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 .
Claims 1-19 and 33 are presented for examination.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on May 4, 2023 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Specification
The disclosure is objected to because of the following informalities: on page 1, line 17 of the specification as originally filed, “a machine a machine” should be “a machine”. Appropriate correction is required.
Claim Objections
Claim 16 is objected to because of the following informalities: “point of converge” should be “point of convergence”. Appropriate correction is required.
Claim Rejections - 35 USC § 112
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.
Claim 9 is 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.
Claim 9 recites “aggregating the one or more parameters of the machine learning model trained by the one or more worker nodes in the previous round of training”. However, claim 8, on which claim 9 depends, already recites that “the one or more parameters of the updated machine learning model comprise an aggregation of the one or more parameters of the machine learning model trained by the one or more worker nodes in the previous round of training”. Therefore, claim 9 fails to limit claim 8 further.
Applicant may cancel the claim(s), amend the claim(s) to place the claim(s) in proper dependent form, rewrite the claim(s) in independent form, or present a sufficient showing that the dependent claim(s) complies with the statutory requirements.
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 1-19 and 33 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The analysis of the claims will follow the 2019 Revised Patent Subject Matter Eligibility Guidance, 84 Fed. Reg. 50 (“2019 PEG”).
Claim 1
Step 1: The claim recites a method; therefore, the claim is directed to the statutory category of processes.
Step 2A Prong 1: The claim recites, inter alia, “selecting one or more worker nodes of a plurality of worker nodes …, wherein the one or more worker nodes are selected to optimize a performance of an updated machine learning model for a validation dataset after [a] round of training”. This limitation could encompass mentally selecting nodes for the purpose of optimizing a performance of an updated machine learning model on a validation dataset.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. The claim further recites that the selection is performed “to train a machine learning model in a round of training” and that “the updated machine learning model has one or more parameters of the machine learning model trained by the one or more worker nodes in a previous round of training.” These limitations merely restrict the field of use of the judicial exception to model training. MPEP § 2106.05(h).
Step 2B: The claim does not contain significantly more than the judicial exception. The analysis at this step mirrors that of step 2A, prong 2. As an ordered whole, the claim is directed to a mentally performable process of selecting worker nodes for the purpose of optimizing a machine learning performance. Nothing in the claim provides significantly more than this. As such, the claim is not patent eligible.
Claim 2
Step 1: A process, as above.
Step 2A Prong 1: The claim recites “selecting a mask indicative of the one or more worker nodes.” This limitation could encompass mentally selecting the mask.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 1 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 1 analysis.
Claim 3
Step 1: A process, as above.
Step 2A Prong 1: The claim recites that “the mask is a binary vector comprising a value of one to indicate the one or more worker nodes and a value of zero to indicate any other worker nodes of the plurality of worker nodes.” Selecting the mask may be performed mentally under these further assumptions.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 2 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 2 analysis.
Claim 4
Step 1: A process, as above.
Step 2A Prong 1: The claim recites that “the one or more worker nodes are selected to optimize the performance of the updated machine learning model by selecting the one or more worker nodes that maximize a reward for the performance of the updated machine learning model.” This limitation could encompass mentally selecting the nodes to maximize a reward for the updated model’s performance.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 1 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 1 analysis.
Claim 5
Step 1: A process, as above.
Step 2A Prong 1: The claim recites that “the reward for the performance of the updated machine learning model is maximized if it is determined to be higher than a reward for a performance of the machine learning model in a previous round of training.” Selecting the node to maximize these rewards remains mentally performable under these further assumptions.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 4 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 4 analysis.
Claim 6
Step 1: A process, as above.
Step 2A Prong 1: The claim recites that “the reward for the performance of the updated machine learning model is based on a performance metric for each of the one or more worker nodes that is indicative of a performance of the worker node.” Selecting the node to maximize these rewards remains mentally performable under these further assumptions.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 4 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 4 analysis.
Claim 7
Step 1: A process, as above.
