:
the second plurality of neural network architectures is to be generated by adjusting one or more configuration settings for each of the neural network architectures of the first subset of neural network architectures (Ganesan, Figure 4, which depicts a process for progressively creating a plurality of candidate models using different configuration settings, those of which are illustrated by Figure 5. Ganesan Figure 4 label 404 shows that a candidate model is created using initial settings and then based on accuracy and adjustments, a new candidate model is created using the adjusted target settings in label 412. Thus, Ganesan clearly teaches generating a second plurality of neural network architectures based on modifying one or more neural network architectures of a first subset. As mentioned above, McDonnell also broadly teaches generating a second plurality of neural network architectures by applying neuroevolutionary operations to the first subset, as supported by McDonnell Col. 5 lines 66-67 & Col. 6 lines 1-12); and
the second subset of neural network architectures is to be selected based, at least in part, on each of the neural network architectures of the second subset having an accuracy at least greater than one or more neural network architectures of the first subset (McDonnell, Col. 19 lines 6-19, “The termination condition can be satisfied when a number of iterations performed is equal to an iteration count threshold, when a fitness value 208 of a particular neural network is greater than or equal to a fitness threshold, when a representative fitness value (e.g., an average of one or more fitness values 208) is greater than or equal to a fitness threshold estimate, when an estimated relative fitness value 312 of a particular neural network is greater than or equal to a relative fitness threshold, when a representative estimated relative fitness value (e.g., an average of one or more estimated relative fitness values 312) is greater than or equal to a relative fitness threshold, or when another condition indicates that further iterations are not warranted.”, thus, the second subset of neural networks may be selected based on the fitness value/accuracy being greater than the threshold associated with one or more neural network architectures of a first subset).
The reasons of obviousness have been noted in the rejection of Claim 1 above and applicable herein.
Regarding Claim 3, McDonnell in view of Ganesan teaches the one or more processors of claim 2, wherein the first subset of neural network architectures is to be trained in parallel using one or more parallel processing units (McDonnell, Col. 23 lines 51-55, “The computer system 600 includes one or more processors 602. Each processor of the one or more processors 602 can include a single processing core or multiple processing cores that operate sequentially, in parallel, or sequentially at times and in parallel at other times.”, thus, one or more parallel processing units comprise the computer system. Furthermore, as mentioned in Col. 18 lines 45-55, the trainer (Figure 3, label 318) may be used to train a subset of the population of first neural networks).
Regarding Claim 4, McDonnell in view of Ganesan teaches the one or more processors of claim 2, wherein the circuitry is to select the second subset based, at least in part, on the first set of one or more accuracy metrics resulting from varying neural network architectures of the first subset (McDonnell, Col. 19 lines 25-32, “In a particular implementation, the neural network selector 324 selects the neural network based on a section value, such as the ranking values 314 associated with neural networks of the final population, the estimated relative fitness values 312 associated with the neural networks of the final population, or fitness values 208 associated with the neural networks of the final population.”, therefore, the second subset of neural networks may be selected based, at least in part, on the first set of one or more accuracy metrics (fitness data) resulting from varying neural network architectures of the first subset).
Regarding Claim 5, McDonnell in view of Ganesan teaches the one or more processors of claim 2, wherein the first subset of neural network architectures are further randomly selected from the first plurality of neural network architectures (McDonnell, Col. 18, lines 32-37, “For example, the evolutionary processor 316 can use a weighted randomization process to select particular neural networks for mutation and crossover operations. In this example, the likelihood of each neural network of the population 302 being randomly selected is weighted according to the estimated fitness data 310.”, therefore, one or more first neural networks may be randomly selected from the set of neural networks/population).
