Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Remarks
This Office Action is responsive to Applicants' Amendment filed on December 24, 2025, in which claims 1, 7, and 10-12 have been amended. No claims have been newly added or cancelled.
Claims 1-12 are currently pending.
Response to Arguments
With regards to the objection of claim 10 for a minor informality, Applicant has amended claim 10 to correct the previously noted informality, and thus the objection is withdrawn.
With regards to the rejections of claims 5-7 and 11 under 35 U.S.C. 112(b), Examiner finds Applicant’s arguments are not persuasive. Applicant argues that the claims have been amended to obviate the rejections. Examiner agrees that claims 7 and 11 have been amended to remove the previously noted causes of indefiniteness, however those amendments have introduced a new cause of indefiniteness, and thus the rejections of claims 7 and 11 are maintained. Examiner additionally notes that claims 5 and 6 are not amended, and thus the 112(b) rejections of those claims are maintained. Examiner further notes that the limitations of claim 6 have been substantially incorporated into claims 1 and 12, and thus those claims and any dependents are newly rejected under 35 U.S.C. 112(b), on analogous grounds.
With regards to the rejections of claims 10 and 11 under 35 U.S.C. 101 as directed to non-statutory subject matter, Examiner notes that Applicant has not explicitly argued as to why these rejections have been overcome by the amendments to the claims. Although claims 10 and 11 have been amended to recite a “A non-transitory machine learning system” and “A non-transitory training system” respectively, these amendments do not cause claims 10 and 11 to be directed towards statutory subject matter, as they still recite computer systems without tangible hardware components, that is, software per se. Thus, the 101 rejections of claims 10 and 11 for being directed to non-statutory subject matter are maintained.
With regards to the rejections of claims 1-12 under 35 U.S.C. 101 as directed to abstract ideas without significantly more, Applicant argues that the claims as amended overcome the 101 rejections. Examiner respectfully disagrees. Applicant presents one set of arguments for independent claims 1, 10, and 12, and another set of arguments for independent claims 7 and 11.
With respect to the rejections of claims 1, 10, and 12 under 35 U.S.C. 101, Applicant first presents a summary of Ex parte Desjardins and alleges that the previous office action does not conform with the standards set forth in Ex parte Desjardins. Applicant then argues that the 101 analysis of the previous office action is invalid at Step 2A, Prong One “because (1) it fails to explain why the human mind is equipped to perform the limitation ‘an output of a hidden layer of the second machine learning system is used as additional input to the third machine learning system for determining the deviation’ and (2) it fails to conform to the admonition of the Desjardins decision to avoid evaluating claim limitations at a high level of generality when performing an eligibility analysis”, on page 10 of the Remarks. Examiner respectfully disagrees with both assertions.
With respect to assertion (1) of Applicant, Applicant states on page 10 of the Remarks that “the Patent Office simply offers the naked assertion that ‘recites an evaluation of the output of a hidden layer, which is a mental process, which is an abstract idea, regardless of if it's performed on a generic computer’” and alleges Examiner’s prior statement does not explain why the limitation in question is a mental process, and merely “declares” it to be so.
Examiner considers their statement that the limitation “recites an evaluation of the output of a hidden layer” is an explanation, albeit a brief one. The limitation verbatim recites an output of a hidden layer of the second machine learning system is used as additional input to the third machine learning system for determining the deviation, which Examiner explained is “an evaluation of the output of a hidden layer”. To be more verbose, Examiner clarifies here that an output of a hidden layer is a numerical value, the deviation is a numerical value, and determining the deviation wherein recites an output of a hidden layer of the second machine learning system is used is just considering one numerical value, the output of a hidden layer, while determining another numerical value, the deviation, which is well within the abilities of a human mind. Considering the output of a hidden layer is an evaluation of the output of the hidden layer, with evaluations being recognized as mental processes within MPEP 2106.04(a)(2).III. That generic machine learning systems produce the values of the output of a hidden layer and the deviation amount to the use of generic computers to perform the mental process, which does not prevent the limitation from reciting a mental process, see MPEP 2106.04(a)(2).III.C.1. Although Examiner has gone into exhaustive detail here, Examiner notes that this is not required in general practice, and Examiner’s earlier explanation that the limitation “recites an evaluation of the output of a hidden layer” is sufficient, as are the similarly concise explanations for all other classifications of limitations as reciting mental processes.
With respect to assertion (2) of Applicant, Applicant states on page 10 of the Remarks that “Rather than apply the Prong One analysis to the claim limitation as it is written, the Patent Office recharacterizes the limitation at an improperly high level of generality by reducing it to ‘an evaluation of the output of a hidden layer’ and then pronounces this bowdlerized version of the full claim limitation (not the actual claim limitation) to be a mental process” and “Here, the claim limitation is directed to the technological interrelationship between three pieces of technology that can only ever exist in a computing environment: a hidden layer and two machine learning systems. No human mind is equipped to execute the way that the output of the hidden layer is used as an additional input for a third machine learning system”, which Applicant further argues on page 11 of the Remarks provides a technological improvement of allowing the machine learning system to self-correct, and that under Desjardins this is an improvement to how a machine learning model itself operates, which would not be an abstract idea.
Examiner earlier analyzed the limitation in exhaustive detail in response to assertion (1) of Applicant. To reiterate, determining the deviation by additionally using an output of a hidden layer amounts to an evaluation of an output of a hidden layer, in addition to the evaluation of the deviation, which is also a mental process, as stated by Examiner in the analysis of an earlier limitation of claim 1. Use of generic machine learning systems to accomplish this is just using generic computers to perform the mental processes, which does not change that mental processes are ineligible judicial exceptions. As Applicant themselves has stated, Desjardins is concerned with ensuring that claims providing an improvement to how a machine learning model itself operates fall under the scope of patent protection. Examiner refers to “Advance notice of change to the MPEP in light of Ex Parte Desjardins”, which states on page 2:
“the ARP upheld the Step 2A Prong One finding that the claims recited an abstract idea (i.e., mathematical concept). In Step 2A Prong Two, the ARP then determined that the specification identified improvements as to how the machine learning model itself operates, including training a machine learning model to learn new tasks while protecting knowledge about previous tasks to overcome the problem of ‘catastrophic forgetting’…Importantly, the ARP evaluated the claims as a whole in discerning at least the limitation ‘adjust the first values of the plurality of parameters to optimize performance of the machine learning model on the second machine learning task while protecting performance of the machine learning model on the first machine learning task’ reflected the improvement disclosed in the specification”.
Examiner notes that Ex Parte Desjardins is not concerned with Step 2A, Prong One, where limitations may be found to recite judicial exceptions, which is at issue here. Examiner further notes that the limitation at issue, the second machine learning system is a neural network and an output of a hidden layer of the second machine learning system is used as additional input to the third machine learning system for determining the deviation, is not concerned with improving “how the machine learning model itself operates”. A generic neural network’s hidden layer output is used as input to a different machine learning system, which determines a deviation. No machine learning model is improved here, rather, the data which is variably input or output of machine learning systems is manipulated. For at least these reasons, the findings of Ex Parte Desjardins are not relevant here.
Applicant further argues with respect to the rejections of claims 1, 10, and 12 under 35 U.S.C. 101, that the 101 analysis of the previous office action is invalid at Step 2A, Prong Two, on pages 12-14 of the Remarks. Specifically, on page 13 of the Remarks, Applicant states: “a goal of the claimed invention is a self-correcting machine learning system that corrects inaccurate predictions before they are output” and “[0030] and [0031] explain that the use of the output of the hidden layer of the second machine learning system as an additional input of a third machine learning system contributes to the accomplishment of this accuracy-improving goal. Moreover, since the claimed use of this hidden layer output as an additional input is recited in the claim, the claim conforms to the stricture applied in Desjardins that ‘the claims reflect such an improvement’”, and thus the limitation an output of a hidden layer of the second machine learning system is used as additional input to the third machine learning system for determining the deviation integrates any recited abstract ideas into a practical application.
Examiner respectfully disagrees. First, as Applicant acknowledged earlier, on page 11 of their Remarks, Desjardins is concerned with ensuring that improvements to how a machine learning model “itself” operates, that would be classified as abstract ideas, fall under the scope of patent protection. Page 2, lines 16-21 of the specification as filed read: “b. Determining an intermediate signal characterizing a classification and/or regression result of the feature representation; c. Predicting, based on the feature representation and the intermediate signal, a deviation of the intermediate signal from a desired output signal of the input signal;”. One of ordinary skill in the art understands a classification and/or regression result to be a result of a machine learning model. Page 4, lines 9-11 of the specification as filed read: “The intermediate signal may be understood as a signal which could possibly be used as output signal also but may be corrected in the method before being put out as output signal”. Examiner thus understands that Applicant’s invention, at least as reflected in claims 1, 10, and 12, is not directed towards how a machine learning model itself operates, but rather a system that corrects the output of a machine learning model via a mechanism external to the machine learning model. For at least that reason, the findings of Ex Parte Desjardins are not relevant here.