Step 2A Prong 1: The claim recites the same judicial exceptions as in claim 6.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. The claim further recites “receiving the performance metric from each of the one or more worker nodes.” This limitation is directed to the insignificant extra-solution activity of mere data gathering and output. MPEP § 2106.05(g).
Step 2B: The claim does not contain significantly more than the judicial exception. The claim further recites “receiving the performance metric from each of the one or more worker nodes.” This limitation is directed to the well-understood, routine, and conventional activity of receiving or transmitting data over a network. MPEP § 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network).
Claim 8
Step 1: A process, as above.
Step 2A Prong 1: The claim recites the same judicial exceptions as in claim 1.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. The claim further recites that “the one or more parameters of the updated machine learning model comprise an aggregation of the one or more parameters of the machine learning model trained by the one or more worker nodes in the previous round of training.” This recitation merely restricts the judicial exception to the field of use of model training. MPEP § 2106.05(h).
Step 2B: The claim does not contain significantly more than the judicial exception. The claim further recites that “the one or more parameters of the updated machine learning model comprise an aggregation of the one or more parameters of the machine learning model trained by the one or more worker nodes in the previous round of training.” This recitation merely restricts the judicial exception to the field of use of model training. MPEP § 2106.05(h).
Claim 9
Step 1: A process, as above.
Step 2A Prong 1: The claim recites “aggregating the one or more parameters of the machine learning model trained by the one or more worker nodes in the previous round of training.” This limitation could encompass mentally aggregating the parameters into a single set.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 8 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 8 analysis.
Claim 10
Step 1: A process, as above.
Step 2A Prong 1: The claim recites that “the aggregation is an average.” Averaging parameters is a mental process and a mathematical concept.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 8 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 8 analysis.
Claim 11
Step 1: A process, as above.
Step 2A Prong 1: The claim recites that “the selection is performed for at least one worker node of the plurality of worker nodes; and for each worker node for which the selection is performed, the one or more worker nodes are selected to optimize the performance of the updated machine learning model for that worker node.” These limitations could encompass mentally performing the selection of the one or more worker nodes to satisfy the claimed criteria.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 1 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 1 analysis.
Claim 12
Step 1: A process, as above.
Step 2A Prong 1: The claim recites that “the selection is performed for at least two worker nodes of the plurality of worker nodes simultaneously.” This limitation could encompass mentally performing the selection.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 11 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 11 analysis.
Claim 13
Step 1: A process, as above.
Step 2A Prong 1: The claim recites that “for each worker node for which the selection is performed, the validation dataset is a validation dataset of that worker node.” Performing the selection remains mentally performable under these further assumptions.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 11 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 11 analysis.
Claim 14
Step 1: A process, as above.
Step 2A Prong 1: The claim recites the same judicial exceptions as in claim 1.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. The claim further recites “initiating transmission of the one or more parameters of the machine learning model trained by the one or more worker nodes in the previous round of training towards the one or more worker nodes for the one or more worker nodes to further train the updated machine learning model.” This limitation is directed to the insignificant extra-solution activity of mere data gathering and output. MPEP § 2106.05(g).
Step 2B: The claim does not contain significantly more than the judicial exception. The claim further recites “initiating transmission of the one or more parameters of the machine learning model trained by the one or more worker nodes in the previous round of training towards the one or more worker nodes for the one or more worker nodes to further train the updated machine learning model.” This limitation is directed to the well-understood, routine, and conventional activity of receiving or transmitting data over a network. MPEP § 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network).
Claim 15
Step 1: A process, as above.
Step 2A Prong 1: The claim recites “repeating the method until a point of convergence is reached.” Since the underlying steps of the method are mentally performable, repeating them is also mentally performable.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 1 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 1 analysis.
Claim 16
Step 1: A process, as above.
