DETAILED ACTION
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 .
This action is responsive to the Application filed on January 23, 2024. Claims 1-8 are pending in the case. Claims 1 and 6 are the independent claims.
This action is non-final.
Claim Rejections – 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-8 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea (mental steps) without significantly more. This judicial exception is not integrated into a practical application because any additional elements amount to implementing the abstract idea on a generic computer. The claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Regarding independent claims 1 and 6, and relying on the evaluation flowchart in MPEP 2106:
Step 1 (Is the claim to a process, machine, manufacture, or composition of matter?): Yes. Claim 1 is a system (machine). Claim 6 is a method (process).
Step 2a Prong One (Does the claim recite an abstract idea?): Yes. Claims 1 and 6 recite:
(in claim 1) calculates a score sir when local data di is applied to each of A neural networks r (a mental process, including a mental process involving a mathematical calculation, including using physical aids such as pen and paper; Examiner notes that the claim does not appear to recite any particular way in which the score is calculated or even that the recited score is the result of applying data to the neural networks; instead the claim only appears to require that the score be calculated at a time when local data is applied to neural networks; such a score could be determined based on mental determination),
(in claim 6) a score calculating step of calculating a score sire when local data di is applied to each of A neural networks r (a mental process, including a mental process involving a mathematical calculation, including using physical aids such as pen and paper; Examiner notes that the claim does not appear to recite any particular way in which the score is calculated or even that the recited score is the result of applying data to the neural networks; instead the claim only appears to require that the score be calculated at a time when local data is applied to neural networks; such a score could be determined based on mental determination);
(in claim 1) selects an optimal neural network (a mental process of determination, such as a human mentally determining a neural network which is optimal),
(in claim 6) selecting an optimal neural network (a mental process of determination, such as a human mentally determining a neural network which is optimal).
Under the broadest reasonable interpretation, these steps may be performed mentally, using mental observation and mental determination, including by a human using a physical aid such as pen and paper, including a human mentally performing observations and mentally performing mathematical calculations, and therefore correspond to the Mental Processes grouping.
Step 2a Prong Two (Does the claim recite additional elements that integrate the judicial exception into a practical application?): No. Claims 1 and 6 additionally recite:
(in claim 1) a learning system comprising: a learning server apparatus; and n processing apparatuses i, wherein, when i = 1, 2,..., n, and r = 0, 1,..., A-1, the processing apparatuses i each include: second processing circuitry configured to…execute (described steps)…the learning server apparatus includes: first processing circuitry configured to….execute (described steps)… (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f));
(in claim 6) using a learning server apparatus that includes first processing circuitry and n processing apparatuses i that include second processing circuitry, the learning method comprising: when i = 1, 2,..., n, and r = 0, 1,..., A-1… (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f));
(in claim 1) execute a second federated learning processing (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f); i.e. the claim recites generic federated learning processing); and
(in claim 1) execute a score calculation processing in which the second processing circuitry calculates (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f))
(in claim 6) calculating, by the second processing circuitry of the processing apparatus I (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f))
(in claim 1) execute a first federated learning processing (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f));
(in claim 1) execute an aggregation processing in which the first processing circuitry aggregates A neural networks using A x n scores sir, and selects (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f); i.e. the claim recites a generic machine learning aggregation operation)
(in claim 6) an aggregation step of aggregating, by the first processing circuitry of the learning server apparatus, A neural networks using A x n scores sir and selecting (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f); i.e. the claim recites a generic machine learning aggregation operation)
(in claim 1) in the first federated learning processing and the second federated learning processes, the first processing circuitry and the second processing circuitries of the n processing apparatuses i cooperate to perform federated learning using the selected optimal neural network as a first global model (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f); i.e. the claim recites a generic federated learning operations using generic computer components)
(in claim 6) a federated learning step of cooperating, by the first processing circuitry of the learning server apparatus and the second processing circuitries of the n processing apparatuses i, to perform federated learning using the selected optimal neural network as a first global model (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f); i.e. the claim recites a generic federated learning operations using generic computer components)
(in claims 1 and 6) wherein the score sir includes an index with which a neural network having an excellent learning effect can be searched for (a field of use and technological environment as discussed in MPEP 2106.05(h))
Therefore, in view of the considerations set forth in MPEP 2106.04(d), 2106.05(a)-(c) and (e)-(h), the additional elements as disclosed above alone or in combination do not integrate the judicial exception into a practical application as they are mere insignificant extra solution activity, combined with implementing the abstract idea using generic computer components.