Regarding Claim 6, McDonnell in view of Ganesan teaches the one or more processors of claim 2, wherein the one or more configuration settings for each of the neural network architectures of the first subset of neural network architectures are adjusted based, at least in part, on configuration settings associated with the first plurality of neural network architectures (Ganesan, Figure 4, which depicts a process for progressively creating a plurality of candidate models using different configuration settings, those of which are illustrated by Figure 5. Ganesan Figure 4 label 404 shows that a candidate model is created using initial settings and then based on accuracy and adjustments, a new candidate model is created using the adjusted target settings in label 412. Thus, Ganesan clearly teaches generating a second plurality of neural network architectures based on modifying one or more neural network architectures of a first subset. As mentioned above, McDonnell also broadly teaches generating a second plurality of neural network architectures by applying neuroevolutionary operations to the first subset, as supported by McDonnell Col. 5 lines 66-67 & Col. 6 lines 1-12).
The reasons of obviousness have been noted in the rejection of Claim 1 above and applicable herein.
Regarding Claim 7, McDonnell in view of Ganesan teaches the one or more processors of claim 1, wherein the first subset is to be selected based, at least in part, on an accuracy associated with a remainder of the neural network architectures in the first plurality being less than the accuracy associated with the first subset (McDonnell, Col. 10 lines 47-54, “During each epoch of the neuroevolutionary process, it is desirable for the population (or at least some members of the population) to be improved (e.g., to become more fit). This goal is generally achieved by retaining, mutating, and/or breeding fitter neural networks, and by avoiding breeding of or discarding less fit neural networks. Calculated fitness values from the fitness calculator 130 can be used to rank the neural networks of a population in order to achieve this goal.”, therefore, the first subset of networks may be selected based, at least in part, on an accuracy (fitness) associated with a remainder of the neural network architectures in the first plurality being less than the accuracy associated with the first subset, as less fit networks are not evolved and/or discarded).
Regarding Claim 8, McDonnell in view of Ganesan teaches the one or more processors of claim 1, wherein the circuitry is to generate the second plurality based, at least in part, on one or more accuracy metrics resulting from varying neural network architectures of the first plurality (McDonnell, Col. 21 lines 7-17, “In the example illustrated in FIG. 4, the operations performed iteratively further include, at 408, generating, based on the population of neural networks and based on the estimated fitness data, a subsequent population of neural networks. For example, the evolutionary processor 316 of the model building engine 126 can perform various evolutionary operations using members of the population 302 to generate the population 320. In this example, the evolutionary operations performed and the neural networks used for the evolutionary operations are based on the estimated fitness data 310 and the parameters 118.”, thus, the second plurality may be generated based, at least in part, on one or more accuracy metrics resulting from varying neural network architectures of the first plurality (through evolutionary operations). See Figure 4 label 408))
Regarding Claim 9, McDonnell in view of Ganesan teaches a system comprising: one or more processors (McDonnell, Figure 6, label 602, which depicts one or more processors within a system (label 600)) to: […]
The rest of the claim language in Claim 9 recites substantially the same limitations as Claim 1, in the form of a system, therefore it is rejected under the same rationale.
The reasons of obviousness have been noted in the rejection of Claim 1 above and applicable herein.
Regarding Claim 10, McDonnell in view of Ganesan teaches the system of claim 9, wherein the one or more processors further determine the second subset by:
performing a first training for the first plurality according to one or more first neural network architecture settings (McDonnell, Figure 3, which depicts the model building engine (label 126) comprising a trainer (label 318) to train a set of neural networks according to one or more settings and parameters (label 118));
selecting the first subset (McDonnell, Figure 3, which depicts the Neural Network Selector (label 324) that selects one or more first neural networks (label 136) from the set/population);
performing a second training for the second plurality according to one or more second neural network architecture settings (McDonnell, Figure 3, which depicts that once the termination check has been performed, a new population may be performed in a new epoch consisting of the same one or more first neural networks but parameters are adjusted and the process (label 300) is repeated such that a second training may be performed by the trainer (label 318)); and
determining the second subset based, at least in part, on accuracy of the second subset having an accuracy at least greater than one or more neural network architectures of the first subset (McDonnell, Col. 19 lines 6-19, “The termination condition can be satisfied when a number of iterations performed is equal to an iteration count threshold, when a fitness value 208 of a particular neural network is greater than or equal to a fitness threshold, when a representative fitness value (e.g., an average of one or more fitness values 208) is greater than or equal to a fitness threshold estimate, when an estimated relative fitness value 312 of a particular neural network is greater than or equal to a relative fitness threshold, when a representative estimated relative fitness value (e.g., an average of one or more estimated relative fitness values 312) is greater than or equal to a relative fitness threshold, or when another condition indicates that further iterations are not warranted.”, thus, the second subset of neural networks may be selected based on the fitness value/accuracy being greater than the threshold associated with one or more neural network architectures of a first subset).