Beyond the findings of Ex Parte Desjardins, MPEP 2106.04(d).III. states “Because a judicial exception alone is not eligible subject matter, if there are no additional claim elements besides the judicial exception, or if the additional claim elements merely recite another judicial exception, that is insufficient to integrate the judicial exception into a practical application”, i.e. that an improvement provided solely by improving an abstract idea itself is ineligible. Examiner further notes that most limitations within claims 1, 10, and 12 have been identified as reflecting abstract ideas. The only further elements within claim 1, 10, and 12 that are not identified as abstract ideas relate to applying generic machine learning systems/models, or mere data outputting, as shown in the 101 rejections below.
Applicant further argues with respect to the rejections of claims 1, 10, and 12 under 35 U.S.C. 101, that the 101 analysis of the previous office action is invalid at Step 2B, on pages 14 and 15 of the Remarks. Applicant argues that analysis of additional elements beyond any recited abstract ideas as well-understood, routine, and conventional activity requires a factual determination, i.e. evidence, in accordance with Berkheimer.
Examiner notes that their 101 analysis, both in the prior office action and in the current office action, is in compliance with Berkheimer. “MPEP 2106.05(d).II., example (i) of WURC computer functions” is recited as evidence for determinations of data outputting limitations as well-understood, routine, and conventional activity. This is necessary as MPEP 2106.05.II instructs examiners to “Re-evaluate any additional element or combination of elements that was considered to be insignificant extra-solution activity per MPEP § 2106.05(g), because if such re-evaluation finds that the element is unconventional or otherwise more than what is well-understood, routine, conventional activity in the field, this finding may indicate that the additional element is no longer considered to be insignificant;”. Classification of claim limitations as mere instructions to apply, however, does not require such a determination, and thus does not require Berkheimer evidence.
With respect to the rejections of claims 7 and 11 under 35 U.S.C. 101, Applicant first argues that the 101 analysis of the previous office action is invalid at Step 2A, Prong One, stating on page 16 of the Remarks that “the analysis of claims 7 and 11 simply labels the training limitations as mental processes, without providing any explanation as to why they fit into the definition of mental processes as given in the MPEP” and “since a claim limitation qualifies for analysis under Prong Two only if it does not recite a judicial exception, the fact that the Director in Desjardins analyzed the training limitation…under Prong Two carries with it the implicit judgment that this limitation does not recite any judicial exception at all”.
Examiner respectfully disagrees. With respect to Applicant’s first point, Examiner considers their statement that the limitation of training (i) the second machine learning model, or (ii) the first machine learning model and the second machine learning model, to determine a desired output signal for a provided first training input signal “recites an evaluation of a desired output signal, which is a mental process, which is an abstract idea, regardless of if it’s performed on a generic computer or using generic machine learning models”, is an explanation, albeit a brief one. [D]etermin[ing] a desired output signal is an evaluation of a desired output signal based on a provided first training input signal, with evaluations being recognized as mental processes within MPEP 2106.04(a)(2).III. That at least one generic machine learning model produces the value of a desired output signal is amount to the use of a generic computer to perform the mental process, which does not prevent the limitation from reciting a mental process, see MPEP 2106.04(a)(2).III.C.1, and generically recited training of the model or models is just mere application of training to perform the mental process.
With respect to Applicant’s second point, “Advance notice of change to the MPEP in light of Ex Parte Desjardins” states on page 2:
“the ARP upheld the Step 2A Prong One finding that the claims recited an abstract idea (i.e., mathematical concept). In Step 2A Prong Two, the ARP then determined that the specification identified improvements as to how the machine learning model itself operates, including training a machine learning model to learn new tasks”.
In the quoted second of Desjardins, it can be seen that the ARP (Appeals Review Panel) did consider the limitation under Step 2A, Prong One, where it was found that the claims did recite an abstract idea. For at least that reason, the findings of Ex Parte Desjardins are not relevant here.
Applicant further argues the rejections of claims 7 and 11 under 35 U.S.C. 101, that the 101 analysis of the previous office action is invalid at Step 2A, Prong Two, stating on page 17 and 18 of the Remarks that “the ‘after training’ claim limitation is an ‘additional element’ that qualifies for analysis under Prong Two” and that this limitation provides a technological improvement as “As further explained in [0012] of the specification, training the third machine learning model to determine a deviation of an intermediate signals permits the claimed system to achieve the improvement of higher accuracy by self-correcting its intermediate signal”.
Examiner respectfully disagrees, and sees several flaws in Applicant’s argument here. First, although claims 7 and 11 recite training the third machine learning model to determine a deviation of an intermediate signal…for a provided second training input signal to a desired output signal, Examiner notes that no correction of the intermediate signal to the desired output signal is actually recited in claim 7 or 11, only determining of a deviation between the two. That is, that the claim does not perform the step that Applicant states provides the technological improvement. Second, the steps the claim actually recites, determin[ing] a desired output signal and determin[ing] a deviation of an intermediate signal…for a provided second training input signal to a desired output signal, are evaluations that could be performed purely mentally by a human mind, and therefore even if they represented an improvement to technology would be ineligible subject matter due to being an improvement to an abstract idea itself, see MPEP 2106.04(d).III, particularly “Because a judicial exception alone is not eligible subject matter, if there are no additional claim elements besides the judicial exception, or if the additional claim elements merely recite another judicial exception, that is insufficient to integrate the judicial exception into a practical application”. Third, even if the training of the generic machine learning models were considered separately beyond being use of generic computers or computer components for performing the abstract ideas, they would be merely applied to perform the abstract ideas, which does not integrate the abstract ideas into a practical application, see MPEP 2106.04(d).I and MPEP 2106.05(f).
Applicant further argues the rejections of claims 7 and 11 under 35 U.S.C. 101, that the 101 analysis of the previous office action is invalid at Step 2B, for the same reasons as for the arguments Applicant previously presented for the rejections of claims 1, 10, and 12. Examiner notes that for the rejections of claims 7 and 11, presentation of Berkheimer evidence is not necessary due to the lack of classification of limitations as extra-solution activity by Examiner.
With regards to the rejections of claims 1, 2, 3, and 12 under 35 U.S.C. 102(a)(1) as being anticipated by Gurney “An Introduction to Neural Networks”, Applicant has amended independent claims 1 and 12 to incorporate limitations from claims 4 and 6 that were acknowledged by the Examiner as not taught by Gurney, and thus the rejections are overcome. However, claims 1, 2, 3, and 12 are now rejected under 35 U.S.C. 103 under the same combination of art as applied for the rejection of claim 6, as presented below, with the change in the grounds of rejection being necessitated by Applicant’s amendments.
Further, Applicant argues with regards to the rejection of claim 6 under 35 U.S.C. 103 as unpatentable over Gurney “An Introduction to Neural Networks” in view of Denil et al. (U.S. Patent Application Publication No. 2019/0220748), further in view of Baker (U.S. Patent Application Publication No. 2021/0342683), as well as with regards to the 102 rejections under Gurney, with arguments against the mappings to Gurney as well as the mappings to the combination of Gurney and Baker. Examiner respectfully disagrees.
Applicant first argues on page 18 of the Remarks that “it is not the case that Gurney discloses ‘predicting, based on the feature representation and the intermediate signal, a deviation of the intermediate signal from a desired output signal of the input signal.’”. Applicant further argues on page 19 of the Remarks that “the Patent Office appears to believe that the claimed deviation corresponds to the value δ of Gurney. Earlier in this rejection, the Patent Office maps the claimed feature representation in Figure 4.10(d), and maps to claimed intermediate signal to the outputs from the ‘hidden units’ of Gurney. Given these mappings, the value of δ that the Patent Office maps to the claimed deviation would have to be predicted based on Figure 4.10(d) and on the output of the hidden units of Gurney, but the above Office Action blurb concerning the mapping of the claimed intermediate signal to δ does not establish δ as being based on these two things”.
Examiner respectfully disagrees. With respect to the deviation being based on a feature representation, Figure 4.10(d) showcases a single example of determining a feature representation, which Gurney refers to as a training pattern, from an image. Examiner’s mapping of Gurney to step (c) of claim 1 in the prior office action reads:
(Gurney Pg. 43) “we expand the main step of ‘train on that pattern’ into the following steps…4. Apply the target pattern to the output layer. 5. Calculate the δs on the output nodes”, (Gurney Pg. 42) “The contribution that node j makes towards the error is, of course, expressed in the ‘δ’ for that node δj”.
As can be seen, application of a pattern, or a feature representation, to an output layer is a preceding step in the calculation of each δ, which represent the contribution of each output node to the total error, which is the deviation. Further, Examiner omitted many elements for brevity, but the algorithm referenced in Gurney reads in full:
“((Gurney Pg. 43) “we expand the main step of ‘train on that pattern’ into the following steps.
1. Present the pattern at the input layer.
2. Let the hidden units evaluate their output using the pattern.
3. Let the output units evaluate their output using the result in step 2 from the hidden units.
4. Apply the target pattern to the output layer.
5. Calculate the δs on the output nodes.
6. Train each output node using gradient descent (6.4).
7. For each hidden node, calculate its δ according to (6.6).
8. For each hidden node, use the δ found in step 7 to train according to gradient descent (6.2)”.
It can be seen that a feature representation pattern is used repeatedly in this algorithm in relation to the output units/layer/nodes, which is/are used to calculate δs. Examiner considers this sufficient to teach that the δs are based on a feature representation, and as the total error is based on the δs as previously recited from Gurney, the deviation is ultimately based on a feature representation. With respect to the deviation being based on an intermediate signal, Examiner likewise refers to the full algorithm on page 43 of Gurney: the output of the hidden units in step 2 is an intermediate signal, in step 3 the output units compute their output from that of the hidden units, and in step 5 the δs are calculated from the output nodes, thus the δs are based on an intermediate signal, and so is the error/deviation. Examiner apologies if this was not clear from the previous mapping; in the current office action the mapping to the algorithm on page 43 of Gurney has been expanded to show it in full for the mapping of step (c).