Step 2b (Does the claim recite additional elements that amount to siqnificantly more than the judicial exception): No. Relying on the same analysis as Step 2a Prong Two (see MPEP 2106.05.I.A: Limitations that the courts have found not to be enough to qualify as “significantly more” when recited in a claim with a judicial exception include:…Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, e.g., a limitation indicating that a particular function such as creating and maintaining electronic records is performed by a computer, as discussed in Alice Corp., 573 U.S. at 225-26, 110 USPQ2d at 1984 (see MPEP 2106.05(f));…Simply appending well-understood, routine, conventional activities previously known to the industry, specified at a high level of generality, to the judicial exception...; Adding insignificant extra-solution activity to the judicial exception, as discussed in MPEP 2106.05(g);…)), claims 1 and 6 do not recite any additional elements that amount to significantly more than the abstract idea. As discussed above, Claims 1 and 6 recite:
(in claim 1) a learning system comprising: a learning server apparatus; and n processing apparatuses i, wherein, when i = 1, 2,..., n, and r = 0, 1,..., A-1, the processing apparatuses i each include: second processing circuitry configured to…execute (described steps)…the learning server apparatus includes: first processing circuitry configured to….execute (described steps)… (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f));
(in claim 6) using a learning server apparatus that includes first processing circuitry and n processing apparatuses i that include second processing circuitry, the learning method comprising: when i = 1, 2,..., n, and r = 0, 1,..., A-1… (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f));
(in claim 1) execute a second federated learning processing (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f); i.e. the claim recites generic federated learning processing); and
(in claim 1) execute a score calculation processing in which the second processing circuitry calculates (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f))
(in claim 6) calculating, by the second processing circuitry of the processing apparatus I (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f))
(in claim 1) execute a first federated learning processing (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f));
(in claim 1) execute an aggregation processing in which the first processing circuitry aggregates A neural networks using A x n scores sir, and selects (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f); i.e. the claim recites a generic machine learning aggregation operation)
(in claim 6) an aggregation step of aggregating, by the first processing circuitry of the learning server apparatus, A neural networks using A x n scores sir and selecting (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f); i.e. the claim recites a generic machine learning aggregation operation)
(in claim 1) in the first federated learning processing and the second federated learning processes, the first processing circuitry and the second processing circuitries of the n processing apparatuses i cooperate to perform federated learning using the selected optimal neural network as a first global model (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f); i.e. the claim recites a generic federated learning operations using generic computer components)
(in claim 6) a federated learning step of cooperating, by the first processing circuitry of the learning server apparatus and the second processing circuitries of the n processing apparatuses i, to perform federated learning using the selected optimal neural network as a first global model (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f); i.e. the claim recites a generic federated learning operations using generic computer components)
(in claims 1 and 6) wherein the score sir includes an index with which a neural network having an excellent learning effect can be searched for (a field of use and technological environment as discussed in MPEP 2106.05(h)).
The additional elements as discussed above, in combination with the abstract idea, are not sufficient to amount to significantly more than the judicial exception as they are well, understood, routine and conventional activity as disclosed in combination with generic computer functions and components used to implement the abstract idea.
Regarding dependent claim 2:
Step 2a Prong One: incorporates the rejection of claim 1. The claim further recites
calculates a variation of the score sir for each neural network r (a mental process, including a mental process involving a mathematical calculation, including using physical aids such as pen and paper; Examiner notes that the claim does not appear to recite any particular way in which the variation of the score is calculated; such a score could be calculated based on mental determination),
calculates a score sr for each neural network in consideration of a number of pieces of data of the local data di in a case where the variation is larger than a predetermined threshold value (a mental process, including a mental process involving a mathematical calculation, including using physical aids such as pen and paper; Examiner notes that the claim does not appear to recite any particular way in which the scores are calculated, other than that there is a consideration of number of pieces of data when the variation is larger than a threshold; such a score could be calculated based on mental determination which includes the calculating of the score considering the number of pieces of data corresponding to a mental determination that the variation is larger than a threshold),
calculates the score sr for each neural network without considering the number of pieces of data of the local data di in a case where the variation is equal to or smaller than the predetermined threshold value (a mental process, including a mental process involving a mathematical calculation, including using physical aids such as pen and paper; Examiner notes that the claim does not appear to recite any particular way in which the scores are calculated, other than that there is no consideration of number of pieces of data when the variation is smaller than a threshold; such a score could be calculated based on mental determination which includes the calculating of the score without considering the number of pieces of data corresponding to a mental determination that the variation is smaller than a threshold).
Step 2a Prong Two: the claims additionally recite
wherein the score sr is a correlation score of a weight when the local data di is applied to the neural network r (a field of use and technological environment as discussed in MPEP 2106.05(h)));
the aggregation processing in which the first processing circuitry calculates… (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Step 2b: the claims additionally recite
wherein the score sr is a correlation score of a weight when the local data di is applied to the neural network r (a field of use and technological environment as discussed in MPEP 2106.05(h)));
the aggregation processing in which the first processing circuitry calculates… (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Regarding dependent claim 3:
Step 2a Prong One: incorporates the rejection of claim 1; the claims further recite
selects Q optimal neural network possibilities in a case where an optimal neural network cannot be selected (a mental process of determining a set of optimal neural network possibilities)
compares accuracies of the Q optimal neural network possibilities after the federated learning, and selects an optimal neural network with the highest accuracy as the optimal neural network (a mental process of evaluation, such as a human mentally comparing accuracies of neural network possibilities, and a mental process of evaluation, such as a human mentally determining/selecting the neural network with the highest accuracy).
