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 .
DETAILED ACTION
The instant application having Application No. 17205632 has a total of 21 claims pending in the application, of which claim 9 has been cancelled.
Claim Rejections - 35 USC § 103
Claims 1, 6-7, 9, 11-12, 16, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Szeto et al (US 20180018590 A1) in view of Guo et al (US 20210099312 A1), Tran et al (“Federated Learning over Wireless Networks: Optimization Model Design and Analysis”) and Russinovich (US 20180227275 A1).
As per claim 1, Szeto discloses, “A first training node comprising:” (Pg.2-3, particularly paragraph 0018; EN: this denotes the system that performs training).
“A processor” (Pg.2-3, particularly paragraph 0018; EN: this denotes the hardware needed to run the system).
“a memory unit operatively connected to the processor, the memory unit including instructions that when executed, cause the processor to” (Pg.2-3, particularly paragraph 0018; EN: this denotes the hardware needed to run the system).
“initiate a training process of a local version of a machine learning (ML) model at the first training node”, and “training process” (Pg.2-3, particularly paragraph 0018; EN: this denotes training a proxy model on the private server (i.e. a local model)).
“elect a second training node to act as a … leader node during the training process” (Pg.3, particularly paragraph 0019; EN: this denotes transmitting the features of the model to another system to create an aggregated model. This system will be the “leader node.”).
“Encrypt” (Pg.8, particularly paragraph 0059; EN: this denotes encrypting communications) “… local parameters derived from the training of the local version of the ML model” (Pg.3, particularly paragraph 0018; EN: this denotes transmitting the features of the model to a non-private server to create an aggregated model).
“transmit the local parameters to the … leader node” (Pg.3, particularly paragraph 0018; EN: this denotes transmitting the features of the model to a non-private server to create an aggregated model).
“receive, from the … leader node, merged parameters derived from a global version of the ML model” (Pg.6, particularly paragraph 0050; EN: this denotes the global model/server returning parameters to the private servers (i.e. the local models) to improve their models).
“apply the merged parameters to the local version of the ML model at the first training node to update the local version of the ML model” (Pg.6, particularly paragraph 0050; EN: this denotes using the global parameters to improve the local model).
“Evaluate the updated local version of the ML model to determine a … validation… value” and “validation … value” (Pg.6, particularly paragraph 0050; EN: this denotes checking to see if the local model improves or worsens. Pg.14, particularly paragraph 0098; EN: this denotes breaking up of training data into training and validation sets).
“…the local parameters…” (Pg.2-3, particularly paragraph 0018; EN: this denotes training a proxy model on the private server (i.e. a local model)).
“determines, based on the global validation … value, whether to continue the training process of the local version of the ML model” (Pg.11, particularly paragraph 0081; EN: this denotes using the incoming global model as a base-line to further train the local model if it wishes to additional training).
“and based on the determination of whether to continue the training process, selectively continues the training process of the local version of the ML model” (Pg.11, particularly paragraph 0081; EN: this denotes using the incoming global model as a base-line to further train the local model if it wishes to do additional training).
However, Szeto fails to explicitly disclose, “temporary leader node”, “wherein the election initiates a transmission of a public key from a key manager to the temporary leader node”, “receive, from the temporary leader node, the public key”, “determine a local … loss value” and “transmit the local …loss value to the temporary leader node”, and “receive, from the temporary leader node, a global … loss value calculated and distributed by the temporary leader node based on the local … loss value wherein in response to receiving the global validation loss value from the temporary leader node, the first training node: instructs the key manager to discard the public key associated with encrypting the local parameters” and “based on the global … loss value.”
Guo discloses, “temporary leader node” (abstract; EN: this denotes the system selecting new leader nodes, making any leader node temporary).
“wherein the election initiates a transmission of a public key from a key manager” (Pg.6-7, particularly paragraph 0058; EN: this denotes keyblock which contains the public key for the validators for the current epoch). “to the temporary leader node” (pg.11, particularly paragraph 0095; EN: this denotes the Leader node working with the public keys generated by the keyblock).
“receive, from the temporary leader node, the public key” (pg.10, particularly paragraph 0087; EN: this denotes common nodes receiving the new public epoch key from the leader node).
“Encrypt, using the public key….” (pg.7, particularly paragraph 0062; EN: this denotes the system using public keys to encrypt).
“wherein in response to receiving the … , the first training node: instructs the key manager to discard the public key associated with encrypting…” (pg.11, particularly paragraph 0095; EN: this denotes the creation of a new public epoch key to be used going forward, which discards the previous public key to no longer be used).