Step 2A Prong 1: The claim recites that “the point of converge[nce] is reached when: a predefined minimum number of training rounds is completed; and/or an increase in the performance of the updated machine learning model for the validation dataset is less than a predefined threshold.” Repeating the method until a point of convergence is reached remains mentally performable under these further assumptions.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 15 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 15 analysis.
Claim 17
Step 1: A process, as above.
Step 2A Prong 1: The claim recites the same judicial exceptions as in claim 1.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. The claim further recites “prior to selecting the one or more worker nodes: initiating transmission of one or more parameters of the machine learning model towards the one or more worker nodes for the one or more worker nodes to train the machine learning model in the previous round of training; and receiving the one or more parameters of the machine learning model trained by the one or more worker nodes in the previous round of training from the one or more worker nodes.” These limitations are directed to the insignificant extra-solution activity of mere data gathering and output. MPEP § 2106.05(g).
Step 2B: The claim does not contain significantly more than the judicial exception. The claim further recites “prior to selecting the one or more worker nodes: initiating transmission of one or more parameters of the machine learning model towards the one or more worker nodes for the one or more worker nodes to train the machine learning model in the previous round of training; and receiving the one or more parameters of the machine learning model trained by the one or more worker nodes in the previous round of training from the one or more worker nodes.” These limitations are directed to the well-understood, routine, and conventional activity of receiving or transmitting data over a network. MPEP § 2106.05(d)(II); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network).
Claim 18
Step 1: A process, as above.
Step 2A Prong 1: The claim recites “selecting a weighting for the one or more worker nodes that controls the amount by which each of the one or more worker nodes contributes to training the machine learning model in the round of training, wherein the weighting is selected to optimize the performance of the updated machine learning model for the validation dataset after the round of training.” This limitation could encompass mentally selecting the weighting based on the claimed criteria.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 1 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 1 analysis.
Claim 19
Step 1: A process, as above.
Step 2A Prong 1: The claim recites “the weighting is selected based on a state of the one or more worker nodes.” This limitation could encompass mentally performing the weighting based on the claimed criteria.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 18 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 18 analysis.
Claim 33
Step 1: The claim recites a master node comprising processing circuitry; therefore, the claim is directed to the statutory category of machines.
Step 2A Prong 1: The claim recites the same judicial exceptions as in claim 1.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. The analysis at this step mirrors that of claim 1, with the exception that claim 33 recites a “master node for managing training of a machine learning model, the master node comprising processing circuitry configured to cause the master node to [perform the method].” However, this is a mere instruction to apply the judicial exception using a generic computer. MPEP § 2106.05(f).
Step 2B: The claim does not contain significantly more than the judicial exception. The analysis at this step mirrors that of claim 1, with the exception that claim 33 recites a “master node for managing training of a machine learning model, the master node comprising processing circuitry configured to cause the master node to [perform the method].” However, this is a mere instruction to apply the judicial exception using a generic computer. MPEP § 2106.05(f).
Claim Rejections - 35 USC § 103
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 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 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 the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1, 4, 6-7, 11-13, 15, 18-19, and 33 are rejected under 35 U.S.C. 103 as being unpatentable over Sathe et al. (US 20220114019) (“Sathe”) in view of Froloff (US 20200117897) (“Froloff”).
Regarding claim 1, Sathe discloses “[a] method performed by a master node for managing training of a machine learning model, the method comprising:
selecting one or more worker nodes of a plurality of worker nodes to train a machine learning model in a round of training, wherein the one or more worker nodes are selected to optimize a performance of an updated machine learning model … (joint optimizer may select worker nodes; joint optimizer may select worker nodes for executing the pipeline based on the predicted pipeline training resources required [i.e., to optimize the performance of the model training/updating] – Sathe, paragraph 42; see also Fig. 1 (showing that the joint optimizer is part of a pipeline training server [master node])), wherein the updated machine learning model has one or more parameters of the machine learning model trained by the one or more worker nodes in a previous round of training (joint optimizer may adjust models [i.e., train the parameters] based on a feedback loop [i.e., there are multiple rounds of training]; joint optimizer may adjust models by determining a loss following the training of one or more pipelines by the one or more worker nodes – Sathe, paragraph 39 and Fig. 2).”
