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 .
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 25 are rejected under 35 U.S.C. 101 because the claimed invention is directed to and abstract idea without significantly more.
Step 1
According to the first part of the analysis, in the instant case, claims 1-19 are directed to a method and claim 25 is directed to an apparatus comprising at least a processor. Thus, each of the claims falls within one of the four statutory categories (i.e. process, machine, manufacture, or composition of matter).
Claim 1 recites:
Step 2A, prong 1
“ii) comparing the aggregated characteristic to an equivalent reference” (This step is a recitation of a mental process that is practical to perform in the human mind (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.).
“iii) identifying whether the first subset of computing nodes are contributing updates that are corrupting the machine learning model, based on the comparison” (This step is a recitation of a mental process that is practical to perform in the human mind (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.).
Step 2A, prong 2
“A computer implemented method for use in a distributed machine learning process for training a machine learning model, wherein the training is distributed across a plurality of computing nodes and updates to the machine learning model, as determined by the plurality of computing nodes, are aggregated using secure multi party computation” (mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
“i) obtaining an aggregated characteristic of updates to the machine learning model provided by a first subset of the plurality of computing nodes” (insignificant extra-solution activity)
This judicial exception is not integrated into a practical application.
Step 2B
“A computer implemented method for use in a distributed machine learning process for training a machine learning model, wherein the training is distributed across a plurality of computing nodes and updates to the machine learning model, as determined by the plurality of computing nodes, are aggregated using secure multi party computation” (mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
“i) obtaining an aggregated characteristic of updates to the machine learning model provided by a first subset of the plurality of computing nodes” (This step appears to be directed to transmitting or receiving information, which is well-understood, routine, and conventional. i. Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information); See MPEP 2106.05 (d) (II).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claim 2 recites:
Step 2A, prong 1
“selecting the first subset of nodes, and wherein the first subset of nodes are selected such that their respective masks, associated with the secure multi party computation, cancel each other out when aggregated” (This step is a recitation of a mental process that is practical to perform in the human mind (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.).
Step 2A, Prong 2 & 2B
The claim does not recite any additional elements.
Claim 3 recites:
Step 2A, prong 1
“wherein the secure multi party computation uses a groupwise secret sharing masking process and the step of selecting the first subset of nodes comprises:selecting a group of nodes from the groupwise secret sharing masking process as the first subset of nodes” (This step is a recitation of a mental process that is practical to perform in the human mind (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.)
Step 2A, Prong 2 & 2B
The claim does not recite any additional elements.
Claim 4 recites:
Step 2A, prong 1
“wherein the aggregated characteristic comprises a measure of convergence or accuracy of the machine learning model when the updates to the machine learning model from the first subset of nodes are aggregated into the machine learning model” (This step is a recitation of a mental process that is practical to perform in the human mind (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.)
Step 2A, Prong 2 & 2B
The claim does not recite any additional elements.
Claim 5 recites:
Step 2A, prong 1
“a measure of convergence or accuracy of the machine learning model when updates from the first subset of nodes are not aggregated into the machine learning model; a measure of convergence or accuracy of the machine learning model as determined using a trusted dataset; or a measure of convergence or accuracy of the machine learning model as determined from updates to the machine learning model provided by a second subset of the plurality of computing nodes” (This step is a recitation of a mental process that is practical to perform in the human mind (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.)
Step 2A, Prong 2 & 2B
The claim does not recite any additional elements.
Claim 6 recites:
Step 2A, prong 1
“wherein the step of identifying whether the first subset of nodes are contributing updates that are corrupting the machine learning model comprises:
determining that the first subset of nodes are contributing updates that are corrupting the machine learning model if the aggregated characteristic indicates that the machine learning model converges faster or is more accurate when updates from the first subset of nodes are not aggregated into the machine learning model compared to when updates from the first subset of nodes are aggregated into the machine learning model” (This step is a recitation of a mental process that is practical to perform in the human mind (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.)
Step 2A, Prong 2 & 2B
The claim does not recite any additional elements.
Claim 7 recites:
Step 2A, prong 1
“wherein the aggregated characteristic comprises an aggregated parameter value obtained using an explainable AI, XAI, process and the equivalent reference is a ground truth value for said parameter” (This step is a recitation of a mental process that is practical to perform in the human mind (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.)