Step 2a Prong Two: the claim additionally recites
in the aggregation processing the first processing circuitry selects (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f))
the first processing circuitry divides the n processing apparatuses I into Q groups, performs federated learning in cooperation with a second processing circuitry of a process apparatus belonging to each group using the Q optimal neural network possibilities as the first global model (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f); i.e. the claim recites generic actions of dividing apparatuses into groups (Examiner notes that there is no limitation on the number of apparatuses or the number of groups) and performing federated learning using a global model, using generic computing components).
Step 2b: the claim additionally recites
in the aggregation processing the first processing circuitry selects (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f))
the first processing circuitry divides the n processing apparatuses I into Q groups, performs federated learning in cooperation with a second processing circuitry of a process apparatus belonging to each group using the Q optimal neural network possibilities as the first global model (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f); i.e. the claim recites generic actions of dividing apparatuses into groups (Examiner notes that there is no limitation on the number of apparatuses or the number of groups) and performing federated learning using a global model, using generic computing components).
Regarding dependent claim 4:
Step 2a Prong One: incorporates the rejection of claim 1.
Step 2a Prong Two: the claim additionally recites a learning server apparatus of the learning system according to claim 1 (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Step 2b: the claim additionally recites a learning server apparatus of the learning system according to claim 1 (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Regarding dependent claim 5:
Step 2a Prong One: incorporates the rejection of claim 1.
Step 2a Prong Two: the claim additionally recites a processing apparatus of the learning system according to claim 1 (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Step 2b: the claim additionally recites a processing apparatus of the learning system according to claim 1 (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Regarding dependent claim 7:
Step 2a Prong One: incorporates the rejection of claim 4.
Step 2a Prong Two: the claim additionally recites a non-transitory computer-readable recording medium having recorded thereon a program for causing a computer to function as the learning server apparatus according to claim 4 (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Step 2b: the claim additionally recites a non-transitory computer-readable recording medium having recorded thereon a program for causing a computer to function as the learning server apparatus according to claim 4 (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Regarding dependent claim 8:
Step 2a Prong One: incorporates the rejection of claim 5.
Step 2a Prong Two: the claim additionally recites a non-transitory computer-readable recording medium having recorded thereon a program for causing a computer to function as the processing apparatus according to claim 5 (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Step 2b: the claim additionally recites a non-transitory computer-readable recording medium having recorded thereon a program for causing a computer to function as the processing apparatus according to claim 5 (mere instructions to apply the exception using generic computer components as discussed in MPEP 2106.05(f)).
Therefore, in view of the considerations set forth in MPEP 2106.04(d), 2106.05(a)-(c) and (e)-(h), the additional elements as recited in the dependent claims discussed above alone or in combination do not integrate the judicial exception into a practical application as they are mere insignificant extra solution activity, combined with implementing the abstract idea using generic computer components, and limitations describing a field of use or technological environment. The additional elements as discussed above, in combination with the abstract idea, are not sufficient to amount to significantly more than the judicial exception as they are well, understood, routine and conventional activity as disclosed in combination with generic computer functions and components used to implement the abstract idea, and limitations describing a field of use or technological environment.
Claim Rejections – 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims under pre-AIA 35 U.S.C. 103(a), the examiner presumes that the subject matter of the various claims was commonly owned at the time any inventions covered therein were made absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and invention dates of each claim that was not commonly owned at the time a later invention was made in order for the examiner to consider the applicability of pre-AIA 35 U.S.C. 103(c) and potential pre-AIA 35 U.S.C. 102€, (f) or (g) prior art under pre-AIA 35 U.S.C. 103(a).
Claims 1 and 3-8 are rejected under 35 U.S.C. 103 as being unpatentable over Satheesh Kumar et al. (US 20230065937 A1) in view of Zhou et al. (US 20220414426 A1).