Sathe appears not to disclose explicitly the further limitations of the claim. However, Froloff discloses “optimiz[ing] a performance of an updated machine learning model for a validation dataset after the round of training (test or validation dataset is used to provide an unbiased evaluation of a final model fit on a training dataset [i.e., validation is performed after training]; some supervised learning algorithms require the user to determine certain program control parameters which may be adjusted by optimizing performance on a validation set – Froloff, paragraph 120) ….”
Froloff and the instant application both relate to validation of machine learning models and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sathe to optimize the performance of the model on a validation dataset after training, as disclosed by Froloff, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would increase the accuracy of the model by ensuring that it has been properly trained. See Froloff, paragraph 120.
Claim 33 is a master node claim corresponding to method claim 1 and is rejected for the same reasons as given in the rejection of that claim.
Regarding claim 4, Sathe, as modified by Froloff, discloses that “the one or more worker nodes are selected to optimize the performance of the updated machine learning model by selecting the one or more worker nodes that maximize a reward for the performance of the updated machine learning model (model captures how each feature impacts an amount of resources required for each of the worker nodes to train a pipeline; once the model is trained, the joint optimizer is capable of applying the trained model to a new set of features to output the predicted performance measures [reward] for each worker node with respect to training a pipeline – Sathe, paragraph 35; joint optimizer may select the best predicted worker node [i.e., the node that maximizes the predicted performance measure/reward] – id. at paragraph 37).”
Regarding claim 6, Sathe, as modified by Froloff, discloses that “the reward for the performance of the updated machine learning model is based on a performance metric for each of the one or more worker nodes that is indicative of a performance of the worker node (performance predictor may receive heartbeat features [performance metrics] from worker nodes; the heartbeat features quantify a busyness and power of each of the worker nodes; joint optimizer may train a model for determining which of the worker nodes can train the pipeline in a least amount of resources based on the heartbeat features collected therein – Sathe, paragraph 29; joint optimizer predicts required performance measures for each of the worker nodes via a performance predictor that receives as inputs, inter alia, the heartbeat features [i.e., the heartbeat features/performance metrics are used to determine the performance measure/reward] – id. at paragraph 35).”
Regarding claim 7, Sathe, as modified by Froloff, discloses “receiving the performance metric from each of the one or more worker nodes (performance predictor may receive heartbeat features [performance metrics] from worker nodes; the heartbeat features quantify a busyness and power of each of the worker nodes; joint optimizer may train a model for determining which of the worker nodes can train the pipeline in a least amount of resources based on the heartbeat features collected therein – Sathe, paragraph 29).”
Regarding claim 11, Sathe, as modified by Froloff, discloses that “the selection is performed for at least one worker node of the plurality of worker nodes (joint optimizer may select at least one of the one or more worker nodes for executing the pipeline based on the training resources required – Sathe, paragraph 37); and
for each worker node for which the selection is performed, the one or more worker nodes are selected to optimize the performance of the updated machine learning model for that worker node (joint optimizer may select the top three predicted workers; once the joint optimizer has made enough evaluations to identify a best performing worker node, the joint optimizer may send all pipelines to the best performing worker node – Sathe, paragraph 37 [i.e., the worker node is selected to optimize the performance of the ML model]).”
Regarding claim 12, Sathe, as modified by Froloff, discloses that “the selection is performed for at least two worker nodes of the plurality of worker nodes simultaneously (joint optimizer may select the top three predicted workers [i.e., it may select three workers simultaneously] – Sathe, paragraph 37).”