Applicant further argues on page 19 of the Remarks that: “Moreover, it is not even the case that δ, the purported ‘deviation,’ is something that is predicted. The above Office Action blurb instead states that δ is calculated in terms of the ‘contribution that node j makes towards the error.’ How that amounts to a prediction of any sort is left unexplained by the Patent Office”.
Examiner notes that “predicting” or “prediction” is not explicitly defined within the instant application’s specification, and thus the broadest reasonable interpretation of the term is used when mapping the claims. Within the art of machine learning, a prediction generally refers to the output of a model. In the algorithm on page 43 of Gurney, this corresponds to step “3. Let the output units evaluate their output using the result in step 2 from the hidden units”, as the output units give the output of a model, the prediction. As the relevant algorithm of Gurney is a training algorithm, this prediction is compared to a target prediction at step “4. Apply the target pattern to the output layer”. Then the difference between the model’s prediction and the desired prediction is computed at step “5. Calculate the δs on the output nodes”. As the δs, which are used to calculate the overall error/deviation, are the result of the difference between two predictions, Examiner considers that the error/deviation is also a prediction under the broadest reasonable interpretation of the term, as it is a mere mathematical transformation of predictions. In the current mapping of claim 1 this has been made more explicit.
Applicant further argues on pages 19 and 20 of the Remarks that “it is not the case that Gurney discloses ‘the second machine learning system is a neural network and an output of a hidden layer of the second machine learning system is used as additional input to the third machine learning system for determining the deviation.’” and “since the Patent Office earlier mapped the claimed ‘deviation’ to value δ of Gurney, then in order to meet the limitation that the output of the hidden layer is used ‘for determining the deviation,’ then the Patent Office had the burden of demonstrating that Baker uses its inner node activation values to determine not just any deviation, but the particular deviation that earlier in the rejection the Patent Office mapped to value δ of Gurney” (emphasis Applicant’s).
Examiner respectfully disagrees. In response to applicant's argument that the particular deviation determined in Gurney differs from the particular deviation determined in Baker, the test for obviousness is not whether the features of a secondary reference may be bodily incorporated into the structure of the primary reference; nor is it that the claimed invention must be expressly suggested in any one or all of the references. Rather, the test is what the combined teachings of the references would have suggested to those of ordinary skill in the art. See In re Keller, 642 F.2d 413, 208 USPQ 871 (CCPA 1981). Gurney teaches a deviation of an intermediate signal from a desired signal, based on a feature representation and an intermediate signal. Baker teaches a deviation determined in part from the output of a hidden layer of a neural network, which is a distinct machine learning model from a first machine learning model. One would have motivation to combine these references so that determining the deviation of Gurney would additionally use the output of another neural network’s hidden layer, even if the deviation of Baker is not calculated in the exact same manner as the deviation of Gurney is calculated. Examiner nevertheless notes that Baker states it uses the backpropagation algorithm for derivatives: (Baker [0003]) “The second machine learning system is trained to optimize a specified analysis objective, and it also may back propagate the derivatives of that analysis objective to the first machine learning system”, and that backpropagation is the same algorithm relied upon in Gurney to teach its deviation: (Gurney Pg. 43) “Step 7 involves propagating the δs back from the output nodes to the hidden units—hence the name backpropagation. The backpropagation (BP) algorithm is also known as error backpropagation or back error propagation”.
With regards to the rejections of claims 7 and 11 under 35 U.S.C. 102(a)(1) as anticipated by Denil et al. (U.S. Patent Application Publication No. 2019/0220748), Applicant argues that Denil does not teach the subject matter of claims 7 and 11 as claimed. Examiner respectfully disagrees.
Applicant first argues on page 21 of the Remarks that “it is not the case that Denil discloses ‘a fourth machine learning system including a first machine learning model, a second machine learning model, and a third machine learning model’”, and elaborates further that “’other machine learning models’ are never described as forming part of a collective of other machine learning systems that together constitute a single overall machine learning system. The blurb does not recite any sort of interoperability among these ‘other machine learning models’…Figure 1B shows, at most, a machine learning model 102 interacting with a recurrent neural network 106, not first through third machine learning models that together constitute a fourth machine learning model”.
Examiner respectfully disagrees. First, claim 7 and 11 both recite a fourth machine learning system including a first machine learning model, a second machine learning model, and a third machine learning model, not a fourth machine learning model. The broadest reasonable interpretation of a machine learning system, consisting of several models, is not limited to the system being a machine learning model. Examiner considers the system of Denil, which includes three separate machine learning models, to be the “fourth machine learning system” of the claim. The claim does not further limit the “fourth machine learning system” beyond the fact that it includes the three machine learning models and that it is trained.
Second, Denil Fig. 1B, reproduced below, clearly shows a “Machine learning Model 102” and a “Recurrent Neural Network 106” interoperating, with the values of various variables being passed between them. Examiner notes that a recurrent neural network is itself a type of machine learning model. Denil further elaborates: (Denil [0060]) “In some implementations a learned update rule, as given by equation (2) above, may be applied to other machine learning models that are configured to perform similar machine learning tasks, e.g., machine learning tasks with a similar structure. For example, the learned update rule may be applied to a second machine learning model that is configured to perform a same machine learning task as the first machine learning model (e.g., the machine learning model 102 of FIG. 1A), but where the second machine learning model includes a different number of hidden units or neural network layers than the first machine learning model”. Examiner submits that the second machine learning model disclosed in Denil [0060], the “Machine learning Model 102”, and the “Recurrent Neural Network 106” explicitly anticipate a first machine learning model, a second machine learning model, and a third machine learning model as described in claims 7 and 11, and that the second machine learning model of Denil interoperates with the “Recurrent Neural Network 106” of Denil, just as the “Machine learning Model 102” does, as explicitly stated in Denil: (Denil [0061]) “Applying the learned update rule to other machine learning models in these examples can be achieved using the coordinate-wise RNN”.
PNG
media_image1.png
727
1025
media_image1.png
Greyscale
Applicant further argues with respect to the 102 rejection of claims 7 and 11, on page 21 of the Remarks, that “Applicant disputes that Denil discloses ‘after training the second machine learning model or after training the first machine learning model and the second machine learning model, training the third machine learning model to determine a deviation of an intermediate signal determined from the first machine learning model and the second machine learning model...’”, and further elaborating on page 22 of the Remarks that “As indicated in the above blurb, the RNN 106 of Denil is mapped to the third machine learning model. However, if that is so, then it is not the case that Denil discloses ‘training the third machine learning model to determine a deviation’ because [0034] of Denil, on which the Patent Office relies to meet the claimed deviation determination, discloses the operation of machine learning model 102, not RNN 106”.
Examiner respectfully disagrees. Although Denil [0034] does state that machine learning model 102 is trained via the determination of a deviation, the model does not train itself, and Denil soon makes clear that the training is done by training module 104 using the RNN 106, in: (Denil [0035]) “The training module 104 communicates with the machine learning models 102 and the RNN 106. The training module 104 is configured to train the machine learning model 102 by determining a learned parameter update rule for the machine learning model parameters using the RNN 106”, with the creation of the update rule being dependent on a determined deviation. Examiner attempted to clarify this by mapping to (Denil [0051]) “Determining the update rule for the machine learning model parameters using the RNN includes training the RNN to determine RNN parameters that minimize the RNN objective function, and using trained RNN parameters to determine a final update rule that is used to generate the trained machine learning model” as well.
Claim Rejections - 35 USC § 112(b)
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-9, 11, and 12 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Regarding claim 1,
Claim 1 recites the limitation and wherein: the second machine learning system is a neural network and an output of a hidden layer of the second machine learning system is used as additional input to the third machine learning system for determining the deviation, which recites the terms “the second machine learning system” and “the third machine learning system”. However, claim 1 earlier recites wherein: the feature representation is determined by a first machine learning system using the input signal as input of the first machine learning system; and/or the intermediate signal is determined by a second machine learning system using the feature representation as input of the second machine learning system; and/or the deviation is determined by a third machine learning system using the feature representation and the intermediate signal as input of the third machine learning system. It is therefore unclear whether the second and third machine learning systems recited later in claim 1 both have antecedent basis, because earlier claim 1 only needs to include one of the recited first, second, and/or third machine learning systems. For examination purposes, the terms will be interpreted as reading “a second machine learning system” and “a third machine learning system”.
In reference to dependent claims 2-9, claims 2-9 do not cure the deficiencies noted in the rejection of claim 1. Therefore, these claims are rejected under the same rationale as claim 1.