Regarding Claim 11, McDonnell in view of Ganesan teaches the system of claim 10, wherein one or more parallel processing units perform the first training and the second training (McDonnell, Col. 23 lines 51-55, “The computer system 600 includes one or more processors 602. Each processor of the one or more processors 602 can include a single processing core or multiple processing cores that operate sequentially, in parallel, or sequentially at times and in parallel at other times.”, thus, one or more parallel processing units comprise the computer system. Furthermore, as mentioned in Col. 18 lines 45-55, the trainer (Figure 3, label 318) may be used to perform first and second training).
Regarding Claim 12, McDonnell in view of Ganesan teaches the system of claim 10, wherein the one or more first neural network architecture settings comprise one or more data values used to initialize each of the neural network architectures of the first plurality (Ganesan, Figure 5, which depicts one or more configuration settings which are used to initialize each neural network architecture. As mentioned above, McDonnell also broadly teaches generating a second plurality of neural network architectures by applying neuroevolutionary operations to the first subset, as supported by McDonnell Col. 5 lines 66-67 & Col. 6 lines 1-12).
The reasons of obviousness have been noted in the rejection of Claim 9 above and applicable herein.
Regarding Claim 13, McDonnell in view of Ganesan teaches the system of claim 12, wherein the one or more second neural network architecture settings comprise one or more adjusted data values from the one or more first neural network architecture settings (Ganesan, Figure 4, which depicts a process for progressively creating a plurality of candidate models using different configuration settings, those of which are illustrated by Figure 5. Ganesan Figure 4 label 404 shows that a candidate model is created using initial settings and then based on accuracy and adjustments, a new candidate model is created using the adjusted target settings in label 412. Thus, Ganesan clearly teaches generating a second plurality of neural network architectures based on modifying one or more neural network architectures of a first subset. As mentioned above, McDonnell also broadly teaches generating a second plurality of neural network architectures by applying neuroevolutionary operations to the first subset, as supported by McDonnell Col. 5 lines 66-67 & Col. 6 lines 1-12).
The reasons of obviousness have been noted in the rejection of Claim 9 above and applicable herein.
Regarding Claim 14, McDonnell in view of Ganesan teaches the system of claim 10, wherein different neural network architectures for each of the first plurality is determined based, at least in part, on an activation key (McDonnell, Col. 8 lines 27-38, “For example, the architecture of each neural network can be randomly or pseudo-randomly selected from among a set of allowable architectures. One way to randomize generation of the initial population of neural networks is to assign a random value to a hyperparameter (or a set of hyperparameters), where the hyperparameters indicate, for example, a number of hidden layers of the neural network, an architecture of a layer or set of layers, a number of nodes in a layer or set of layers, an activation function of a node or layer, etc. In such implementations, the random values assigned to the hyperparameters specify the architecture of the neural network.”, therefore, referencing applicant’s specification Par. [0070] an activation key may be a vector, set, group, array, or any data structure that indicates one or more items/components which are included or activated in specific neural network architectures. The cited portion of McDonnell above similarly specifies a hyperparameter which may indicate the architecture of a neural network including various components such as number of hidden layers, number of nodes, activation function, etc. Hence, the different neural network architectures is determined based, at least in part, on an activation key).