Regarding claim 13, the rejection of claim 11 is incorporated. Sathe further discloses that “for each worker node for which the selection is performed, the … dataset is a … dataset of that worker node (embodiments may include predicting an amount of one or more resources required for each of the one or more worker nodes to train the one or more pipelines using the one or more datasets based on models that correlate worker node features, pipeline features, and dataset features with resources and identifying a worker node that requires a least amount of resources for training the pipelines [i.e., the dataset is associated with the worker node that trains the pipelines]– Sathe, paragraph 3).”
Froloff further discloses “validation dataset[s]”, as shown in the rejection of claim 1. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sathe to optimize performance on validation datasets, as disclosed by Froloff, for substantially the same reasons as given in the rejection of claim 1.
Regarding claim 15, Sathe, as modified by Froloff, discloses “repeating the method until a point of convergence is reached (joint optimizer may adjust models based on a feedback loop [i.e., repeatedly until the feedback indicates that the loop is to be terminated, i.e., until convergence]; the joint optimizer may adjust models by determining a loss following the training of one or more pipelines by the worker nodes – Sathe, paragraph 39).”
Regarding claim 18, the rejection of claim 1 is incorporated. Sathe further discloses “selecting a weighting for the one or more worker nodes that controls the amount by which each of the one or more worker nodes contributes to training the machine learning model in the round of training,
wherein the weighting is selected to optimize the performance of the updated machine learning model (joint optimizer may select the top three predicted workers; once the joint optimizer has made enough evaluations to identify a best performing worker node, the joint optimizer may send all pipelines to the best performing worker node – Sathe, paragraph 37 [i.e., the weight of the best performing node is 1 and that of all others is 0, this selection being made to optimize the performance of the model]) ….”
Sathe appears not to disclose explicitly the further limitations of the claim. However, Froloff discloses “optimiz[ing] the performance of the updated machine learning model for the validation dataset after the round of training (test or validation dataset is used to provide an unbiased evaluation of a final model fit on a training dataset [i.e., validation is performed after training]; some supervised learning algorithms require the user to determine certain program control parameters which may be adjusted by optimizing performance on a validation set – Froloff, paragraph 120) ….” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sathe to optimize the performance of the model on a validation dataset after training, as disclosed by Froloff, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would increase the accuracy of the model by ensuring that it has been properly trained. See Froloff, paragraph 120.
Regarding claim 19, Sathe, as modified by Froloff, discloses that “the weighting is selected based on a state of the one or more worker nodes (joint optimizer may select the top three predicted workers; once the joint optimizer has made enough evaluations to identify a best performing worker node, the joint optimizer may send all pipelines to the best performing worker node – Sathe, paragraph 37 [i.e., the weight of the best performing node is 1 and that of all others is 0, this selection being made based on the status of the best performing node as the best performing node]).”
Claims 2-3 are rejected under 35 U.S.C. 103 as being unpatentable over Sathe in view of Froloff and further in view of Douglas et al. (US 20190180841) (“Douglas”).
Regarding claim 2, the rejection of claim 1 is incorporated. Sathe further discloses “worker nodes”, as shown in the rejection of claim 1.
Neither Sathe nor Froloff appears to disclose explicitly the further limitations of the claim. However, Douglas discloses that “selecting the one or more … nodes comprises:
selecting a mask indicative of the one or more … nodes (in order for the network to approximate the marginal posteriors at test time, each sample Si is partially masked; the network receives as input a binary vector where a subset of the nodes initially observed were hidden, or masked – Douglas, paragraph 152).”
Douglas and the instant application both relate to machine learning and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Sathe and Froloff to select a mask indicative of nodes, as disclosed by Douglas, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would allow the model to produce data that are robust to the provision of incomplete information. See Douglas, paragraph 15.
Regarding claim 3, the rejection of claim 2 is incorporated. Sathe further discloses “worker nodes”, as shown in the rejection of claim 1.