Regarding claim 5,
Claim 5 recites the limitation wherein the first machine learning system, the second machine learning system, and the third machine learning system are parts of a fourth machine learning system. However, the parent claim of claim 5, claim 4, recites the feature representation is determined by a first machine learning system…and/or the intermediate signal is determined by a second machine learning system…and/or the deviation is determined by a third machine learning system. It is therefore unclear whether the first, second, and third machine learning systems recited in claim 5 all have antecedent basis, because the method of claim 4, upon which claim 5 depends, only needs to include one of the recited machine learning systems. Therefore, the scope of the claim is indefinite. For examination purposes, the limitation will be interpreted as reading “wherein a first machine learning system, a second machine learning system, and a third machine learning system are parts of a fourth machine learning system”.
Regarding claim 6,
Claim 6, which depends on claim 4, recites the terms “the second machine learning system” and “the third machine learning system”. It is therefore indefinite for the same reason claim 5 is indefinite. For examination purposes, the terms will be interpreted as reading “a second machine learning system” and “a third machine learning system”.
Regarding claim 7,
Claim 7 recites the limitation wherein the first machine learning model and the second machine learning model are not trained during the training of the third machine learning system. However, it is unclear what the term “the third machine learning system” refers to, as no “third machine learning system” is referred to previously within claim 7, although the similar term “a third machine learning model” is recited earlier in claim 7. The term “the third machine learning system” thus lacks antecedent basis. For examination purposes, the limitation will be interpreted as reading “wherein the first machine learning model and the second machine learning model are not trained during the training of the third machine learning model”.
Regarding claim 11,
Claim 11 recites a system for performing the function of the method of claim 7. All other limitations in claim 11 are substantially the same as those in claim 7, therefore the claim is considered indefinite with an equivalent rationale and is interpreted for examination purposes in the same way.
Regarding claim 12,
Claim 12 recites a non-transitory machine-readable storage medium for performing the function of the method of claim 1. All other limitations in claim 12 are substantially the same as those in claim 1, therefore the claim is considered indefinite with an equivalent rationale and is interpreted for examination purposes in the same way.
Claim Rejections - 35 USC § 112(d)
The following is a quotation of 35 U.S.C. 112(d):
(d) REFERENCE IN DEPENDENT FORMS.—Subject to subsection (e), a claim in dependent form shall contain a reference to a claim previously set forth and then specify a further limitation of the subject matter claimed. A claim in dependent form shall be construed to incorporate by reference all the limitations of the claim to which it refers.
The following is a quotation of pre-AIA 35 U.S.C. 112, fourth paragraph:
Subject to the following paragraph [i.e., the fifth paragraph of pre-AIA 35 U.S.C. 112], a claim in dependent form shall contain a reference to a claim previously set forth and then specify a further limitation of the subject matter claimed. A claim in dependent form shall be construed to incorporate by reference all the limitations of the claim to which it refers.
Claims 4 and 6 are rejected under 35 U.S.C. 112(d) or pre-AIA 35 U.S.C. 112, 4th paragraph, as being of improper dependent form for failing to further limit the subject matter of the claim upon which it depends, or for failing to include all the limitations of the claim upon which it depends.
Regarding claim 4,
Claim 4 recites the limitations the feature representation is determined by a first machine learning system using the input signal as input of the first machine learning system; and/or the intermediate signal is determined by a second machine learning system using the feature representation as input of the second machine learning system; and/or the deviation is determined by a third machine learning system using the feature representation and the intermediate signal as input of the third machine learning system. However, its parent claim, claim 1, recites the limitations the feature representation is determined by a first machine learning system using the input signal as input of the first machine learning system; and/or the intermediate signal is determined by a second machine learning system using the feature representation as input of the second machine learning system; and/or the deviation is determined by a third machine learning system using the feature representation and the intermediate signal as input of the third machine learning system. Therefore, all limitations in claim 4 recite only elements already recited within the claim upon which it depends.
Therefore, claim 4 fails to further limit the subject matter of its parent claim.
Regarding claim 6,
Claim 6 recites the limitations wherein the second machine learning system is a neural network and an output of a hidden layer of the second machine learning system is used as additional input to the third machine learning system for determining the deviation. However, claim 1, which claim 6 ultimately depends on, recites the limitations and wherein: the second machine learning system is a neural network and an output of a hidden layer of the second machine learning system is used as additional input to the third machine learning system for determining the deviation. Therefore, all limitations in claim 6 recite only elements already recited within a claim upon which it depends.
Therefore, claim 6 fails to further limit the subject matter of its parent claims.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a No therefor, subject to the conditions and requirements of this title.
Claims 10 and 11 are rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter. The claims do not fall within at least one of the four categories of patent eligible subject matter because they recite a computer system without reciting any hardware components to implement the system, and thus the current scope of the claims falls under software per se.
Additionally, claims 1-12 are rejected under 35 U.S.C. 101 because the claimed invention is directed to abstract ideas without significantly more.
Regarding claim 1,
Step 1 - “Is the claim to a process, machine, manufacture or composition of matter?”
Yes, the claim is directed towards a process.
Step 2A, Prong 1 - “Is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?”:
The limitation of a. determining a feature representation characterizing the input signal; recites an evaluation of the input signal, which is a mental process, which is an abstract idea, regardless of if it’s performed on a generic computer.
The limitation of b. determining an intermediate signal characterizing a classification and/or regression result of the feature representation; recites an evaluation of the feature representation, which is a mental process, which is an abstract idea, regardless of if it’s performed on a generic computer.
The limitation of c. predicting, based on the feature representation and the intermediate signal, a deviation of the intermediate signal from a desired output signal of the input signal; recites an evaluation of a deviation from a desired output, which is a mental process, which is an abstract idea, regardless of if it’s performed on a generic computer.
The limitation of d. adapting the intermediate signal according to the determined deviation thereby determining an adapted signal; recites, in view of the specification (Pg. 13, lines 20-24) “In the depicted embodiment, a gradient (g) of the deviation with respect to the intermediate signal (i) is determined and the adapted signal is then determined based a gradient descent of the deviation (d)”, a mathematical calculation, which is a mathematical concept, which is an abstract idea.
The limitation of and an output of a hidden layer of the second machine learning system is used as additional input to the third machine learning system for determining the deviation recites an evaluation of the output of a hidden layer, which is a mental process, which is an abstract idea, regardless of if it’s performed on a generic computer.
Step 2A, Prong 2 - “Does the claim recite additional elements that integrate the judicial exception into a practical application?”:
The limitation of and e. providing the adapted signal as the output signal, recites the mere extra-solution activity of data outputting, which does not integrate the exception into a practical application, MPEP 2106.05(d) and 2106.05(g).
The limitation of wherein: the feature representation is determined by a first machine learning system using the input signal as input of the first machine learning system; recites mere instructions to apply the exception with a generic machine learning system, which does not integrate the exception into a practical application, MPEP 2106.05(d) and 2106.05(f).
The limitation of and/or the intermediate signal is determined by a second machine learning system using the feature representation as input of the second machine learning system; recites mere instructions to apply the exception with a generic machine learning system, which does not integrate the exception into a practical application, MPEP 2106.05(d) and 2106.05(f).
The limitation of and/or the deviation is determined by a third machine learning system using the feature representation and the intermediate signal as input of the third machine learning system, recites mere instructions to apply the exception with a generic machine learning system, which does not integrate the exception into a practical application, MPEP 2106.05(d) and 2106.05(f).
The limitation of and wherein: the second machine learning system is a neural network recites mere additional details on one of the machine learning systems used to perform the abstract ideas, without changing that performing the judicial exceptions with generic machine learning systems constitutes mere instructions to apply the exceptions, MPEP 2106.05(d) and 2106.05(f).
Step 2B - “Does the claim recite additional elements that amount to significantly more than the judicial exception?”:
The limitation of and e. providing the adapted signal as the output signal, recites transmitting data over a network, which is well-understood, routine, and conventional, MPEP 2106.05(d).II., example (i) of WURC computer functions.
The limitation of wherein: the feature representation is determined by a first machine learning system using the input signal as input of the first machine learning system; recites mere instructions to apply the exception with a generic machine learning system, which does not integrate the exception into a practical application, MPEP 2106.05(f).
The limitation of and/or the intermediate signal is determined by a second machine learning system using the feature representation as input of the second machine learning system; recites mere instructions to apply the exception with a generic machine learning system, which does not integrate the exception into a practical application, MPEP 2106.05(f).
The limitation of and/or the deviation is determined by a third machine learning system using the feature representation and the intermediate signal as input of the third machine learning system, recites mere instructions to apply the exception with a generic machine learning system, which does not integrate the exception into a practical application, MPEP 2106.05(f).
The limitation of and wherein: the second machine learning system is a neural network recites mere additional details on one of the machine learning systems used to perform the abstract ideas, without changing that performing the judicial exceptions with generic machine learning systems constitutes mere instructions to apply the exceptions, MPEP 2106.05(f).
Therefore, claim 1 is found to be ineligible subject matter under 35 U.S.C. 101.
Regarding claim 2,
Claim 2 adds the additional limitations to claim 1:
wherein the steps c. and d. are repeated iteratively until an exit criterion is fulfilled recites a judgement of an exit criterion being fulfilled, which is a mental process, which is an abstract idea, regardless of if it’s performed on a generic computer.
and wherein the adapted signal is used as intermediate signal in a next iteration recites mere additional details on how the intermediate signal is determined, without changing that the limitation b. determining an intermediate signal characterizing a classification and/or regression result of the feature representation;, recited in claim 1, is an evaluation, which is a mental process, which is an abstract idea, regardless of if it’s performed on a generic computer.