Regarding Claim 15, McDonnell in view of Ganesan teaches the system of claim 14, wherein different neural network architectures for each of the first plurality comprises one or more neural network layers indicated by the activation key (McDonnell, Col. 8 lines 29-38, “One way to randomize generation of the initial population of neural networks is to assign a random value to a hyperparameter (or a set of hyperparameters), where the hyperparameters indicate, for example, a number of hidden layers of the neural network, an architecture of a layer or set of layers, a number of nodes in a layer or set of layers, an activation function of a node or layer, etc. In such implementations, the random values assigned to the hyperparameters specify the architecture of the neural network.”, thus, the neural network architecture may comprise one or more neural network layers indicated by the activation key/hyperparameter, as explained above by the rejection of Claim 14).
Regarding Claim 16, McDonnell in view of Ganesan teaches the system of claim 14, wherein different neural network architectures for each of the first plurality comprises one or more neural network blocks indicated by the activation key (McDonnell, Col. 8 lines 29-38, “One way to randomize generation of the initial population of neural networks is to assign a random value to a hyperparameter (or a set of hyperparameters), where the hyperparameters indicate, for example, a number of hidden layers of the neural network, an architecture of a layer or set of layers, a number of nodes in a layer or set of layers, an activation function of a node or layer, etc. In such implementations, the random values assigned to the hyperparameters specify the architecture of the neural network.”, thus, the neural network architecture may comprise one or more neural network blocks indicated by the activation key/hyperparameter. Applicant’s specification Par. [0081] defines a “neural network block” as any module and/or block comprising at least convolutional blocks/layers, or any other type of neural network block capable of use in performing neural network operations related to neural network functions).
Regarding Claim 17, McDonnell in view of Ganesan teaches a non-transitory machine-readable medium having stored thereon a set of instructions, when performed by one or more processors (McDonnell, Col. 2 lines 15-18, “In another particular aspect, a computer-readable storage device stores instructions that, when executed by a processor, cause the processor to iteratively perform a set of operations until a termination condition is satisfied.”, therefore, a machine-readable medium having stored thereon a set of instructions to be performed by processors is disclosed), cause the one or more processors to at least: […]
The rest of the claim language in Claim 17 recites substantially the same limitations as Claim 1, in the form of a non-transitory machine-readable medium, therefore it is rejected under the same rationale.
The reasons of obviousness have been noted in the rejection of Claim 1 above and applicable herein.
Regarding Claim 18, McDonnell in view of Ganesan teaches the non-transitory machine-readable medium of claim 17, wherein the set of instructions when performed by one or more processors, further cause the one or more processors to select the one or more of the neural network architectures by:
training, based on one or more settings, the first plurality to determine an accuracy of the neural network architectures of the first plurality (McDonnell, Figure 3, which depicts the model building engine (label 126) comprising a trainer (label 318) to train a set of neural networks according to one or more settings and parameters (label 118). Further, See Ganesan Figure 5 for teaching of specific configuration settings);
selecting the first subset from the first plurality of neural network architectures (McDonnell, Figure 3, which depicts the Neural Network Selector (label 324) that selects one or more first neural networks (label 136) from the set/population); and
training, based on one or more adjusted settings, one or more architectures of the first subset to determine an accuracy of the one or more architectures of the first subset (McDonnell, Figure 3, which depicts that once the termination check has been performed, a new population may be processed in a new epoch consisting of the same one or more first neural networks but parameters are adjusted and the process (label 300) is repeated such that training may be performed by the trainer (label 318)).