With respect to claim 1, Satheesh Kumar teaches a learning system comprising:
a learning server apparatus (e.g. paragraph 0033, Fig. 1, central server 102 participating in federated learning process); and
n processing apparatuses i, wherein, when i = 1, 2,..., n, and r = 0, 1,..., A-1, the processing apparatuses i each include: second processing circuitry (e.g. paragraph 0033, Fig. 1, local client computing devices 104 participating in federated learning process; paragraph 0068, apparatus such as client device 104 including processing circuitry) configured to:
execute a second federated learning processing (e.g. paragraph 0033, Fig. 1, local client computing devices 104 participating in federated learning process; paragraphs 0051-0054, method of Fig. 4 performed at local client computing devices 104, including training a local model resulting local model update, sending the local model update to the central server computing device, receiving from the central server computing device a first updated global model, etc.); and
execute a score calculation processing in which the second processing circuitry calculates a score sir when local data di is applied to each of A neural networks r (e.g. paragraph 0047, each local client device computing score based on global model, such as determining an error of its local model; paragraphs 0055, 0058, and 0059, determining that the first updated global model does not meet a local criteria, which comprises computing a score based on the first updated global model wherein the score exceeds a threshold, and where the computed score may comprise error information in a prediction),
the learning server apparatus includes: first processing circuitry (e.g. paragraph 0068, apparatus such as central server 102 including processing circuitry) configured to:
execute a first federated learning processing (e.g. paragraph 0033, Fig. 1, central server 102 participating in federated learning process; paragraph 0046-0047, central server using local models and received context to obtain updated global model, and sending updated global model to local client computing devices; paragraphs 0060-0067, method of Fig. 5 performed by central server including receiving local model updates from local client devices, training a global model using the local updates resulting in the first updated global model, sending to local devices, receiving context information, training the global model using local updates and context information resulting in second updated global model, and sending to the local client devices); and
execute an aggregation processing in which the first processing circuitry aggregates A neural networks using A x n scores sir, and selects an optimal neural network (e.g. paragraph 0034, global model obtained by aggregating weights from local models; paragraph 0049, upon receiving computed score, central server adjusting the global mode such as by modifying the objective function to include the computed scores and other context information as inputs, and using this to recompute the global model; paragraph 0067, modifying objective function of global model to include computed score and other context information as inputs; i.e. where obtaining the global model by aggregating weights from local models, including by performing this iteratively using local model updates, and further by also utilizing the calculated scores received from local devices, is analogous to performing aggregation processing to aggregate the A neural networks (i.e. the corresponding local models, using their respective weights) using the respective scores, to generate/select an optimal neural network (i.e. the updated global model which is based on local model updates and corresponding scores for each); Examiner notes that the claims do not appear to require more than one processing apparatus (i.e. n may be equal to 1) and one neural network (i.e. A may be equal to 1)),
in the first federated learning processing and the second federated learning processes, the first processing circuitry and the second processing circuitries of the n processing apparatuses i cooperate to perform federated learning using the selected optimal neural network as a first global model (e.g. paragraph 0033, indicating that the federated learning process is repeated iteratively across multiple rounds; paragraphs 0045-0049, describing federated learning processing between local devices and central server performed cooperatively using a global model which is continuously updated/optimized; paragraph 0050, iteratively repeating the process until convergence or stopping factor reached; paragraph 0057, receiving, at the local device, the updated global model from the server; paragraph 0066, sending, at the server, the updated global model to the local device; i.e. the federated learning processing/processes are performed cooperatively using the updated/optimized global model on an iterative, repeated basis over a number of training rounds, analogous to first and second processes/processing circuitries cooperating to perform federated learning using a selected optimal neural network as a first global model).
Satheesh Kumar does not explicitly disclose the score sir includes an index with which a neural network having an excellent learning effect can be searched for. However, Zhou teaches the score sir includes an index with which a neural network having an excellent learning effect can be searched for (e.g. paragraph 0009, performing inference on each child model on first hardware to obtain an evaluation indicator value of the trained child model on the first hardware; obtaining evaluation indicator values of the plurality of child model and determining based on evaluation indicator values and neural network architectures corresponding to the child models, a first target neural network architecture that meets a preset condition; searching does not need to depend on actual evaluation indicator value of previous child model, and neural architecture search process and training process of initial child model can be process in parallel and the neural architecture search process is decoupled form the training process of the initial child model; paragraph 0012, searcher includes evaluator and controller, where the evaluator is trained based on neural network architectures and evaluation indicator values, and the searcher can then determine a first target neural network architecture; paragraph 0013, evaluation indicator values of child models may be used to predict an evaluation indicator value corresponding to a neural network architecture; paragraph 0025, performing federated learning on child models; paragraphs 0061-0062 evaluation indicator values include metric values obtained by evaluating the child model, such as hardware-related performance values (inference time, quantity of activations, throughput, power consumption, video RAM usage) and hardware-irrelevant performance values such as accuracy, precision, and recall; paragraph 0118, searching for neural network architecture based on neural network architectures corresponding to child models and corresponding evaluation indicator values; paragraph 0154, performing federated learning on child models to obtain evaluation indicator values; performing search for first target neural network architecture based on the obtained evaluation indicator values; i.e. the evaluation indicator value (which is itself a performance-based evaluation score/metric) for each model evaluated in a federated learning process may be used as an index for performing neural architecture search to find an optimal neural network architecture).
Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention having the teachings of Satheesh Kumar and Zhou in front of him to have modified the teachings of Satheesh Kumar (directed to context-level federated learning), to incorporate the teachings of Zhou (directed to neural architecture search, including using federated learning) to include the capability to use the score (i.e. score of Satheesh Kumar, such as an error of the relevant model and evaluation indicator of Zhou, such as an accuracy/precision of the relevant model) as an index with which a neural network having an excellent learning effect can be searched for (as taught by Zhou). One of ordinary skill would have been motivated to perform such a modification in order to reduce duration of, and improve efficiency of, neural architecture search as taught by Zhou (paragraph 0069).
With respect to claim 6, Satheesh Kumar teaches a learning method using a learning server apparatus that includes first processing circuitry and n processing apparatuses i that include second processing circuitry (e.g. paragraph 0033, Fig. 1, local client computing devices 104 and central server 102 participating in federated learning process; paragraph 0068, apparatus such as client device 104 and central server 102 including processing circuitry), the learning method comprising:
when i = 1, 2,..., n, and r = 0, 1,..., A-1, a score calculation step of calculating, by the second processing circuitry of the processing apparatus i, a score sir when local data di is applied to each of A neural networks r (e.g. paragraph 0047, each local client device computing score based on global model, such as determining an error of its local model; paragraphs 0055, 0058, and 0059, determining that the first updated global model does not meet a local criteria, which comprises computing a score based on the first updated global model wherein the score exceeds a threshold, and where the computed score may comprise error information in a prediction);
an aggregation step of aggregating, by the first processing circuitry of the learning server apparatus, A neural networks using A x n scores sir, and selecting an optimal neural network (e.g. paragraph 0034, global model obtained by aggregating weights from local models; paragraph 0049, upon receiving computed score, central server adjusting the global mode such as by modifying the objective function to include the computed scores and other context information as inputs, and using this to recompute the global model; paragraph 0067, modifying objective function of global model to include computed score and other context information as inputs; i.e. where obtaining the global model by aggregating weights from local models, including by performing this iteratively using local model updates, and further by also utilizing the calculated scores received from local devices, is analogous to performing aggregation processing to aggregate the A neural networks (i.e. the corresponding local models, using their respective weights) using the respective scores, to generate/select an optimal neural network (i.e. the updated global model which is based on local model updates and corresponding scores for each); Examiner notes that the claims do not appear to require more than one processing apparatus (i.e. n may be equal to 1) and one neural network (i.e. A may be equal to 1)); and
a federated learning step of cooperating, by the first processing circuitry of the learning server apparatus and the second processing circuitries of the n processing apparatuses i, to perform federated learning using the selected optimal neural network as a first global model (e.g. paragraph 0033, indicating that the federated learning process is repeated iteratively across multiple rounds; paragraphs 0045-0049, describing federated learning processing between local devices and central server performed cooperatively using a global model which is continuously updated/optimized; paragraph 0050, iteratively repeating the process until convergence or stopping factor reached; paragraph 0057, receiving, at the local device, the updated global model from the server; paragraph 0066, sending, at the server, the updated global model to the local device; i.e. the federated learning processing/processes are performed cooperatively using the updated/optimized global model on an iterative, repeated basis over a number of training rounds, analogous to first and second processes/processing circuitries cooperating to perform federated learning using a selected optimal neural network as a first global model).
Satheesh Kumar does not explicitly disclose the score sir includes an index with which a neural network having an excellent learning effect can be searched for. However, Zhou teaches the score sir includes an index with which a neural network having an excellent learning effect can be searched for (e.g. paragraph 0009, performing inference on each child model on first hardware to obtain an evaluation indicator value of the trained child model on the first hardware; obtaining evaluation indicator values of the plurality of child model and determining based on evaluation indicator values and neural network architectures corresponding to the child models, a first target neural network architecture that meets a preset condition; searching does not need to depend on actual evaluation indicator value of previous child model, and neural architecture search process and training process of initial child model can be process in parallel and the neural architecture search process is decoupled form the training process of the initial child model; paragraph 0012, searcher includes evaluator and controller, where the evaluator is trained based on neural network architectures and evaluation indicator values, and the searcher can then determine a first target neural network architecture; paragraph 0013, evaluation indicator values of child models may be used to predict an evaluation indicator value corresponding to a neural network architecture; paragraph 0025, performing federated learning on child models; paragraphs 0061-0062 evaluation indicator values include metric values obtained by evaluating the child model, such as hardware-related performance values (inference time, quantity of activations, throughput, power consumption, video RAM usage) and hardware-irrelevant performance values such as accuracy, precision, and recall; paragraph 0118, searching for neural network architecture based on neural network architectures corresponding to child models and corresponding evaluation indicator values; paragraph 0154, performing federated learning on child models to obtain evaluation indicator values; performing search for first target neural network architecture based on the obtained evaluation indicator values; i.e. the evaluation indicator value (which is itself a performance-based evaluation score/metric) for each model evaluated in a federated learning process may be used as an index for performing neural architecture search to find an optimal neural network architecture).
Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention having the teachings of Satheesh Kumar and Zhou in front of him to have modified the teachings of Satheesh Kumar (directed to context-level federated learning), to incorporate the teachings of Zhou (directed to neural architecture search, including using federated learning) to include the capability to use the score (i.e. score of Satheesh Kumar, such as an error of the relevant model and evaluation indicator of Zhou, such as an accuracy/precision of the relevant model) as an index with which a neural network having an excellent learning effect can be searched for (as taught by Zhou). One of ordinary skill would have been motivated to perform such a modification in order to reduce duration of, and improve efficiency of, neural architecture search as taught by Zhou (paragraph 0069)
With respect to claim 3, Satheesh Kumar in view of Zhou teaches all of the limitations of claim 1 as previously discussed, and Zhou further teaches wherein
in the aggregation processing the first processing circuitry selects Q optimal neural network possibilities in a case where an optimal neural network cannot be selected (e.g. paragraph 0089, Fig. 4, generating plurality of neural network architectures based on a search space; paragraph 0103, sending the plurality of neural network architectures to model training platform; i.e. where no candidate optimal neural networks yet exist for evaluation, such as a beginning of a process, this is analogous to a case where an optimal neural network cannot be selected (i.e. because none exist to select); in this case, the system selects a set of potential neural network architectures such as using neural network architecture search in a search space, analogous to selecting Q optimal neural network possibilities), and
the first processing circuitry divides the n processing apparatuses i into Q groups, performs federated learning in cooperation with a second processing circuitry of a processing apparatus belonging to each group using the Q optimal neural network possibilities as a first global model, compares accuracies of the Q optimal neural network possibilities after the federated learning, and selects an optimal neural network with the highest accuracy as the optimal neural network (e.g. paragraph 0025, indicating that the model training platform performs federated learning on each of initial child models; paragraph 0082, preset condition such as the evaluation indicator of the child model reaching a preset value, determining corresponding child model as the target neural network; paragraph 0086, indicating that the generator, searcher, model training platform, and model inference platform may be deployed in a variety of configurations such as in different or same cloud computing clusters, on physical devices, etc.; paragraphs 0105-0106, obtaining plurality of child models based on the plurality of neural network architectures, including performing weight initialization on the architectures to obtain the child models and then training the plurality of initial child models using training data to obtain the plurality of child models; paragraph 0111-0113, sending plurality of child models to model inference platform and performing inference on the plurality of child models to obtain evaluation indicator values of the plurality of child models; evaluation indicator values such as precision; paragraph 0115, sending, to the searcher the neural network architectures corresponding to the plurality of child models and the evaluation indicator values; paragraph 0117, determining based on the neural network architectures corresponding to the plurality of child models and the evaluation indicator values, a first target neural network architecture that meets a preset condition; i.e. using a federated learning process (including within the model training platform itself and also between different components of the system deployed across corresponding devices/clusters of devices), where child/candidate model is trained and evaluated with respect to an evaluation indicator such as precision/accuracy, and the neural networks are subsequently evaluated with respect to the evaluation indicator (i.e. precision/accuracy) in order to select the one which best meets the preset condition (i.e. such as meeting a preset value, where one of the networks may meet the preset value and the other may not, such that it has the highest accuracy) as a target neural network (analogous to selecting an optimal neural network with highest accuracy as the optimal neural network); Examiner notes that the claim does not appear to require any particular number n of processing apparatuses or any particular number Q of groups, and therefore, that these are interpreted as at least one device and one group).
Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention having the teachings of Satheesh Kumar and Zhou in front of him to have modified the teachings of Satheesh Kumar (directed to context-level federated learning), to incorporate the teachings of Zhou (directed to neural architecture search, including using federated learning) to include the capability to, where an optimal neural network cannot be selected (such as at the beginning of a process when none yet exist for selection), select a plurality of candidate neural network architectures via a corresponding search space, and then performing federated learning using the candidate neural network architectures including training them and determining corresponding evaluation indictors such as accuracy/precision, the subsequently evaluating each of the candidate neural network architectures using the indicator (accuracy/precision) with respect to the indicator (accuracy/precision) meeting a preset value (i.e. where a network meeting a preset accuracy value/criteria would have a higher accuracy as compared to other networks failing to meet the accuracy value/criteria), and selecting the neural network architecture which meets the preset value (i.e. instead of those which do not) as a target/optimal neural network architecture (as taught by Zhou). One of ordinary skill would have been motivated to perform such a modification in order to reduce duration of, and improve efficiency of, neural architecture search as taught by Zhou (paragraph 0069)
With respect to claim 4, Satheesh Kumar in view of Zhou teaches all of the limitations of claim 1 as previously discussed, and Satheesh Kumar further teaches a learning server apparatus of the learning system according to claim 1 (e.g. paragraph 0033, Fig. 1, central server 102 participating in federated learning process; paragraph 0068, apparatus such as central server 102 including processing circuitry).