Douglas discloses that “the mask is a binary vector comprising [one] value … to indicate the one or more … nodes and [another] value … to indicate any other … nodes of the plurality of … nodes (in order for the network to approximate the marginal posteriors at test time, each sample Si is partially masked; the network receives as input a binary vector where a subset of the nodes initially observed were hidden, or masked – Douglas, paragraph 152; see also Fig. 7, ref. chars. S503 and S505 (disclosing that the unmasked nodes take a value of 0 or 1 and the masked nodes take a value of *)).” Douglas further discloses values of 0 and 1 (Douglas Fig. 7, ref. chars S503, S505). It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Sathe and Froloff to employ a binary vector to mask certain nodes of the system, as disclosed by Douglas, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would allow the model to produce data that are robust to the provision of incomplete information. See Douglas, paragraph 15.
Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over Sathe in view of Froloff and further in view of Mohassel et al. (WO 2018174873) (“Mohassel”).
Regarding claim 5, neither Sathe nor Froloff appears to disclose explicitly the further limitations of the claim. However, Mohassel discloses that “the reward for the performance of the updated machine learning model is maximized if it is determined to be higher than a reward for a performance of the machine learning model in a previous round of training (after one epoch, the accuracy [reward for performance of the model] of the current set of weights is tested; if the accuracy decreases, the training starts over; otherwise the data are reshuffled and the next epoch of training is executed; then the difference in accuracy compared to the previous epoch is below a small threshold [i.e., the accuracy/reward is higher than in a previous round but not significantly higher], w converges to the minimum and the algorithm terminates [i.e., accuracy is thereby maximized] – Mohassel, paragraphs 74-75).”
Mohassel and the instant application both relate to machine learning and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Sathe and Froloff to maximize the reward by ensuring that it is higher than in previous epochs, as disclosed by Mohassel, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would ensure that the model is continually improving and not wasting resources on incremental improvements not affecting performance. See Mohassel, paragraphs 74-75.
Claims 8-10 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Sathe in view of Froloff and further in view of Lenc et al. (US 11676035) (“Lenc”).
Regarding claim 8, neither Sathe nor Froloff appears to disclose explicitly the further limitations of the claim. However, Lenc discloses that “the one or more parameters of the updated machine learning model comprise an aggregation of the one or more parameters of the machine learning model trained by the one or more worker nodes in the previous round of training (each worker can determine the fitnesses and then the updates to the distribution parameters [i.e., parameters obtained in a previous round of training] for the samples assigned to the worker in parallel with each other worker, i.e., instead of sending the fitnesses or utilities back to a central server; thus, the central server only needs to combine [aggregate], e.g., average or sum, the updates received from each worker and then apply the final update to the current distribution parameters – Lenc, col. 13, l. 66-col. 14, l. 6).”
Lenc and the instant application both relate to distributed machine learning and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Sathe and Froloff to aggregate the parameters obtained by the worker nodes, as disclosed by Lenc, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would exploit parallelism to allow learning to take place in less time than its serial counterpart. See Lenc, col. 13, l. 66-col. 14, l. 6.
Regarding claim 9, Sathe, as modified by Froloff and Lenc, discloses “aggregating the one or more parameters of the machine learning model trained by the one or more worker nodes in the previous round of training (each worker can determine the fitnesses and then the updates to the distribution parameters [i.e., parameters obtained in a previous round of training] for the samples assigned to the worker in parallel with each other worker, i.e., instead of sending the fitnesses or utilities back to a central server; thus, the central server only needs to combine [aggregate], e.g., average or sum, the updates received from each worker and then apply the final update to the current distribution parameters – Lenc, col. 13, l. 66-col. 14, l. 6).” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Sathe and Froloff to aggregate the parameters obtained by the worker nodes, as disclosed by Lenc, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would exploit parallelism to allow learning to take place in less time than its serial counterpart. See Lenc, col. 13, l. 66-col. 14, l. 6.