Regarding Claim 19, McDonnell in view of Ganesan teaches the non-transitory machine-readable medium of claim 18, wherein one or more graphics processing units perform training, as a first parallel operation, of the first plurality and perform training, as a second parallel operation, of the one or more first subset (McDonnell, Col. 23 lines 51-55, “The computer system 600 includes one or more processors 602. Each processor of the one or more processors 602 can include a single processing core or multiple processing cores that operate sequentially, in parallel, or sequentially at times and in parallel at other times.”, thus, one or more parallel processing units (which are analogous to the graphics processing unit, as supported by Applicant’s specification Par. [0102]) comprise the computer system. Furthermore, as mentioned in Col. 18 lines 45-55, the trainer (Figure 3, label 318) may be used to train neural networks in parallel).
Regarding Claim 20, McDonnell in view of Ganesan teaches the non-transitory machine-readable medium of claim 18, wherein the one or more settings are determined based, at least in part, on a visualization (McDonnell, Col. 7 lines 64-67 – Col. 8 lines 1-7, “The data set analyzer 124 uses heuristics, a data classifier, or both, to determine characteristics of the input data that indicate a data type of the input data. For example, the data set 110 could include time-series data, text, image data, other data types, or combinations thereof (e.g., time-series data with associated text labels). In some implementations, the data set analyzer 124 also, or in the alternative, uses heuristics, the data classifier, or both, to identify a problem to be solved based on the input data, such as whether the neural network 136 is to be configured to classify input data or to predict a future state or value.”, thus, the data set analyzer may utilize image data/visualization to determine settings/configurations that are provided to the model building engine to configure the neural networks – further exemplified in Col. 8 lines 11-23.).
Regarding Claim 21, McDonnell in view of Ganesan teaches the non-transitory machine-readable medium of claim 20, wherein the visualization comprises one or more images comprising information about one or more previously selected neural network architectures (McDonnell, Col. 8 lines 39- 46, “In some implementations, information provided by the data set analyzer 124 is used to weight the randomization process used by the model building engine 126. For example, when the data set analyzer 124 indicates that the data set 110 has particular characteristics, the model building engine 126 may favor one type of neural network architecture over another type of neural network architecture.”, thus, the image data/visualization may comprise information about one or more previously selected neural networks, as the data set analyzer may forward this information to the model builder to favor a certain type of architecture from a previously selected network. Further, Col. 11 also mentions that the data set may comprise historical data).
Regarding Claim 22, McDonnell in view of Ganesan teaches the non-transitory machine-readable medium of claim 18, wherein the one or more settings comprise data values to initialize each neural network architecture of the first plurality (Ganesan, Figure 5, which depicts one or more configuration settings which are used to initialize each neural network architecture. As mentioned above, McDonnell also broadly teaches generating a second plurality of neural network architectures by applying neuroevolutionary operations to the first subset, as supported by McDonnell Col. 5 lines 66-67 & Col. 6 lines 1-12).
The reasons of obviousness have been noted in the rejection of Claim 17 above and applicable herein.
Regarding Claim 23, McDonnell in view of Ganesan teaches the non-transitory machine-readable medium of claim 22, wherein the one or more adjusted settings comprise data values from the one or more settings modified to change the accuracy of the one or more architectures of the first subset (Ganesan, Figure 4, which depicts how the settings of the initial model may be adjusted/modified, in order to change accuracy of the one or more architectures. In Figure 4, first a candidate model is created using initial settings (label 404) then accuracy is measured (label 406), depending on the accuracy evaluation, settings are adjusted or not adjusted (label 408) based on said evaluation of accuracy).
The reasons of obviousness have been noted in the rejection of Claim 17 above and applicable herein.
Regarding Claim 25, McDonnell in view of Ganesan teaches a method (McDonnell, Claim 1, “A method of generating a neural network based on a data set, the method comprising: […]”, thus, a method is disclosed) comprising: […]
The rest of the claim language in Claim 25 recites substantially the same limitations as Claim 1, in the form of a method, therefore it is rejected under the same rationale.
The reasons of obviousness have been noted in the rejection of Claim 1 above and applicable herein.