Therefore, claim 2 is found to be ineligible subject matter under 35 U.S.C. 101.
Regarding claim 3,
Claim 3 adds the additional limitations to claim 1:
wherein the deviation is predicted by a differentiable model recites mere instructions to apply the exception with a generic model, which does not integrate the exception into a practical application, MPEP 2106.05(d) and 2106.05(f).
and the intermediate signal is adapted based on a gradient of the deviation with respect to the intermediate signal a mathematical formula, which is a mathematical concept, which is an abstract idea.
Therefore, claim 3 is found to be ineligible subject matter under 35 U.S.C. 101.
Regarding claim 4,
Claim 4 adds the additional limitations to claim 1:
the feature representation is determined by a first machine learning system using the input signal as input of the first machine learning system; recites mere instructions to apply the exception with a generic machine learning system, which does not integrate the exception into a practical application, MPEP 2106.05(d) and 2106.05(f).
and/or the intermediate signal is determined by a second machine learning system using the feature representation as input of the second machine learning system; recites mere instructions to apply the exception with a generic machine learning system, which does not integrate the exception into a practical application, MPEP 2106.05(d) and 2106.05(f).
and/or the deviation is determined by a third machine learning system using the feature representation and the intermediate signal as input of the third machine learning system recites mere instructions to apply the exception with a generic machine learning system, which does not integrate the exception into a practical application, MPEP 2106.05(d) and 2106.05(f).
Therefore, claim 4 is found to be ineligible subject matter under 35 U.S.C. 101.
Regarding claim 5,
Claim 5 adds the additional limitation to claim 4:
wherein the first machine learning system, the second machine learning system, and the third machine learning system are parts of a fourth machine learning system recites mere additional details on the machine learning systems used to perform the abstract ideas, without changing that performing the judicial exceptions with the generic machine learning systems constitutes mere instructions to apply the exceptions, MPEP 2106.05(d) and 2106.05(f).
Therefore, claim 5 is found to be ineligible subject matter under 35 U.S.C. 101.
Regarding claim 6,
Claim 6 adds the additional limitations to claim 4:
wherein the second machine learning system is a neural network recites mere additional details on one of the machine learning systems used to perform the abstract ideas, without changing that performing the judicial exceptions with generic machine learning systems constitutes mere instructions to apply the exceptions, MPEP 2106.05(d) and 2106.05(f).
and an output of a hidden layer of the second machine learning system is used as additional input to the third machine learning system for determining the deviation recites an evaluation of the output of a hidden layer, which is a mental process, which is an abstract idea, regardless of if it’s performed on a generic computer.
Therefore, claim 6 is found to be ineligible subject matter under 35 U.S.C. 101.
Regarding claim 7,
Step 1 - “Is the claim to a process, machine, manufacture or composition of matter?”
Yes, the claim is directed towards a process.
Step 2A, Prong 1 - “Is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?”:
The limitation of training (i) the second machine learning model, or (ii) the first machine learning model and the second machine learning model, to determine a desired output signal for a provided first training input signal; recites an evaluation of a desired output signal, which is a mental process, which is an abstract idea, regardless of if it’s performed on a generic computer or using generic machine learning models.
The limitation of after training the second machine learning model or after training the first machine learning model and the second machine learning model, training the third machine learning model to determine a deviation of an intermediate signal determined from the first machine learning model and the second machine learning model for a provided second training input signal to a desired output signal for the provided second training input signal recites an evaluation of a deviation from a desired output signal, which is a mental process, which is an abstract idea, regardless of if it’s performed on a generic computer or using generic machine learning models.
Step 2A, Prong 2 - “Does the claim recite additional elements that integrate the judicial exception into a practical application?”:
The limitation of wherein the first machine learning model and the second machine learning model are not trained during the training of the third machine learning system recites mere addition details on the generic machine learning models, without changing that the generic machine learning models are used to apply the judicial exceptions, MPEP 2106.05(d) and 2106.05(f).
Step 2B - “Does the claim recite additional elements that amount to significantly more than the judicial exception?”:
The limitation of wherein the first machine learning model and the second machine learning model are not trained during the training of the third machine learning system recites mere addition details on the generic machine learning models, without changing that the generic machine learning models are used to apply the judicial exceptions, MPEP 2106.05(f).
Therefore, claim 7 is found to be ineligible subject matter under 35 U.S.C. 101.
Regarding claim 8,
Claim 8 adds the additional limitation to claim 1:
wherein the input signal includes a sensor signal recites mere additional details on the input signal, without changing that the limitation a. determining a feature representation characterizing the input signal;, recited in claim 1, recites an evaluation of the input signal, which is a mental process, which is an abstract idea, regardless of if it’s performed on a generic computer.
Therefore, claim 8 is found to be ineligible subject matter under 35 U.S.C. 101.
Regarding claim 9,
Claim 9 adds the additional limitation to claim 1:
wherein a robot is controlled based on the output signal recites a mere indication of a field of use for the judicial exceptions, MPEP 2106.05(d) and 2106.05(h).
Therefore, claim 9 is found to be ineligible subject matter under 35 U.S.C. 101.
Regarding claim 10,
Step 1 - “Is the claim to a process, machine, manufacture or composition of matter?”
No, the claim is directed towards software per se.
Step 2A, Prong 1 - “Is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?”:
The limitation of wherein the first machine system is configured to determine a feature representation characterizing an input signal, recites an evaluation of the input signal, which is a mental process, which is an abstract idea, regardless of if it’s performed on a generic computer.
The limitation of the second machine learning system is configured to determine an intermediate signal characterizing a classification and/or regression result of the feature representation, recites an evaluation of the feature representation, which is a mental process, which is an abstract idea, regardless of if it’s performed on a generic computer.
The limitation of and the third machine learning system is configured to determine, based on the feature representation and the intermediate signal, a deviation of the intermediate signal from a desired output signal of the input signal, recites an evaluation of a deviation from a desired output, which is a mental process, which is an abstract idea, regardless of if it’s performed on a generic computer.
The limitation of and wherein the fourth machine learning system is configured to adapt the intermediate signal according to the determined deviation thereby determining an adapted signal, recites, in view of the specification (Pg. 13, lines 20-24) “In the depicted embodiment, a gradient (g) of the deviation with respect to the intermediate signal (i) is determined and the adapted signal is then determined based a gradient descent of the deviation (d)”, a mathematical calculation, which is a mathematical concept, which is an abstract idea.
The limitation of and an output of a hidden layer of the second machine learning system is used as additional input to the third machine learning system for determining the deviation recites an evaluation of the output of a hidden layer, which is a mental process, which is an abstract idea, regardless of if it’s performed on a generic computer.
Step 2A, Prong 2 - “Does the claim recite additional elements that integrate the judicial exception into a practical application?”:
The limitation of a fourth machine learning system including a first machine learning system, a second machine learning system, and a third machine learning system, recites mere additional details on the machine learning systems used to perform the abstract ideas, without changing that performing the judicial exceptions with the generic machine learning systems constitutes mere instructions to apply the exceptions, MPEP 2106.05(d) and 2106.05(f).
The limitation of and provide the adapted signal as an output signal characterizing a classification result and/or a regression result of the input signal recites the mere extra-solution activity of data outputting, which does not integrate the exception into a practical application, MPEP 2106.05(d) and 2106.05(g).
The limitation of wherein the second machine learning system is a neural network recites mere additional details on one of the machine learning systems used to perform the abstract ideas, without changing that performing the judicial exceptions with generic machine learning systems constitutes mere instructions to apply the exceptions, MPEP 2106.05(d) and 2106.05(f).
Step 2B - “Does the claim recite additional elements that amount to significantly more than the judicial exception?”:
The limitation of a fourth machine learning system including a first machine learning system, a second machine learning system, and a third machine learning system, recites mere additional details on the machine learning systems used to perform the abstract ideas, without changing that performing the judicial exceptions with the generic machine learning systems constitutes mere instructions to apply the exceptions, MPEP 2106.05(f).
The limitation of and provide the adapted signal as an output signal characterizing a classification result and/or a regression result of the input signal recites transmitting data over a network, which is well-understood, routine, and conventional, MPEP 2106.05(d).II., example (i) of WURC computer functions.
The limitation of wherein the second machine learning system is a neural network recites mere additional details on one of the machine learning systems used to perform the abstract ideas, without changing that performing the judicial exceptions with generic machine learning systems constitutes mere instructions to apply the exceptions, MPEP 2106.05(f).
Therefore, claim 10 is found to be ineligible subject matter under 35 U.S.C. 101.
Regarding claim 11,
Claim 11 discloses a system implementing the method of claim 7, with substantially the same limitations. Therefore the same analysis and rejection applied to claim 7 applies to claim 11. Additionally, claim 11 is ineligible under 35 U.S.C. 101 for being directed to a computer system without hardware components, i.e. software per se.
Therefore, claim 11 is found to be ineligible subject matter under 35 U.S.C. 101.
Regarding claim 12,
Claim 12 discloses a non-transitory machine-readable medium, i.e. a manufacture, implementing the method of claim 1, with substantially the same limitations. Therefore the same analysis and rejection applied to claim 1 applies to claim 12.