Step 2A, Prong 2 & 2B
The claim does not recite any additional elements.
Claim 8 recites:
Step 2A, prong 1
“wherein the parameter value obtained using the XAI process is a measure of feature importance of an input feature to the machine learning model” (This step is a recitation of a mental process that is practical to perform in the human mind (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.)
Step 2A, Prong 2 & 2B
The claim does not recite any additional elements.
Claim 9 recites:
Step 2A, prong 1
“wherein the step of identifying whether the first subset of nodes are contributing updates that are corrupting the machine learning model comprises: determining that the first subset of nodes are contributing updates that are corrupting the machine learning model if the feature importance as determined by the first subset of nodes is different to the ground truth feature importance.” (This step is a recitation of a mental process that is practical to perform in the human mind (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.)
Step 2A, Prong 2 & 2B
The claim does not recite any additional elements.
Claim 10 recites:
Step 2A, prong 1
“determining that the first subset of nodes are not contributing updates that are corrupting the machine learning model; and repeating steps i)-iii) in an iterative manner for other subsets of nodes.” (This step is a recitation of a mental process that is practical to perform in the human mind (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.)
Step 2A, Prong 2 & 2B
The claim does not recite any additional elements.
Claim 11 recites:
Step 2A, prong 1
“wherein the method is performed responsive to detecting a reduction in performance of the machine learning model” (This step is a recitation of a mental process that is practical to perform in the human mind (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.)
Step 2A, Prong 2 & 2B
The claim does not recite any additional elements.
Claim 12 recites:
Step 2A, prong 1
“wherein the first subset of nodes are selected from nodes associated with a common aggregation point in the distributed machine learning process” (This step is a recitation of a mental process that is practical to perform in the human mind (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.)
Step 2A, Prong 2 & 2B
The claim does not recite any additional elements.
Claim 13 recites:
Step 2A, prong 1
“wherein the first subset of nodes are selected from nodes associated with different aggregation points in the distributed machine learning process” (This step is a recitation of a mental process that is practical to perform in the human mind (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.)
Step 2A, Prong 2 & 2B
The claim does not recite any additional elements.
Claim 14 recites:
Step 2A, prong 1
Claim 14 recites at least the abstract idea identified above in claim 1.
Step 2A, Prong 2
“quarantining the first subset of nodes from the distributed machine learning process, if the first subset of nodes are identified as contributing updates that are corrupting the machine learning model” (linking judicial exception to a field of use. See MPEP 2106.05(h).)
This judicial exception is not integrated into a practical application.
Step 2B
“quarantining the first subset of nodes from the distributed machine learning process, if the first subset of nodes are identified as contributing updates that are corrupting the machine learning model” (linking judicial exception to a field of use. See MPEP 2106.05(h).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claim 15 recites:
Step 2A, prong 1
Claim 15 recites at least the abstract idea identified above in claim 1.
Step 2A, Prong 2
“wherein the method is performed by a first node in a communications network” (linking judicial exception to a field of use. See MPEP 2106.05(h).)
This judicial exception is not integrated into a practical application.
Step 2B
“wherein the method is performed by a first node in a communications network” (linking judicial exception to a field of use. See MPEP 2106.05(h).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claim 16 recites:
Step 2A, prong 1
Claim 16 recites at least the abstract idea identified above in claim 1.
Step 2A, Prong 2
“wherein the first node is configured to send a message to a second node, the second node being an aggregation point in the distributed machine learning process for the plurality of computing nodes, and wherein the message instructs the second node to determine the characteristic for the first subset of nodes” (insignificant extra-solution activity)
This judicial exception is not integrated into a practical application.
Step 2B
“wherein the first node is configured to send a message to a second node, the second node being an aggregation point in the distributed machine learning process for the plurality of computing nodes, and wherein the message instructs the second node to determine the characteristic for the first subset of nodes” (This step appears to be directed to transmitting or receiving information, which is well-understood, routine, and conventional. i. Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information); See MPEP 2106.05 (d) (II).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claim 17 recites:
Step 2A, prong 1
“determine the characteristic for other nodes in the plurality of computing nodes that are not in the first subset of nodes; and wherein the characteristic for the other nodes in the plurality of computing nodes is used as the equivalent reference” (This step is a recitation of a mental process that is practical to perform in the human mind (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.)