Regarding Claim 26, McDonnell in view of Ganesan teaches the method of claim 25, further comprising:
modifying one or more settings associated with one or more neural network architectures of the first plurality (McDonnell, Col. 18 lines 52-55, “For example, at least the new neural networks can undergo some training (e.g., backpropagation training) via the trainer 318 to adjust link weights of the new neural networks.”, thus, the new neural network settings, such as link weights, may be adjusted by the trainer.);
training the one or more neural network architectures of the first plurality (McDonnell, Figure 3, which depicts the model building engine (label 126) comprising a trainer (label 318) to train a set of neural networks according to one or more settings and parameters (label 118)); and
selecting the first subset based, at least in part, on one or more architectures of the first plurality having an accuracy less than the neural network architectures of the first subset (McDonnell, Col. 10 lines 47-54, “During each epoch of the neuroevolutionary process, it is desirable for the population (or at least some members of the population) to be improved (e.g., to become more fit). This goal is generally achieved by retaining, mutating, and/or breeding fitter neural networks, and by avoiding breeding of or discarding less fit neural networks. Calculated fitness values from the fitness calculator 130 can be used to rank the neural networks of a population in order to achieve this goal.”, therefore, the first subset of networks may be selected based, at least in part, on an accuracy (fitness) associated with a remainder of the neural network architectures in the first plurality being less than the accuracy associated with the first subset, as less fit networks are not evolved and/or discarded).
Regarding Claim 27, McDonnell in view of Ganesan teaches the method of claim 26, wherein the one or more neural network architectures of the first plurality are trained in parallel using one or more parallel processing units (McDonnell, Col. 23 lines 51-55, “The computer system 600 includes one or more processors 602. Each processor of the one or more processors 602 can include a single processing core or multiple processing cores that operate sequentially, in parallel, or sequentially at times and in parallel at other times.”, thus, one or more parallel processing units comprise the computer system. Furthermore, as mentioned in Col. 18 lines 45-55, the trainer (Figure 3, label 318) may be used to train first neural networks in parallel).
Regarding Claim 28, McDonnell in view of Ganesan teaches the method of claim 26, wherein the one or more settings comprise one or more data values usable to initialize one or more components in each neural network architectures of the first plurality (Ganesan, Figure 5, which depicts one or more configuration settings which are used to initialize each neural network architecture. As mentioned above, McDonnell also broadly teaches generating a second plurality of neural network architectures by applying neuroevolutionary operations to the first subset, as supported by McDonnell Col. 5 lines 66-67 & Col. 6 lines 1-12).
The reasons of obviousness have been noted in the rejection of Claim 25 above and applicable herein.
Regarding Claim 29, McDonnell in view of Ganesan teaches the method of claim 26, wherein each of the one or more neural network architectures of the first plurality comprises an architecture determined based, at least in part, on an activation key (McDonnell, Col. 8 lines 27-38, “For example, the architecture of each neural network can be randomly or pseudo-randomly selected from among a set of allowable architectures. One way to randomize generation of the initial population of neural networks is to assign a random value to a hyperparameter (or a set of hyperparameters), where the hyperparameters indicate, for example, a number of hidden layers of the neural network, an architecture of a layer or set of layers, a number of nodes in a layer or set of layers, an activation function of a node or layer, etc. In such implementations, the random values assigned to the hyperparameters specify the architecture of the neural network.”, therefore, referencing applicant’s specification Par. [0070] an activation key may be a vector, set, group, array, or any data structure that indicates one or more items/components which are included or activated in specific neural network architectures. The cited portion of McDonnell above similarly specifies a hyperparameter which may indicate the architecture of a neural network including various components such as number of hidden layers, number of nodes, activation function, etc.).