Therefore, claim 12 is found to be ineligible subject matter under 35 U.S.C. 101.
Claim Rejections - 35 USC § 102
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claims 7 and 11 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Denil et al. (U.S. Patent Application Publication No. 2019/0220748), hereinafter Denil.
Regarding claim 7,
Denil teaches A computer-implemented method for training a fourth machine learning system including a first machine learning model, a second machine learning model, and a third machine learning model, the method comprising the following steps: (Denil Fig. 1A shows the machine learning system that is trained, (Denil [0060]) “In some implementations a learned update rule, as given by equation (2) above, may be applied to other machine learning models that are configured to perform similar machine learning tasks, e.g., machine learning tasks with a similar structure. For example, the learned update rule may be applied to a second machine learning model”, the second machine learning model, the Machine learning model 102, and the Recurrent Neural Network 106 constitute first, second, and third machine learning models)
PNG
media_image2.png
690
561
media_image2.png
Greyscale
training (i) the second machine learning model, or (ii) the first machine learning model and the second machine learning model, to determine a desired output signal for a provided first training input signal; ((Denil [0034]) “The machine learning model 102 can be trained to perform the machine learning task using gradient descent techniques to optimize a machine learning model objective function. For example, in cases where the machine learning model 102 is a neural network, the machine learning model may be trained to perform a respective machine learning task using backpropagation of errors. During a backpropagation training process, training inputs are processed by the neural network to generate respective neural network outputs. The outputs are then compared to a desired or known output using an objective function, e.g., a loss function, and error values are determined”, machine learning model 102 is a second machine learning model)
after training the second machine learning model or after training the first machine learning model and the second machine learning model, ((Denil [0036]) “To determine the above learned update rule for time t+1, the training module 104 is configured to compute or obtain a gradient of the machine learning model objective function at time t with respect to the machine learning model parameters at time t”, determining an update rule for time t+1 based on parameters at time t means that the machine learning model has been trained at time t) training the third machine learning model to determine a deviation ((Denil [0051]) “Determining the update rule for the machine learning model parameters using the RNN includes training the RNN to determine RNN parameters that minimize the RNN objective function, and using trained RNN parameters to determine a final update rule that is used to generate the trained machine learning model”, (Denil [0034]) “The outputs are then compared to a desired or known output using an objective function, e.g., a loss function, and error values are determined. The error values are used to calculate a gradient of the objective function with respect to the neural network parameters. The gradient is then used as input to an update rule to determine an update for the neural network parameters that minimizes the objective function”, a gradient calculated from error values is a deviation, an RNN is a third machine learning model) of an intermediate signal determined from the first machine learning model and the second machine learning model ((Denil [0034]) “The machine learning model 102 can be trained to perform the machine learning task using gradient descent techniques to optimize a machine learning model objective function. For example, in cases where the machine learning model 102 is a neural network, the machine learning model may be trained to perform a respective machine learning task using backpropagation of errors. During a backpropagation training process, training inputs are processed by the neural network to generate respective neural network outputs”, neural network outputs are an intermediate signal) for a provided second training input signal to a desired output signal for the provided second training input signal, ((Denil [0034]) “During a backpropagation training process, training inputs are processed by the neural network to generate respective neural network outputs. The outputs are then compared to a desired or known output using an objective function, e.g., a loss function, and error values are determined”, training inputs include at least two training input signals)
wherein the first machine learning model and the second machine learning model are not trained during the training of the third machine learning system ((Denil [0036]) “The training module 104 is configured to determine the learned update rule for time t+1 in equation (2) above using the values of the RNN parameters ϕ for time t and gradients of respective machine learning model objective functions f. The learned update rule 110 may then be applied to the machine learning model parameters to update the machine learning model 102”, determining the learned update rule at time t+1 for the machine learning model 102 means that the machine learning model is not updated, i.e. trained, until the update rule is learned)
Regarding claim 11,
Claim 11 recites a system for performing the function of the method of claim 7. All other limitations in claim 11 are substantially the same as those in claim 7, therefore the same rationale for rejection applies.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-6 are rejected under 35 U.S.C. 103 as being unpatentable over Gurney “An Introduction to Neural Networks”, hereinafter Gurney, in view of Denil, further in view of Baker (U.S. Patent Application Publication No. 2021/0342683), hereinafter Baker.
Regarding claim 1,
Gurney teaches A computer-implemented method for determining an output signal characterizing a classification and/or a regression result of an input signal comprising the following steps: ((Gurney Pgs. 2-3) “One type of network is shown in Figure 1.3. Each node is now shown by only a circle but weights are implicit on all connections. The nodes are arranged in a layered structure in which each signal emanates from an input and passes via two nodes before reaching an output beyond which it is no longer transformed. This feedforward structure is only one of several available and is typically used to place an input pattern into one of several classes according to the resulting pattern of outputs. For example, if the input consists of an encoding of the patterns of light and dark in an image of handwritten letters, the output layer (topmost in the figure) may contain 26 nodes—one for each letter of the alphabet—to flag which letter class the input character is from”)
a. determining a feature representation characterizing the input signal; ((Gurney Pg. 31) “We will make this concrete by way of an example which assumes a network that is being used to classify visual images. The sequence of events for making a single training pattern in this case is shown in Figure 4.10”, Gurney Fig. 4.10 shows a feature representation in (d) characterizing an input image in (a))
PNG
media_image3.png
337
676
media_image3.png
Greyscale
b. determining an intermediate signal characterizing a classification and/or regression result of the feature representation; ((Gurney Pg. 43) “One iteration of the for loop results in a single presentation of each pattern in the training set and is sometimes referred to as an epoch. What constitutes an ‘acceptably low’ error is treated in Section 6.4 but for the moment we expand the main step of ‘train on that pattern’ into the following steps. 1. Present the pattern at the input layer. 2. Let the hidden units evaluate their output using the pattern. 3. Let the output units evaluate their output using the result in step 2 from the hidden units”, an output from the hidden units for an input pattern corresponds to an intermediate signal characterizing a classification result of a feature representation)
c. predicting, based on the feature representation and the intermediate signal, a deviation of the intermediate signal from a desired output signal of the input signal; ((Gurney Pg. 43) “we expand the main step of ‘train on that pattern’ into the following steps. 1. Present the pattern at the input layer. 2. Let the hidden units evaluate their output using the pattern. 3. Let the output units evaluate their output using the result in step 2 from the hidden units. 4. Apply the target pattern to the output layer. 5. Calculate the δs on the output nodes. 6. Train each output node using gradient descent (6.4). 7. For each hidden node, calculate its δ according to (6.6). 8. For each hidden node, use the δ found in step 7 to train according to gradient descent (6.2)”, (Gurney Pg. 42) “The contribution that node j makes towards the error is, of course, expressed in the ‘δ’ for that node δj”, an error based on contributions of output nodes produced from a training pattern input corresponds to a deviation based on a feature representation, the error being based on contributions of output nodes which use the output of hidden units corresponds to a deviation being based on an intermediate signal, each node contribution to the error δ being based on a difference between target prediction and actual prediction to predicting a deviation from the desired output)
d. adapting the intermediate signal according to the determined deviation thereby determining an adapted signal; ((Gurney Pg. 43) “we expand the main step of “train on that pattern” into the following steps…6. Train each output node using gradient descent (6.4). 7. For each hidden node, calculate its δ according to (6.6). 8. For each hidden node, use the δ found in step 7 to train according to gradient descent (6.2)”, training based on a determined deviation corresponds to adapting an intermediate signal according to the determined deviation)
and e. providing the adapted signal as the output signal, ((Gurney Pgs. 42-43) “This basic algorithm loop structure is the same as for the perception rule or delta rule initialize weights repeat for each training pattern train on that pattern end for loop until the error is acceptably low”, (Gurney Pg. 3) “When the required weight updates have been made another pattern is presented, the output compared with the target, and new changes made. This sequence of events is repeated iteratively many times until (hopefully) the network’s behaviour converges so that its response to each pattern is close to the corresponding target”, the output of the network after the error has become acceptably low corresponds to an adapted signal that is provided as an output signal)
Denil teaches the following further limitation that Gurney does not teach:
wherein: the feature representation is determined by a first machine learning system using the input signal as input of the first machine learning system;
and/or the intermediate signal is determined by a second machine learning system using the feature representation as input of the second machine learning system;
and/or the deviation is determined by a third machine learning system ((Denil [0051]) “Determining the update rule for the machine learning model parameters using the RNN includes training the RNN to determine RNN parameters that minimize the RNN objective function, and using trained RNN parameters to determine a final update rule that is used to generate the trained machine learning model”, (Denil [0034]) “The outputs are then compared to a desired or known output using an objective function, e.g., a loss function, and error values are determined. The error values are used to calculate a gradient of the objective function with respect to the neural network parameters. The gradient is then used as input to an update rule to determine an update for the neural network parameters that minimizes the objective function”, a gradient calculated from error values is a deviation, an RNN is a third machine learning model) using the feature representation and the intermediate signal as input of the third machine learning system, ((Denil [0034]) “The machine learning model 102 can be trained to perform the machine learning task using gradient descent techniques to optimize a machine learning model objective function. For example, in cases where the machine learning model 102 is a neural network, the machine learning model may be trained to perform a respective machine learning task using backpropagation of errors. During a backpropagation training process, training inputs are processed by the neural network to generate respective neural network outputs. The outputs are then compared to a desired or known output using an objective function, e.g., a loss function, and error values are determined”, neural network outputs are an intermediate signal, and training inputs correspond to a feature representation)
At the time of filing, one of ordinary skill in the art would have motivation to combine Gurney and Denil by taking the method for determining an output signal for classification or regression, including predicting a deviation from an intermediate signal from a desired output signal, taught by Gurney, and having a machine learning system determine the deviation, taught by Denil, as Denil teaches: (Denil [0023]) “machine learning models that have been trained using a recurrent neural network may perform respective machine learning tasks more accurately and efficiently”. Such a combination would be obvious.