Step 2A, Prong 2 & 2B
The claim does not recite any additional elements.
Claim 18 recites:
Step 2A, prong 1
“identifying nodes in the plurality of computing nodes that are performing a data poison attack” (This step is a recitation of a mental process that is practical to perform in the human mind (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.)
Step 2A, Prong 2 & 2B
The claim does not recite any additional elements.
Claim 19 recites:
Step 2A, prong 1
“determining actions that should be performed in a safety critical system” (This step is a recitation of a mental process that is practical to perform in the human mind (i.e., observation, evaluation, judgement, opinion). See MPEP § 2106.04(a)(2), subsection III.)
Step 2A, prong 2
“wherein the method is used in training the machine learning model for use in a determining actions that should be performed in a safety critical system” (mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
This judicial exception is not integrated into a practical application.
Step 2B
“wherein the method is used in training the machine learning model for use in a determining actions that should be performed in a safety critical system” (mere instructions to apply the exception using a generic computer component. See 2106.05(f).)
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claim 25 recites: See the rejection of claim 1 above. Same rationale applies.
2A Prong 2 & 2B: The claim recites another additional element “An apparatus, a memory comprising instruction data representing a set of instructions ; and a processor configured to communicate with the memory and to execute the set of instructions” (mere instructions to apply the exception using a generic computer component)
Claim Rejections - 35 USC § 102
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claims 1, 10-19, and 25 are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Sharad et al. (US-20200285980-A1).
Regarding Claim 1,
Sharad (US 20200285980 A1) teaches a computer implemented method for use in a distributed machine learning process for training a machine learning model, wherein the training is distributed across a plurality of computing nodes and updates to the machine learning model, as determined by the plurality of computing nodes (para [0018]-[0019] participants and server (i.e., nodes) update ML model.), are aggregated using secure multi party computation (para [0013] “The federated learning system is a collaborative machine learning method which alleviates privacy and security concerns by performing the machine learning training process in a distributed manner, without the need of centralizing private data.”), the method comprising:
i) obtaining an aggregated characteristic of updates to the machine learning model provided by a first subset of the plurality of computing nodes (para [0020], [0040]);
ii) comparing the aggregated characteristic to an equivalent reference (para [0041]-[0042]); and
iii) identifying whether the first subset of computing nodes are contributing updates that are corrupting the machine learning model, based on the comparison (para [0041]-[0042]).
Regarding Claim 10,
Sharad teaches a method as in claim 1 further comprising:
determining that the first subset of nodes are not contributing updates that are corrupting the machine learning model (para [0041] “Upon completion of this procedure, the server is left with a subset of aggregated updates {AU′.sup.1, . . . , AU′.sup.g′}⊂{AU.sup.1, . . . , AU.sup.g}(g′≤g), for which the server has a higher level of confidence were not submitted by a malicious actor.”); and
repeating steps i)-iii) in an iterative manner for other subsets of nodes (para [0036] “FIG. 3 illustrates a process for performing federated learning iterations according to an embodiment. The training process consists of a number of rounds R, as illustrated in FIG. 2.”).
Regarding Claim 11,
Sharad teaches a method as in claim 1 wherein the method is performed responsive to detecting a reduction in performance of the machine learning model (para [0048] “The model makes predictions to complete sentences typed by users. In this scenario a malicious device may try to poison the model such that a chosen business or restaurant is suggested when someone searches for food in a particular street. For instance, “best burger in 5th Avenue . . . ” can be completed by inputting a chosen business, rather than the business with the best burger. In such a scenario it is important to make the model robust for people to trust it.”).
Regarding Claim 12,
Sharad teaches a method as in claim 1 wherein the first subset of nodes are selected from nodes associated with a common aggregation point in the distributed machine learning process (para [0050] “At step 504, The server selects n out of N participants, according to some filtering criteria, to contribute the training for round r. At step 506 the server partitions the selected contributors into s groups. At step 508, the server informs each contributor about the other participants belonging to the same group.”).