Regarding Claim 30, McDonnell in view of Ganesan teaches the method of claim 29, wherein the activation key comprises one or more numerical values to indicate a number of layers for each neural network architecture in the first plurality (McDonnell, Col. 8 lines 29-38, “One way to randomize generation of the initial population of neural networks is to assign a random value to a hyperparameter (or a set of hyperparameters), where the hyperparameters indicate, for example, a number of hidden layers of the neural network, an architecture of a layer or set of layers, a number of nodes in a layer or set of layers, an activation function of a node or layer, etc. In such implementations, the random values assigned to the hyperparameters specify the architecture of the neural network.”, thus, the neural network architecture may comprise a number of layers as indicated by the activation key/hyperparameter).
Regarding Claim 31, McDonnell in view of Ganesan teaches the method of claim 29, wherein the activation key comprises one or more numerical values to indicate one or more neural network blocks to be used in one or more layers for each neural network architecture in the first plurality (McDonnell, Col. 8 lines 29-38, “One way to randomize generation of the initial population of neural networks is to assign a random value to a hyperparameter (or a set of hyperparameters), where the hyperparameters indicate, for example, a number of hidden layers of the neural network, an architecture of a layer or set of layers, a number of nodes in a layer or set of layers, an activation function of a node or layer, etc. In such implementations, the random values assigned to the hyperparameters specify the architecture of the neural network.”, thus, the neural network architecture may comprise a number of neural network blocks as indicated by the activation key/hyperparameter. Applicant’s specification Par. [0081] defines a “neural network block” as any module and/or block comprising at least convolutional blocks/layers, or any other type of neural network block capable of use in performing neural network operations related to neural network functions).
Regarding Claim 33, McDonnell in view of Ganesan teaches the one or more processors of claim 1, wherein the circuitry is further to update the one or more parameters of the one or more first neural networks by training the one or more first neural networks to perform a task and identify the first set of one or more accuracy metrics associated with performing the task (McDonnell, Col. 21 lines 14-20, “In this example, the evolutionary operations performed and the neural networks used for the evolutionary operations are based on the estimated fitness data 310 and the parameters 118. Generating the population 320 of neural networks can also include using the trainer 318 to train (e.g., adapt link weights) of one or more of the neural networks.”, thus, the one or more parameters (weights) may be updated of the one or more first neural networks by training the networks to perform a task (to produce an inference/prediction) and identify accuracy metrics (fitness data) associated with performing the task (producing an inference/prediction)).
Regarding Claim 34, McDonnell in view of Ganesan teaches the one or more processors of claim 1, wherein the one or more parameters of the one or more first neural networks comprise one or more neural network weights that are to be modified by training the one or more first neural networks (McDonnell, Col. 21 lines 14-20, “In this example, the evolutionary operations performed and the neural networks used for the evolutionary operations are based on the estimated fitness data 310 and the parameters 118. Generating the population 320 of neural networks can also include using the trainer 318 to train (e.g., adapt link weights) of one or more of the neural networks.”, thus, the one or more parameters of the one or more first neural networks of the population may comprise one or more weights that are to be modified during training).
12. Claims 24 and 32 are rejected under 35 U.S.C. 103 as being unpatentable over McDonnell et al. (hereinafter McDonnell) (US Patent 10685286), in view of Ganesan et al. (hereinafter Ganesan) (US Patent 11494587), further in view of Weng et al. (hereinafter Weng) (“NAS-Unet: Neural Architecture Search for Medical Image Segmentation”)
Regarding Claim 24, McDonnell in view of Ganesan teaches the non-transitory machine-readable medium of claim 17 including the set of instructions to be performed by one or more processors.