Baker teaches the following further limitation that neither Gurney nor Denil teaches:
wherein the second machine learning system is a neural network and an output of a hidden layer of the second machine learning system is used as additional input to the third machine learning system ((Baker [0016]) “If the first machine learning system 202 is a neural network, then activation values of inner layer nodes of first machine learning system 202 may also be directly provided as input to the second machine learning system 206”, activation values of inner layer nodes correspond to outputs of a hidden layer) for determining the deviation ((Baker [0020] “the computer system 300 may compute the partial derivative of the primary objective 203 with respect to any of the following: (i) the activation value of a node”) (Baker [0023]) “With these partial derivatives, the computer system 300 may compute in the second machine learning system 206 an estimate of the likelihood of the first machine learning system 202 making an error due to a change in the input 201 or a change in one of the elements in the first machine learning system 202”, determining the likelihood of an error due to a change in the input or in an element based at least in part on a derivative of an output of a hidden layer corresponds to determining a deviation based on an additional input of an output of a hidden layer of a machine learning system)
At the time of filing, one of ordinary skill in the art would have motivation to combine Gurney, Denil, and Baker by taking the method for determining an output signal for classification or regression, including predicting a deviation from an intermediate signal from a desired output signal, the deviation being determined by another machine learning system, taught by Gurney and Denil, and adding the use of hidden layer outputs from another machine learning model to predict a deviation from a correct output, taught by Baker, as additional information from the intermediate states of a neural network can help pinpoint the origin of an error. Such a combination would be obvious.
Regarding claim 2,
Gurney, Denil, and Baker jointly teach The method according to claim 1,
Gurney further teaches:
wherein the steps c. and d. are repeated iteratively until an exit criterion is fulfilled, ((Gurney Pgs. 42-43) “This basic algorithm loop structure is the same as for the perception rule or delta rule initialize weights repeat for each training pattern train on that pattern end for loop until the error is acceptably low”, the training step including steps c. and d. loop until the exit criterion of an acceptably low error is fulfilled)
and wherein the adapted signal is used as intermediate signal in a next iteration ((Gurney Pgs. 42-43) “This basic algorithm loop structure is the same as for the perception rule or delta rule…repeat for each training pattern train on that pattern end for loop”, (Gurney Pg. 43) “we expand the main step of “train on that pattern” into the following steps. 1. Present the pattern at the input layer. 2. Let the hidden units evaluate their output using the pattern. 3. Let the output units evaluate their output using the result in step 2 from the hidden units”, each iteration of the loop uses the signal adapted based on the error as the signal for the next iteration)
At the time of filing, one of ordinary skill in the art would have motivation to combine the method jointly taught by Gurney, Denil, and Baker for the parent claim of claim 2, claim 1. No new embodiments are introduced, so the reason to combine is the same as for the parent claim.
Regarding claim 3,
Gurney, Denil, and Baker jointly teach The method according to claim 1,
Gurney further teaches:
wherein the deviation is predicted by a differentiable model and the intermediate signal is adapted based on a gradient of the deviation with respect to the intermediate signal ((Gurney Pg. 35) “By way of terminology dy/dx is also known as the differential or derivative of y with respect to x…This technique is called, not surprisingly, gradient descent and its effectiveness hinges, of course, on the ability to calculate, or make estimates of, the quantities like dy/dx”, (Gurney Pg. 36) “We now apply gradient descent to the minimization of a network error function”, using gradient descent which uses a differentiable function to minimize a network error function corresponds to adapting an intermediate signal based on a gradient of deviation predicted by a differentiable model)
At the time of filing, one of ordinary skill in the art would have motivation to combine the method jointly taught by Gurney, Denil, and Baker for the parent claim of claim 3, claim 1. No new embodiments are introduced, so the reason to combine is the same as for the parent claim.
Regarding claim 4,
Gurney, Denil, and Baker jointly teach The method according to claim 1, wherein:
Denil further teaches:
the feature representation is determined by a first machine learning system using the input signal as input of the first machine learning system;
and/or the intermediate signal is determined by a second machine learning system using the feature representation as input of the second machine learning system;
and/or the deviation is determined by a third machine learning system ((Denil [0051]) “Determining the update rule for the machine learning model parameters using the RNN includes training the RNN to determine RNN parameters that minimize the RNN objective function, and using trained RNN parameters to determine a final update rule that is used to generate the trained machine learning model”, (Denil [0034]) “The outputs are then compared to a desired or known output using an objective function, e.g., a loss function, and error values are determined. The error values are used to calculate a gradient of the objective function with respect to the neural network parameters. The gradient is then used as input to an update rule to determine an update for the neural network parameters that minimizes the objective function”, a gradient calculated from error values is a deviation, an RNN is a third machine learning model) using the feature representation and the intermediate signal as input of the third machine learning system ((Denil [0034]) “The machine learning model 102 can be trained to perform the machine learning task using gradient descent techniques to optimize a machine learning model objective function. For example, in cases where the machine learning model 102 is a neural network, the machine learning model may be trained to perform a respective machine learning task using backpropagation of errors. During a backpropagation training process, training inputs are processed by the neural network to generate respective neural network outputs. The outputs are then compared to a desired or known output using an objective function, e.g., a loss function, and error values are determined”, neural network outputs are an intermediate signal, and training inputs correspond to a feature representation)
At the time of filing, one of ordinary skill in the art would have motivation to combine the method jointly taught by Gurney, Denil, and Baker for the parent claim of claim 4, claim 1. No new embodiments are introduced, so the reason to combine is the same as for the parent claim.
Regarding claim 5,
Gurney, Denil, and Baker jointly teach The method according to claim 4,
Denil further teaches:
wherein the first machine learning system, the second machine learning system, and the third machine learning system are parts of a fourth machine learning system (Denil Fig. 1A shows the machine learning system corresponding to the fourth machine learning system, including a machine learning model 102 corresponding to a second machine learning system and a recurrent neural network 106 that corresponds to a third machine learning system, (Denil [0060]) “In some implementations a learned update rule, as given by equation (2) above, may be applied to other machine learning models that are configured to perform similar machine learning tasks, e.g., machine learning tasks with a similar structure”, the machine learning models that a learned update rule is applied to could include a first machine learning model)
At the time of filing, one of ordinary skill in the art would have motivation to combine the method jointly taught by Gurney, Denil, and Baker for the parent claim of claim 5, claim 4. No new embodiments are introduced, so the reason to combine is the same as for the parent claim.
Regarding claim 6,
Gurney, Denil, and Baker jointly teach The method according to claim 4,
Baker further teaches:
wherein the second machine learning system is a neural network and an output of a hidden layer of the second machine learning system is used as additional input to the third machine learning system ((Baker [0016]) “If the first machine learning system 202 is a neural network, then activation values of inner layer nodes of first machine learning system 202 may also be directly provided as input to the second machine learning system 206”, activation values of inner layer nodes correspond to outputs of a hidden layer) for determining the deviation ((Baker [0020] “the computer system 300 may compute the partial derivative of the primary objective 203 with respect to any of the following: (i) the activation value of a node”) (Baker [0023]) “With these partial derivatives, the computer system 300 may compute in the second machine learning system 206 an estimate of the likelihood of the first machine learning system 202 making an error due to a change in the input 201 or a change in one of the elements in the first machine learning system 202”, determining the likelihood of an error due to a change in the input or in an element based at least in part on a derivative of an output of a hidden layer corresponds to determining a deviation based on an additional input of an output of a hidden layer of a machine learning system)
At the time of filing, one of ordinary skill in the art would have motivation to combine the method jointly taught by Gurney, Denil, and Baker for the parent claim of claim 6, claim 4. No new embodiments are introduced, so the reason to combine is the same as for the parent claim.
Regarding claim 12,
Claim 12 recites a non-transitory computer-readable medium storing a program for performing the function of the method of claim 1. Specifically, claim 12 recites A non-transitory machine-readable storage medium on which is stored a computer program for determining an output signal characterizing a classification and/or a regression result of an input signal, the computer program, when executed by a processor, causing the processor to perform the following steps: [The method of claim 1]. Gurney teaches (Gurney Pg. 32) “When we build a neural network do we go to our local electronic hardware store, buy components and then assemble them? The answer, in most cases, is ‘no’. Usually we simulate the network on a conventional computer such as a PC or workstation”, with simulation on a conventional computer including a program stored on the computer’s non-transitory machine-readable memory. All other limitations in claim 12 are substantially the same as those in claim 1, therefore the same rationale for rejection applies.