Regarding Claim 13,
Sharad teaches a method as in claim 1 wherein the first subset of nodes are selected from nodes associated with different aggregation points in the distributed machine learning process (para [0019] “In the same or other embodiment, the aggregated group updates are obtained from a locally trained model L_i using local training data by each participant; relevant updates U_i needed to obtain a local model L_i from the previous global model G{circumflex over ( )}(r−1) are generated by each participant;”).
Regarding Claim 14,
Sharad teaches a method as in claim 1 further comprising:
quarantining the first subset of nodes from the distributed machine learning process, if the first subset of nodes are identified as contributing updates that are corrupting the machine learning model (para [0030] “combining the aggregated group updates by excluding the updates identified as suspicious, to obtain an aggregated update U{circumflex over ( )}final;”).
Regarding Claim 15,
Sharad teaches a method as in claim 1 wherein the method is performed by a first node in a communications network (para [0051] “The network interface 612 may connect to a wired network or cellular network and to a local area network or wide area network, such as the internet. Further, the network interface 612 may include a transmitter and a receiver to implement Optical-Wireless-Communication links as described above.”).
Regarding Claim 16,
Sharad teaches a method as in claim 15 wherein the first node is configured to send a message to a second node, the second node being an aggregation point in the distributed machine learning process for the plurality of computing nodes, and wherein the message instructs the second node to determine the characteristic for the first subset of nodes (para [0018] “obtaining, by the server, aggregated group updates AU{circumflex over ( )}1, . . . , AU{circumflex over ( )}g from each group; comparing, by the server, the aggregated group updates and identifying suspicious aggregated group updates; combining, by the server, the aggregated group updates by excluding the updates identified as suspicious, to obtain an aggregated update U{circumflex over ( )}final;”).
Regarding Claim 17,
Sharad teaches a method as in claim 16 wherein the message further instructs the second node to
determine the characteristic for other nodes in the plurality of computing nodes that are not in the first subset of nodes (para [0028] “partitioning the selected participants n into s groups; informing each participant about the other participants belonging to the same group; obtaining aggregated group updates AU{circumflex over ( )}1, . . . , AU{circumflex over ( )}g from each group; comparing the aggregated group updates and identifying suspicious aggregated group updates; combining the aggregated group updates by excluding the updates identified as suspicious, to obtain an aggregated update U{circumflex over ( )}final;”); and
wherein the characteristic for the other nodes in the plurality of computing nodes is used as the equivalent reference (para [0041]).
Regarding Claim 18,
Sharad teaches a method as in claim 1 wherein the method is for use in identifying nodes in the plurality of computing nodes that are performing a data poison attack (para [0048] “The model makes predictions to complete sentences typed by users. In this scenario a malicious device may try to poison the model such that a chosen business or restaurant is suggested when someone searches for food in a particular street.”).
Regarding Claim 19,
Sharad teaches method as in claim 1 wherein the method is used in training the machine learning model for use in a determining actions that should be performed in a safety critical system (para [0050] “Next, at step 518 the server combines the aggregated group updates by excluding, or reducing the impact of, the updates identified as suspicious, obtaining an aggregated update U.sup.final.”).
Regarding Claim 25,
Claim 25 is the apparatus corresponding to the method of claim 1. Claim 25 is substantially similar to claim 1 and is rejected on the same grounds.
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.
Claims 2-3 are rejected under 35 U.S.C. 103 as being unpatentable over Sharad et al. (US-20200285980-A1) in view of Gama et al. (US-20200304293-A1).
Regarding Claim 2,
Sharad teaches a method as in claim 1.
Sharad does not explicitly disclose
further comprising selecting the first subset of nodes, and wherein the first subset of nodes are selected such that their respective masks, associated with the secure multi party computation, cancel each other out when aggregated.
However, Gama (US 20200304293 A1) teaches
further comprising selecting the first subset of nodes, and wherein the first subset of nodes are selected such that their respective masks, associated with the secure multi party computation, cancel each other out when aggregated (para [0135] “They also generate secret shares of a common mask, that they share between each other via the broadcast channel, but which remains secret to the dealer. The players then mask their secret shares with the common mask and send them to the dealer, who evaluates the non-linear parts (product in the first method and power in the second method). The dealer generates new additive shares for the result and sends these values back to each player via the private channel. This way, the players don't know each other's shares. Finally, the players, who know the common mask, can independently unmask their secret shares, and obtain their final share of the numerical masking data, which is therefore unknown to the dealer.”).