With respect to claim 7, Satheesh Kumar in view of Zhou teaches all of the limitations of claim 4 as previously discussed, and Satheesh Kumar further teaches a non-transitory computer-readable recording medium having recorded thereon a program for causing a computer to function as the learning server apparatus according to claim 4 (e.g. paragraph 0068, Fig. 6, diagram of apparatus 600, such as central server computing device 102, including computer readable medium storing a computer program comprising instructions which, when executed, cause the apparatus to perform the disclosed processes/steps).
With respect to claim 5, Satheesh Kumar in view of Zhou teaches all of the limitations of claim 1 as previously discussed, and Satheesh Kumar further teaches a processing apparatus of the learning system according to claim 1 (e.g. paragraph 0033, Fig. 1, local client computing devices 104 participating in federated learning process; paragraph 0068, apparatus such as client device 104 including processing circuitry).
With respect to claim 8, Satheesh Kumar in view of Zhou teaches all of the limitations of claim 5 as previously discussed, and Satheesh Kumar further teaches a non-transitory computer-readable recording medium having recorded thereon a program for causing a computer to function as the processing apparatus according to claim 5 (e.g. paragraph 0068, Fig. 6, diagram of apparatus 600, such as local client computing device 104, including computer readable medium storing a computer program comprising instructions which, when executed, cause the apparatus to perform the disclosed processes/steps).
Claim 2 is rejected under 35 U.S.C. 103 as being unpatentable over Satheesh Kumar in view of Zhou, further in view of .
With respect to claim 2, Satheesh Kumar in view of Zhou teaches all of the limitations of claim 1 as previously discussed, and Satheesh Kumar further teaches wherein
the score sir is a correlation score of a weight when the local data di is applied to the neural network r (e.g. paragraph 0034, global model obtained by aggregating weights from local models; paragraph 0036, local devices sending context to central server including weights and additional information; paragraph 0037, additional context such as the amount of error; paragraph 0042, context includes sending prediction error; paragraphs 0047-0049, local computing device computing score based on global model, such as determining error, and sending computed score/error to central server, which adjusts the global model based on the computed score/error; i.e. local devices applies local data to the neural network and sends back context information, where this context information may include both weights of the model and the computed score, such as the error, of the model; since these are included in the same context, the score/error is correlated with the weights, and is therefore analogous to a correlation score of a weight, which corresponds to the local data being applied to the neural network/model).
Satheesh Kumar and Zhou do not explicitly disclose:
the aggregation processing in which the first processing circuitry calculates a variation of the score sir for each neural network r,
performs a first action in a case where the variation is larger than a predetermined threshold value, and
performs a second action in a case where the variation is equal to or smaller than the predetermined threshold value.
However, Takasaki teaches:
the aggregation processing in which the first processing circuitry calculates a variation of the score sir for each neural network r (e.g. paragraph 0033, in each group, local device receives local models, and each local device uses local data to validate accuracies of all local models and a global model; paragraph 0036, Fig. 2(a) step 206, calculating variance of accuracies; i.e. where model accuracy/error is analogous to a score as previously cited),
performs a first action in a case where the variation is larger than a predetermined threshold value (e.g. paragraph 0036, Fig. 2(a) step 207 is no, and continuing to step 209; in response to determining that the variance in a group is not below the predetermined variance threshold, selecting groups whose variances are not below predetermined variance level and executing step 209 and further steps), and
performs a second action in a case where the variation is equal to or smaller than the predetermined threshold value (e.g. paragraph 0036, Fig. 2(a) step 207 is yes, and continuing to step 208; in response to determining that the variance in a group is below the predetermined variance threshold, performing step 208, such as excluding local models whose variance is below threshold).
Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention having the teachings of Satheesh Kumar, Zhou, and Takasaki in front of him to have modified the teachings of Satheesh Kumar (directed to context-level federated learning) and Zhou (directed to neural architecture search, including using federated learning), to incorporate the teachings of Takasaki (directed to parallel cross validation in collaborative machine learning) to include the capability to calculate variations in scores/accuracies for each neural network, and perform a first action when the variation is larger than a threshold value and perform a second action when the variation is smaller than the threshold value (as taught by Takasaki). One of ordinary skill would have been motivated to detect data bias in collaborative machine learning in advance without sharing local data and metadata containing confidential information, securing versatility and reliability of the global model as taught by Takasaki (paragraph 0019).
As previously discussed, Satheesh Kumar teaches that the second action is calculates the score Sr for each neural network r without considering the number of pieces of data of the local data di (e.g. paragraphs 0047-0049, local computing device computing score based on global model, such as determining error, and sending computed score/error to central server, which adjusts the global model based on the computed score/error; i.e. the score is determined using a error of the model without taking into account characteristics of the local dataset such as size/amount).