Claims 8 and 9 are rejected under 35 U.S.C. 103 as being unpatentable over Gurney in view of Denil, further in view of Baker, further in view of Kozuka (U.S. Patent Application Publication No. 2021/0016439), hereinafter Kozuka.
Regarding claim 8,
Gurney, Denil, and Baker jointly teach The method according to claim 1,
Kozuka teaches the following further limitation that neither Gurney, nor Denil, nor Baker teaches:
wherein the input signal includes a sensor signal ((Kozuka [0029]) “The gripping force sensor 15 detects a gripping force of the hand 13 when the hand 13 grips the work W, and then transmits a gripping force signal indicating the gripping force to the robot control device 3”, (Kozuka [0037]) “The storage 353 stores therein a learning model MD1 trained. The learning model MD1 is generated by machine learning. Captured image data PD is input to the learning model MD1, and thereby the learning model MD1 outputs gripping force data GD”)
At the time of filing, one of ordinary skill in the art would have motivation to combine Gurney, Denil, Baker, and Kozuka by taking the method of claim 1, including determination of an output signal for classification or regression of an input signal, jointly taught by Gurney, Denil, and Baker, and having the input signal include a sensor signal, taught by Kozuka, as using sensor input for a machine learning model allows for integration of machine learning into variety of practical systems such as smart homes or security systems to improve them. Such a combination would be obvious.
Regarding claim 9,
Gurney, Denil, and Baker jointly teach The method according to claim 1,
Kozuka teaches the following further limitation that neither Gurney, nor Denil, nor Baker teaches:
wherein a robot is controlled based on the output signal ((Kozuka [0046]) “when the robotic device 1 grips the works W to convey the works W to a predetermined position one at a time, the imaging section 37 captures an image of a work W to be gripped by the hand 13. The gripping controller 51 then inputs the captured image data PD generated by the imaging section 37 to the learning model MD1. The learning model MD1 then outputs the gripping force data GD. The gripping controller 51 then controls the hand 13 based on the gripping force data GD. The robotic device 1 can therefore grip the work W with an appropriate force and place the work W at the predetermined position”, a robotic device including a hand that is controlled based on data output by a learning model corresponds to a robot controlled based on an output signal)
At the time of filing, one of ordinary skill in the art would have motivation to combine Gurney, Denil, Baker, and Kozuka by taking the method of claim 1, including determination of an output signal for classification or regression of an input signal, jointly taught by Gurney, Denil, and Baker, and having the output signal be used to control a robot, taught by Kozuka, as machine learning models are cheaper to develop and more adaptable than former means used for controlling robots. Such a combination would be obvious.
Claim 10 is rejected under 35 U.S.C. 103 as being unpatentable over Denil in view of Gurney, further in view of Baker.
Regarding claim 10,
Denil teaches A machine learning system comprising:
a fourth machine learning system including a first machine learning system, a second machine learning system, and a third machine learning system, (Denil Fig. 1A shows the machine learning system corresponding to the fourth machine learning system, including a machine learning model 102 corresponding to a second machine learning system and a recurrent neural network 106 that corresponds to a third machine learning system, (Denil [0060]) “In some implementations a learned update rule, as given by equation (2) above, may be applied to other machine learning models that are configured to perform similar machine learning tasks, e.g., machine learning tasks with a similar structure. For example, the learned update rule may be applied to a second machine learning model”, the second machine learning model corresponds to a first machine learning system)
the second machine learning system is configured to determine an intermediate signal characterizing a classification and/or regression result of the feature representation, ((Denil [0034]) “The machine learning model 102 can be trained to perform the machine learning task using gradient descent techniques to optimize a machine learning model objective function. For example, in cases where the machine learning model 102 is a neural network, the machine learning model may be trained to perform a respective machine learning task using backpropagation of errors. During a backpropagation training process, training inputs are processed by the neural network to generate respective neural network outputs”, neural network outputs correspond to an intermediate signal characterizing classification or regression results)
and the third machine learning system is configured to determine, based on the feature representation and the intermediate signal, ((Denil [0034]) “The machine learning model 102 can be trained to perform the machine learning task using gradient descent techniques to optimize a machine learning model objective function. For example, in cases where the machine learning model 102 is a neural network, the machine learning model may be trained to perform a respective machine learning task using backpropagation of errors. During a backpropagation training process, training inputs are processed by the neural network to generate respective neural network outputs”, neural network outputs are an intermediate signal) a deviation of the intermediate signal from a desired output signal of the input signal, ((Denil [0051]) “Determining the update rule for the machine learning model parameters using the RNN includes training the RNN to determine RNN parameters that minimize the RNN objective function, and using trained RNN parameters to determine a final update rule that is used to generate the trained machine learning model”, (Denil [0034]) “The outputs are then compared to a desired or known output using an objective function, e.g., a loss function, and error values are determined. The error values are used to calculate a gradient of the objective function with respect to the neural network parameters. The gradient is then used as input to an update rule to determine an update for the neural network parameters that minimizes the objective function”, a gradient calculated from error values is a deviation, an RNN is a third machine learning model)
and wherein the fourth machine learning system is configured to adapt the intermediate signal according to the determined deviation thereby determining an adapted signal, ((Denil [0035]) “The training module 104 is configured to train the machine learning model 102 by determining a learned parameter update rule for the machine learning model parameters using the RNN 106. The learned parameter update rule for the machine learning model parameters can be implemented over a sequence of time steps t=1, ..., T to adjust the values of the machine learning model parameters from initial or current values, e.g., at time t=1, to trained values, e.g., at time t=T”, the fourth machine learning system includes a training module, adjusting the values of machine learning model parameters based on deviation associated with an intermediate signal corresponds to adapting the intermediate signal to determine an adapted signal)
Gurney teaches the following further limitations that Denil does not teach:
wherein the first machine system is configured to determine a feature representation characterizing an input signal, ((Gurney Pg. 31) “We will make this concrete by way of an example which assumes a network that is being used to classify visual images. The sequence of events for making a single training pattern in this case is shown in Figure 4.10”, Gurney Fig. 4.10 shows a feature representation in (d) characterizing an input image in (a))
and provide the adapted signal as an output signal characterizing a classification result and/or a regression result of the input signal, ((Gurney Pg. 3) “In one training paradigm called supervised learning, used in conjunction with nets of the type shown in Figure 1.3, an input pattern is presented to the net and its response then compared with a target output. In terms of our previous letter recognition example, an “A”, say, may be input and the network output compared with the classification code for A. The difference between the two patterns of output then determines how the weights are altered. Each particular recipe for change constitutes a learning rule, details of which form a substantial part of subsequent chapters. When the required weight updates have been made another pattern is presented, the output compared with the target, and new changes made. This sequence of events is repeated iteratively many times until (hopefully) the network’s behaviour converges so that its response to each pattern is close to the corresponding target”)
At the time of filing, one of ordinary skill in the art would have motivation to combine Gurney and Denil by taking the method for determining a deviation of an intermediate signal from a desired signal and adapting the signal, taught by Denil, and having a first machine learning system determine a feature representation for the signal, as well as using the adapted signal as an output signal characterizing a classification result, taught by Gurney, as both creating feature representations and modifying signals to create more accurate output signals for classification are very well-known within the art, providing the predictable benefit of more accurate machine learning systems. Such a combination would be obvious.
Baker teaches the following further limitation that neither Denil nor Gurney teaches:
wherein the second machine learning system is a neural network and an output of a hidden layer of the second machine learning system is used as additional input to the third machine learning system ((Baker [0016]) “If the first machine learning system 202 is a neural network, then activation values of inner layer nodes of first machine learning system 202 may also be directly provided as input to the second machine learning system 206”, activation values of inner layer nodes correspond to outputs of a hidden layer) for determining the deviation ((Baker [0020] “the computer system 300 may compute the partial derivative of the primary objective 203 with respect to any of the following: (i) the activation value of a node”) (Baker [0023]) “With these partial derivatives, the computer system 300 may compute in the second machine learning system 206 an estimate of the likelihood of the first machine learning system 202 making an error due to a change in the input 201 or a change in one of the elements in the first machine learning system 202”, determining the likelihood of an error due to a change in the input or in an element based at least in part on a derivative of an output of a hidden layer corresponds to determining a deviation based on an additional input of an output of a hidden layer of a machine learning system)
At the time of filing, one of ordinary skill in the art would have motivation to combine Denil, Gurney, and Baker by taking the method for determining a deviation of an intermediate signal from a desired signal and adapting the signal, including a first machine learning system determine a feature representation for the signal and using the adapted signal as an output signal characterizing a classification result, taught by Denil and Gurney, and adding the use of hidden layer outputs from another machine learning model to predict a deviation from a correct output, taught by Baker, as additional information from the intermediate states of a neural network can help pinpoint the origin of an error. Such a combination would be obvious.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Rickard et al. (U.S. Patent No. 11,816,555) teaches systems, media and methods for chaining together multiple discrete machine learning models to model real-world situations.
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to VICTOR A NAULT whose telephone number is (703) 756-5745. The examiner can normally be reached M - F, 12 - 8.
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, Miranda Huang can be reached at (571) 270-7092. 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.
/V.A.N./Examiner, Art Unit 2124
/Kevin W Figueroa/Primary Examiner, Art Unit 2124
Prosecution Insights