“relinquishes acting as the temporary leader node” (pg.11, particularly paragraph 0095; EN: this denotes the creation of a new public epoch key to be used going forward, which discards the previous public key to no longer be used along with selecting a new leader node).
“returns to acting as the second training node, wherein the election of the second training node as the temporary leader node ensures privacy by changing ownership of the public key” (pg.11, particularly paragraph 0095; EN: this denotes the creation of a new public epoch key to be used going forward, which discards the previous public key to no longer be used along with selecting a new leader node).
Tran discloses, “determine a local … loss value” (Pg.1388, particularly paragraph C2, section III; EN: this denotes the loss function of individual clients (i.e. local models)).
“transmit the local … loss value to the …leader node” (Pg.1389, particularly C1, Communication; EN: this denotes transmitting the data to BS, the base station, which would be the leader node similar to the global model of the Szeto reference).
“receive from the …leader node, a global … loss value calculated and distributed by the … leader node based on the local … loss value” and “global … loss” (Pg.1388, Particularly C2, Section III; EN: This denotes combining the loss functions of the clients to get a combined (i.e. global) loss function).
“based on the global … loss value” (Pg.1388, Particularly C2, Section III; EN: This denotes combining the loss functions of the clients to get a combined (i.e. global) loss function).
Russinovich discloses, “the first training node instructs the key manager to discard the public key…” (Pg.10, particularly paragraph 0105; EN: The new claim amendments require that the first training node, and not the leader, request the changing of the public key. Russinovich discloses that any member may request a new public key as needed).
Szeto and Guo are analogous art because both involve blockchain.
Before the effective filing date it would have been obvious to one skilled in the art of blockchain to combine the work of Szeto and Guo in order to change public keys with different leaders.
The motivation for doing so would be to “encrypt… using the current epoch public key, and transmitted to the current leader node, who can validate the encrypted new-epoch keys using the current epoch private key” (Guo, Pg.11, paragraph 0095) or in the case of Szeto, allow the use of public keys among the nodes to ensure privacy of the encrypted data as needed.
Therefore before the effective filing date it would have been obvious to one skilled in the art of blockchain to combine the work of Szeto and Guo in order to change public keys with different leaders.
Szeto and Tran are analogous art because both involve distributed machine learning.
Before the effective filing date it would have been obvious to one skilled in the art of distributed machine learning to combine the work of Szeto and Tran in order to use and share loss functions between the local and global models.
The motivation for doing so would be to “find the model parameter w that characterizes the output y … with the loss function” (Tran, Pg.1388, C2, Section III, second paragraph) or in the case of Szeto, allow the system to use loss functions to train the local/global models, and share that information among the local/global system.
Therefore before the effective filing date it would have been obvious to one skilled in the art of distributed machine learning to combine the work of Szeto and Tran in order to use and share loss functions between the local and global models.
Szeto and Russinovich are analogous art because both involve blockchain.
Before the effective filing date it would have been obvious to one skilled in the art of blockchain to combine the work of Szeto and Russinovich in order to allow any member of the blockchain to request a new public key.
The motivation for doing so would be to cause “the network [to] not accept any transactions referencing the old KBK” (Russinovich, Pg.10, paragraph 0105) or in the case of Szeto, allow the system to have members
Therefore before the effective filing date it would have been obvious to one skilled in the art of blockchain to combine the work of Szeto and Russinovich in order to allow any member of the blockchain to request a new public key.
As per claim 6, Szeto discloses, “validation … value” (Pg.6, particularly paragraph 0050; EN: this denotes checking to see if the local model improves or worsens. Pg.14, particularly paragraph 0098; EN: this denotes breaking up of training data into training and validation sets).
Guo discloses, “temporary leader node” (abstract; EN: this denotes the system selecting new leader nodes, making any leader node temporary).
Tran discloses, “wherein the memory unit includes instructions that when executed further cause the processor to transmit the local … loss value to the … leader node” (Pg.1389, particularly C1, Communication; EN: this denotes transmitting the data to BS, the base station, which would be the leader node similar to the global model of the Szeto reference).
As per claim 7, Szeto discloses, “validation … value” (Pg.6, particularly paragraph 0050; EN: this denotes checking to see if the local model improves or worsens. Pg.14, particularly paragraph 0098; EN: this denotes breaking up of training data into training and validation sets).
Guo discloses, “temporary leader node” (abstract; EN: this denotes the system selecting new leader nodes, making any leader node temporary).
Tran discloses, “wherein the memory unit includes instructions that when executed further cause the processor to receive an averaged … loss value from the leader node” (Pg.1388, Particularly C2, Section III; EN: This denotes combining the loss functions of the clients to get a combined (i.e. global) loss function) The equation denotes this to be an average of the various local loss functions).