Satheesh Kumar, Zhou, and Takasaki do not explicitly disclose:
the first action is calculates a score Sir for each neural network r in consideration of a number of pieces of data of the local data di.
However, Ding teaches:
the first action is calculates a score Sir for each neural network r in consideration of a number of pieces of data of the local data di (e.g. paragraph 0037, performing aggregation of ML model using partial ML models using weighted average, where the weights of each partial ML model may be based on characteristics including a size of a client dataset; paragraph 0052, aggregating includes performing weighted average of structural parameter values; assigning weights to parameter sets corresponding to various partial ML models; weights assigned based on client information; if a first client data has significantly larger relative size than second client data, assigning relatively high weight to modified first parameter set and relatively low weight to the modified second parameter set; after assigning the weights, the model aggregator performs weighted average of structural parameter values to construct the aggregate ML model parameters; i.e. the system may assign a second score, different from a first score, for each neural network corresponding to the size/number of pieces of data in the corresponding local dataset; as discussed in paragraph 0052, because the assigned weights are relative to the corresponding dataset sizes, where there is a large variance between the sizes, the assigned weights effectively provide scores which consider the number of pieces of data of the local data);
the second action is calculates the score Sr for each neural network r without considering the number of pieces of data of the local data di (e.g. paragraph 0037, performing aggregation of ML model using partial ML models using weighted average, where the weights of each partial ML model may be based on characteristics including a size of a client dataset; paragraph 0052, aggregating includes performing weighted average of structural parameter values; assigning weights to parameter sets corresponding to various partial ML models; weights assigned based on client information; if a first client data has significantly larger relative size than second client data, assigning relatively high weight to modified first parameter set and relatively low weight to the modified second parameter set; after assigning the weights, the model aggregator performs weighted average of structural parameter values to construct the aggregate ML model parameters; i.e. the system may assign a second score, different from a first score, for each neural network corresponding to the size/number of pieces of data in the corresponding local dataset; as discussed in paragraph 0052, because the assigned weights are relative to the corresponding dataset sizes, where there is a small or no variance between the sizes, the assigned weights will be approximately equal to one another, providing the same effect as if the weights/scores are not considered at all).
Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the invention having the teachings of Satheesh Kumar, Zhou, Takasaki, and Ding in front of him to have modified the teachings of Satheesh Kumar (directed to context-level federated learning), Zhou (directed to neural architecture search, including using federated learning), and Takasaki (directed to parallel cross validation in collaborative machine learning), to incorporate the teachings of Ding (directed to privacy preserving cooperative learning in untrusted environments) to include the capability to calculate a weight/score for each model corresponding to a local dataset, where the score provides a relative weighting of model for aggregation purposes, is based on relative dataset sizes, such that the score/weighting is calculated in consideration of the dataset sizes/number of pieces of data in local datasets corresponding to each model (as taught by Takasaki). One of ordinary skill would have been motivated to solve problems associated with federated learning and split learning in untrusted environments, particularity ones with unbalanced computing resource access as taught by Ding (paragraph 0025).
It is noted that any citation to specific pages, columns, lines, or figures in the prior art references and any interpretation of the references should not be considered to be limiting in any way. “The use of patents as references is not limited to what the patentees describe as their own inventions or to the problems with which they are concerned. They are part of the literature of the art, relevant for all they contain,” In re Heck, 699 F.2d 1331, 1332-33, 216 USPQ 1038, 1039 (Fed. Cir. 1983) (quoting in re Lemelson, 397 F.2d 1006, 1009, 158 USPQ 275, 277 (GCPA 1968)). Further, a reference may be relied upon for all that it would have reasonably suggested to one having ordinary skill the art, including nonpreferred embodiments. Merck & Co, v. Biocraft Laboratories, 874 F.2d 804, 10 USPQ2d 1843 (Fed. Cir.), cert, denied, 493 U.S. 975 (1989). See also Upsher-Smith Labs. v. Pamlab, LLC, 412 F,3d 1319, 1323, 75 USPQ2d 1213, 1215 (Fed. Cir, 2005): Celeritas Technologies Ltd. v. Rockwell International Corp., 150 F.3d 1354, 1361, 47 USPQ2d 1516, 1522-23 (Fed. Cir. 1998).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JEREMY L STANLEY whose telephone number is (469)295-9105. The examiner can normally be reached on Monday-Friday from 9:00 AM to 5:00 PM CST.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Abdullah Al Kawsar, can be reached at telephone number (571) 270-3169. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from Patent Center and the Private Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from Patent Center or Private PAIR. Status information for unpublished applications is available through Patent Center and Private PAIR for authorized users only. Should you have questions about access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
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) Form at https://www.uspto.gov/patents/uspto-automated- interview-request-air-form.
/JEREMY L STANLEY/
Primary Examiner, Art Unit 2127
Prosecution Insights