Regarding claim 10, Sathe, as modified by Froloff and Lenc, discloses that “the aggregation is an average (each worker can determine the fitnesses and then the updates to the distribution parameters [i.e., parameters obtained in a previous round of training] for the samples assigned to the worker in parallel with each other worker, i.e., instead of sending the fitnesses or utilities back to a central server; thus, the central server only needs to combine [aggregate], e.g., average or sum, the updates received from each worker and then apply the final update to the current distribution parameters – Lenc, col. 13, l. 66-col. 14, l. 6).” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Sathe and Froloff to average the parameters obtained by the worker nodes, as disclosed by Lenc, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would exploit parallelism to allow learning to take place in less time than its serial counterpart. See Lenc, col. 13, l. 66-col. 14, l. 6.
Regarding claim 16, Sathe, as modified by Froloff and Lenc, discloses that “the point of converge[nce] is reached when:
a predefined minimum number of training rounds is completed; and/or
an increase in the performance of the updated machine learning model for the validation dataset is less than a predefined threshold (system can repeatedly perform the process on different sets of training inputs from the training data to train the neural network; the system can perform the process until certain termination criteria are satisfied, e.g., until the performance of the neural network on the task has reached a threshold or until a threshold number of iterations of the process have been performed – Lenc, col. 9, ll. 9-17).” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Sathe and Froloff to continue the process until a threshold number of iterations has been reached, as disclosed by Lenc, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would reduce processor usage by providing a determinate stopping point for the training process. See Lenc, col. 9, ll. 9-17.
Claims 14 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Sathe in view of Froloff and further in view of Lan et al. (US 20220121465) (“Lan”).
Regarding claim 14, neither Sathe nor Froloff appears to disclose explicitly the further limitations of the claim. However, Lan discloses “initiating transmission of the one or more parameters of the machine learning model trained by the one or more worker nodes in the previous round of training towards the one or more worker nodes for the one or more worker nodes to further train the updated machine learning model (a method of processing flow includes reducers receiving gradients from all workers involved in the job; the reducer performs reduction operations and broadcasts [initiates transmission of] its results [parameters] to all workers – Lan, paragraph 86; worker processes may adjust weight coefficients in the context of training the neural network [i.e., the results are used for training] – id. at paragraph 10).”
Lan and the instant application both relate to distributed machine learning and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Sathe and Froloff to transmit parameters to worker nodes for further training, as disclosed by Lan, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would allow the data processing to be distributed across multiple machines, thereby speeding up training. See Lan, paragraph 3.
Regarding claim 17, neither Sathe nor Froloff appears to disclose explicitly the further limitations of the claim. However, Lan discloses that “prior to selecting the one or more worker nodes:
initiating transmission of one or more parameters of the machine learning model towards the one or more worker nodes for the one or more worker nodes to train the machine learning model in the previous round of training (a method of processing flow includes reducers receiving gradients from all workers involved in the job; the reducer performs reduction operations and broadcasts [initiates transmission of] its results [parameters] to all workers – Lan, paragraph 86; worker processes may adjust weight coefficients in the context of training the neural network [i.e., the results are used for training] – id. at paragraph 10; see also Fig. 10 (showing that the gradients are received from all workers at each iteration, i.e., all worker nodes are selected for the following iteration after the result is broadcast to the workers in the previous iteration)); and
receiving the one or more parameters of the machine learning model trained by the one or more worker nodes in the previous round of training from the one or more worker nodes (after broadcasting results to workers, a reducer determines whether processing is terminated; if processing is not terminated, the reducers performs another iteration of processing that includes receiving the gradients [parameters] from all workers – Lan, paragraph 86).” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Sathe and Froloff to transmit parameters to worker nodes for further training, as disclosed by Lan, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would allow the data processing to be distributed across multiple machines, thereby speeding up training. See Lan, paragraph 3.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to RYAN C VAUGHN whose telephone number is (571)272-4849. The examiner can normally be reached M-R 7:00a-5:00p ET.
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, Kamran Afshar, can be reached at 571-272-7796. 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.
/RYAN C VAUGHN/ Primary Examiner, Art Unit 2125
Prosecution Insights