As per claim 9, Tran discloses, “wherein the memory unit includes instructions that when executed further cause the processor to one of continue the training of the local version of the ML model or end the training of the local version of the ML model based on the global validation loss value” (Pg.1389, particularly C1, Communication section; EN: this denotes continuing the process until a global accuracy is achieved).
As per claim 11, Szeto discloses, “A system comprising a training node, the training node comprising” (Pg.2-3, particularly paragraph 0018; EN: this denotes the system that performs training).
“A processor” (Pg.2-3, particularly paragraph 0018; EN: this denotes the hardware needed to run the system).
“a memory unit operatively connected to the processor, the memory unit including instructions that when executed, cause the processor to” (Pg.2-3, particularly paragraph 0018; EN: this denotes the hardware needed to run the system).
“Initiate a training process of a local version of a machine learning (ML) model” and “training process” (Pg.2-3, particularly paragraph 0018; EN: this denotes training a proxy model on the private server (i.e. a local model)).
“receive an election to act as a … leader node to other training nodes; (Pg.3, particularly paragraph 0018; EN: this denotes transmitting the features of the model to another system to create an aggregated model. This system will be the “leader node.”).
“receive, from the other training nodes, local parameters derived from training of respective local version of the mL model at the other training nodes” (Pg.2-3, particularly paragraph 0018; EN: this denotes training a proxy model on the private server (i.e. a local model, with a plurality of different private servers).
“… encrypting” (Pg.8, particularly paragraph 0059; EN: this denotes encrypting communications) “the received local parameters of the other training nodes” (Pg.3, particularly paragraph 0018; EN: this denotes transmitting the features of the model to a non-private server to create an aggregated model).
“Merge the received local parameters” (Pg.3, particularly paragraph 0019; EN: this denotes transmitting the features of the model to a to create an aggregated model).
“Build a global version of the ML model using the merged local parameters” (Pg.3, particularly paragraph 0019; EN: this denotes transmitting the features of the model to create an aggregated model).
“transmit the merged local parameters to each of the other training nodes” (Pg.6, particularly paragraph 0050; EN: this denotes the global model/server returning parameters to the private servers (i.e. the local models) to improve their models).
“receive from each of the other training nodes, local validation … values derived from local evaluation of the respective local version of the ML model” and “validation … value” (Pg.6, particularly paragraph 0050; EN: this denotes checking to see if the local model improves or worsens. Pg.14, particularly paragraph 0098; EN: this denotes breaking up of training data into training and validation sets).
“in response to receiving the local validation… values form the other training nodes;” (Pg.3, particularly paragraph 0019; EN: this denotes transmitting the features of the model to a to create an aggregated model).
“Calculate, based on the local validation … values, a global validation … value” (Pg.3, particularly paragraph 0019; EN: this denotes transmitting the features of the model to create an aggregated model).
“distribute the global validation …. Value to the other training nodes” (Pg.6, particularly paragraph 0050; EN: this denotes the global model/server returning parameters to the private servers (i.e. the local models) to improve their models).
“Determine, based on the global validation … value, whether to continue the training process of the respective local version of the ML model” (Pg.11, particularly paragraph 0081; EN: this denotes using the incoming global model as a base-line to further train the local model if it wishes to additional training).
“based on the determination of whether to continue the training process, selectively continue the training process of the respective local version of the ML model” (Pg.11, particularly paragraph 0081; EN: this denotes using the incoming global model as a base-line to further train the local model if it wishes to do additional training).
However, Szeto fails to explicitly disclose, “a temporary leader node”, “receive, from a key manager, a public key for encrypting the received….”, “Receive, from each of the other training nodes, local … loss values”, “the local … loss values”, “a global … loss value”, “the system further comprising a second training node configured to: instruct the key manager to discard the public key”
Guo discloses, “a temporary leader node” (abstract; EN: this denotes the system selecting new leader nodes, making any leader node temporary).
“receive, from a key manager, a public key for encrypting the received….” (Pg.6-7, particularly paragraph 0058; EN: this denotes keyblock which contains the public key for the validators for the current epoch).
“the system further comprising a … node configured to: instruct the key manager to discard the public key” (pg.11, particularly paragraph 0095; EN: this denotes the creation of a new public epoch key to be used going forward, which discards the previous public key to no longer be used).
“relinquish acting as the temporary leader node” (pg.11, particularly paragraph 0095; EN: this denotes the creation of a new public epoch key to be used going forward, which discards the previous public key to no longer be used along with selecting a new leader node). “wherein the election to act as the temporary leader node ensures privacy by changing ownership of the public key” (pg.11, particularly paragraph 0095; EN: this denotes the creation of a new public epoch key to be used going forward, which discards the previous public key to no longer be used along with selecting a new leader node).