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the distributed machine learning model of Sharad with the masks of Gama.
Doing so would allow for performing privacy-preserving or secure multi-party computations enables multiple parties to collaborate to produce a shared result while preserving the privacy of input data contributed by individual parties (Gama abs.).
Regarding Claim 3,
Sharad and Gama teach a method as in claim 2. Gama further teaches wherein the secure multi party computation uses a groupwise secret sharing masking process and the step of selecting the first subset of nodes comprises:
selecting a group of nodes from the groupwise secret sharing masking process as the first subset of nodes (para [0135] “They also generate secret shares of a common mask, that they share between each other via the broadcast channel, but which remains secret to the dealer. The players then mask their secret shares with the common mask and send them to the dealer, who evaluates the non-linear parts (product in the first method and power in the second method). The dealer generates new additive shares for the result and sends these values back to each player via the private channel. This way, the players don't know each other's shares. Finally, the players, who know the common mask, can independently unmask their secret shares, and obtain their final share of the numerical masking data, which is therefore unknown to the dealer.”).
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the distributed machine learning model of Sharad with the masks of Gama.
Doing so would allow for performing privacy-preserving or secure multi-party computations enables multiple parties to collaborate to produce a shared result while preserving the privacy of input data contributed by individual parties (Gama abs.).
Claims 4-6 are rejected under 35 U.S.C. 103 as being unpatentable over Sharad et al. (US-20200285980-A1) in view of Shaloudegi et al. (US-20220237508-A1).
Regarding Claim 4,
Sharad teaches a method as in claim 1.
Sharad does not explicitly disclose
wherein the aggregated characteristic comprises a measure of convergence or accuracy of the machine learning model when the updates to the machine learning model from the first subset of nodes are aggregated into the machine learning model.
However, Shaloudegi (US 20220237508 A1)
wherein the aggregated characteristic comprises a measure of convergence or accuracy of the machine learning model when the updates to the machine learning model from the first subset of nodes are aggregated into the machine learning model (para [0118] “The aggregation operation then begins a new iteration: the aggregation and update block 620 performs the first step to compute x.sub.j+1 by using the information obtained from the client nodes 102. The steps of the aggregation operation may be iterated until a convergence condition is satisfied, thereby ending the round of training. The convergence condition may be defined based on the values or gradients of the global learned parameters, based on a performance metric, or based on a maximum threshold for iterations, time, communication cost, or some other resource being reached.”).
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the distributed machine learning model of Sharad with the convergence condition of Shaloudegi.
Doing so would allow for minimizing the communication cost, in order to reduce the use of network resources, processing resources and/or monetary costs (e.g., the monetary cost associated with network use), thereby improving the functioning of the system (Shaloudegi para [0083]).
Regarding Claim 5,
Sharad and Shaloudegi teach a method as in claim 4.
Shaloudegi further teaches
wherein the equivalent reference comprises:
a measure of convergence or accuracy of the machine learning model when updates from the first subset of nodes are not aggregated into the machine learning model;
a measure of convergence or accuracy of the machine learning model as determined using a trusted dataset (para [0006] “In the specific context of FL, averaging approaches such as FedAvg may attempt to account for the bias described above using two techniques: first, client nodes may be configured to not fully fit their local models to the respective local datasets (i.e., local learned parameter values are not learned locally to the point of convergence),”); or
a measure of convergence or accuracy of the machine learning model as determined from updates to the machine learning model provided by a second subset of the plurality of computing nodes.
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the distributed machine learning model of Sharad with the convergence condition of Shaloudegi.
Doing so would allow for minimizing the communication cost, in order to reduce the use of network resources, processing resources and/or monetary costs (e.g., the monetary cost associated with network use), thereby improving the functioning of the system (Shaloudegi para [0083]).
Regarding Claim 6,
Sharad teaches a method as in claim 1.
Sharad does not explicitly disclose
wherein the step of identifying whether the first subset of nodes are contributing updates that are corrupting the machine learning model comprises:
determining that the first subset of nodes are contributing updates that are corrupting the machine learning model if the aggregated characteristic indicates that the machine learning model converges faster or is more accurate when updates from the first subset of nodes are not aggregated into the machine learning model compared to when updates from the first subset of nodes are aggregated into the machine learning model.