McDonnell in view of Ganesan does not explicitly disclose select[ing] the one or more of the neural network architectures based, at least in part, on a time to perform segmentation of one or more medical images
However, Weng teaches select[ing] the one or more of the neural network architectures based, at least in part, on a time to perform segmentation of one or more medical images (Weng, Pg. 1, Abstract, “In this paper, we design three types of primitive operation set on search space to automatically find two cell architecture DownSC and UpSC for semantic image segmentation especially medical image segmentation. Inspired by the U-net architecture and its variants successfully applied to various medical image segmentation, we propose NAS-Unet which is stacked by the same number of DownSC and UpSC on a U-like backbone network.” & Pg. 3, “We show that the performance of NAS-Unet outper forms U-Net and one of its variants (FC-Densenet) in all types of medical image segmentation datasets we evaluated without using any pre-trained backbone. The training time of NAS-Unet closes to U-Net, but parameters amount is only 6%. FC-Densenet has twice memory cost than ours.”, thus, Weng teaches using neural architecture search (NAS) to iteratively generate and select optimal design for a neural network (See Weng Pg. 4 Section B. Neural Architecture Search) based on a time (training time of the network) to perform segmentation of one or more medical images).
It would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the medium of Claim 17, as disclosed by McDonnell in view of Ganesan to include selecting the one or more of the neural network architectures based, at least in part, on a time to perform segmentation of one or more medical images, as disclosed by Weng. One of ordinary skill in the art would have been motivated to make this modification to enable the use of deep learning for medical image segmentation, such that the selected neural network may intuitively perform medical image analysis to improve diagnostic efficiency (Weng, Pg. 1, Section 1 Introduction, “Medical image analysis is the first step in analyzing medical images, which helps to make images more intuitive and improves diagnostic efficiency. Medical image segmentation is a critical step in the field of medical image analysis. In order to provide a reliable basis for clinical diagnosis and pathology research, and assist doctor to make a more accurate diagnosis, it need to segment the parts of medical images we focus and extract relevant features.”)
Regarding Claim 32, McDonnell in view of Ganesan teaches the method of claim 26.
McDonnell in view of Ganesan does not explicitly disclose wherein one or more architectures of the second subset performs segmentation of medical images with maximal accuracy
However, Weng teaches wherein one or more architectures of the second subset performs segmentation of medical images (Weng, Pg. 1, Abstract, “In this paper, we design three types of primitive operation set on search space to automatically find two cell architecture DownSC and UpSC for semantic image segmentation especially medical image segmentation. Inspired by the U-net architecture and its variants successfully applied to various medical image segmentation, we propose NAS-Unet which is stacked by the same number of DownSC and UpSC on a U-like backbone network.”, thus, the neural networks may perform segmentation on one or more medical images) with maximal accuracy (Weng, Pg. 2, Section I. Introduction, “In addition, U-net uses the skip connection operation to connect each pair of down-sampling layer and the up-sampling layer, which makes the spatial information directly applied to much deeper layers and a more accurate segmentation result.”, therefore, the neural networks may perform segmentation of medical images with maximal accuracy)
It would have been obvious for one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the method of claim 26, as disclosed by McDonnell in view of Ganesan to include wherein one or more architectures of the second subset performs segmentation of medical images with maximal accuracy, as disclosed by Weng. One of ordinary skill in the art would have been motivated to make this modification to enable the use of deep learning for medical image segmentation, such that the selected neural network may intuitively perform medical image analysis to improve diagnostic efficiency (Weng, Pg. 1, Section 1 Introduction, “Medical image analysis is the first step in analyzing medical images, which helps to make images more intuitive and improves diagnostic efficiency. Medical image segmentation is a critical step in the field of medical image analysis. In order to provide a reliable basis for clinical diagnosis and pathology research, and assist doctor to make a more accurate diagnosis, it need to segment the parts of medical images we focus and extract relevant features.”)
Conclusion
13. Any inquiry concerning this communication or earlier communications from the examiner should be directed to Devika S Maharaj whose telephone number is (571)272-0829. The examiner can normally be reached Monday - Thursday 8:30am - 5:30pm.
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, Alexey Shmatov can be reached on (571)270-3428. 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.
/DEVIKA S MAHARAJ/Examiner, Art Unit 2123
Prosecution Insights