However, Shaloudegi (US 20220237508 A1)
determining that the first subset of nodes are contributing updates that are corrupting the machine learning model if the aggregated characteristic indicates that the machine learning model converges faster or is more accurate when updates from the first subset of nodes are not aggregated into the machine learning model compared to when updates from the first subset of nodes are aggregated into the machine learning model (para [0118] “The steps of the aggregation operation may be iterated until a convergence condition is satisfied, thereby ending the round of training. The convergence condition may be defined based on the values or gradients of the global learned parameters, based on a performance metric, or based on a maximum threshold for iterations, time, communication cost, or some other resource being reached. In some embodiments, changes in the value of (Equation 10) are monitored by the aggregation and update block 620; if the changes in two consecutive iterations (or over several consecutive iterations) of the aggregation operation are below a threshold, the current round of training is terminated.”).
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the distributed machine learning model of Sharad with the convergence condition of Shaloudegi.
Doing so would allow for minimizing the communication cost, in order to reduce the use of network resources, processing resources and/or monetary costs (e.g., the monetary cost associated with network use), thereby improving the functioning of the system (Shaloudegi para [0083]).
Claim(s) 7-9 are rejected under 35 U.S.C. 103 as being unpatentable over Sharad et al. (US-20200285980-A1) in view of Dalli et al. (US-20210350211-A1).
Regarding Claim 7,
Sharad teaches a method as in claim 1.
Sharad does not explicitly disclose
wherein the aggregated characteristic comprises an aggregated parameter value obtained using an explainable AI, XAI, process and the equivalent reference is a ground truth value for said parameter.
However, Dalli (US 20210350211 A1) teaches
wherein the aggregated characteristic comprises an aggregated parameter value obtained using an explainable AI, XAI, process and the equivalent reference is a ground truth value for said parameter (para [0083] “The trace path analysis may be in the form of a backmap process whereby the output of the neural network is projected back to the input in order to analyze and perform an impact assessment of the partition, feature importance, and data in the explainable model and data via human knowledge injection (HKI) processes, against a number of criteria and thresholds and values set against those criteria.”).
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the distributed machine learning model of Sharad with the model explainability of Dalli.
Doing so would allow for providing information to improve the data collection process or fix the weakness in the data and enhance the resulting model performance with better generalization (Dalli para [0064]).
Regarding Claim 8,
Sharad and Dalli teach a method as in claim 7.
However, Dalli (US 20210350211 A1) teaches
wherein the parameter value obtained using the XAI process is a measure of feature importance of an input feature to the machine learning model (para [0083] “The trace path analysis may be in the form of a backmap process whereby the output of the neural network is projected back to the input in order to analyze and perform an impact assessment of the partition, feature importance, and data in the explainable model and data via human knowledge injection (HKI) processes, against a number of criteria and thresholds and values set against those criteria.”).
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the distributed machine learning model of Sharad with the model explainability of Dalli.
Doing so would allow for providing information to improve the data collection process or fix the weakness in the data and enhance the resulting model performance with better generalization (Dalli para [0064]).
Regarding Claim 9,
Sharad and Dalli a method as in claim 8. Dalli further teaches wherein the step of identifying whether the first subset of nodes are contributing updates that are corrupting the machine learning model comprises:
determining that the first subset of nodes are contributing updates that are corrupting the machine learning model if the feature importance as determined by the first subset of nodes is different to the ground truth feature importance (para [0083] “If the impact assessment concludes that such data points will result into different model behavior, various mitigation strategies may be applied. For example, one mitigation strategy may involve updating weights to minimize or take out a path without the need for re-training”).
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the distributed machine learning model of Sharad with the model explainability of Dalli.
Doing so would allow for providing information to improve the data collection process or fix the weakness in the data and enhance the resulting model performance with better generalization (Dalli para [0064]).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to HENRY K NGUYEN whose telephone number is (571)272-0217. The examiner can normally be reached Mon - Fri 7:00am-4:30pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Li B Zhen can be reached at 5712723768. 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.
/HENRY NGUYEN/Examiner, Art Unit 2121
Prosecution Insights