Tran discloses, “Receive from each of the other training nodes, local … loss values” and “the local … loss values” (Pg.1388, particularly paragraph C2, section III; EN: this denotes the loss function of individual clients (i.e. local models)).
“a global … loss value” (Pg.1388, Particularly C2, Section III; EN: This denotes combining the loss functions of the clients to get a combined (i.e. global) loss function).
Russinovich discloses, “the second training node configured to: instruct the key manager to discard the public key…” (Pg.10, particularly paragraph 0105; EN: The new claim amendments require that the first training node, and not the leader, request the changing of the public key. Russinovich discloses that any member may request a new public key as needed).
Szeto and Guo are analogous art because both involve blockchain.
Before the effective filing date it would have been obvious to one skilled in the art of blockchain to combine the work of Szeto and Guo in order to change public keys with different leaders.
The motivation for doing so would be to “encrypt… using the current epoch public key, and transmitted to the current leader node, who can validate the encrypted new-epoch keys using the current epoch private key” (Guo, Pg.11, paragraph 0095) or in the case of Szeto, allow the use of public keys among the nodes to ensure privacy of the encrypted data as needed.
Therefore before the effective filing date it would have been obvious to one skilled in the art of blockchain to combine the work of Szeto and Guo in order to change public keys with different leaders.
Szeto and Tran are analogous art because both involve distributed machine learning.
Before the effective filing date it would have been obvious to one skilled in the art of distributed machine learning to combine the work of Szeto and Tran in order to use and share loss functions between the local and global models.
The motivation for doing so would be to “find the model parameter w that characterizes the output y … with the loss function” (Tran, Pg.1388, C2, Section III, second paragraph) or in the case of Szeto, allow the system to use loss functions to train the local/global models, and share that information among the local/global system.
Therefore before the effective filing date it would have been obvious to one skilled in the art of distributed machine learning to combine the work of Szeto and Tran in order to use and share loss functions between the local and global models.
Szeto and Russinovich are analogous art because both involve blockchain.
Before the effective filing date it would have been obvious to one skilled in the art of blockchain to combine the work of Szeto and Russinovich in order to allow any member of the blockchain to request a new public key.
The motivation for doing so would be to cause “the network [to] not accept any transactions referencing the old KBK” (Russinovich, Pg.10, paragraph 0105) or in the case of Szeto, allow the system to have members
Therefore before the effective filing date it would have been obvious to one skilled in the art of blockchain to combine the work of Szeto and Russinovich in order to allow any member of the blockchain to request a new public key.
As per claim 12, Szeto discloses, “wherein the instructions that when executed cause the processor to build the global version of the ML model further causes the processor to build the global version of the ML model based on a local parameter derived from the training of the local version of the ML model at the training node in addition to the local parameters derived from the training of the respective local versions of the ML model at the other training nodes” (Pg.11, particularly paragraph 0081; EN: this denotes allowing other private networks within the system to be the global model, which would make use of its own local model along with the rest).
As per claim 16, Szeto discloses, “Wherein the training node and the other training nodes comprise a distributed ML network” (Pg.11, particularly paragraph 0081; EN: this denotes allowing other private networks within the system to be the global model, which would make use of its own local model along with the rest).
Tran discloses, “local … loss values” (Pg.1388, particularly paragraph C2, section III; EN: this denotes the loss function of individual clients (i.e. local models)).
As per claim 20, Szeto discloses, “wherein the memory unit includes instructions that when executed cause the processor to transmit the … local validation … values to the other training nodes as a training performance indicator of the respective local version of the ML model at the other training nodes” (Pg.6, particularly paragraph 0050; EN: this denotes the global model/server returning parameters to the private servers (i.e. the local models) to improve their models).
Tran discloses, “Averaged local … loss values” (Pg.1388, Particularly C2, Section III; EN: This denotes combining the loss functions of the clients to get a combined (i.e. global) loss function) The equation denotes this to be an average of the various local loss functions. Pg.1389, particularly C1, Communication section; EN: this denotes continuing the process until a global accuracy is achieved).
Claim Rejections - 35 USC § 103
Claims 2-5 and 13-15 are rejected under 35 U.S.C. 103 as being unpatentable over Szeto et al (US 20180018590 A1) in view of Guo et al (US 20210099312 A1), Tran et al (“Federated Learning over Wireless Networks: Optimization Model Design and Analysis”) and Russinovich (US 20180227275 A1) and further in view of Dirac et al (US 20150379072 A1).
As per claim 2, Szeto fails to explicitly disclose, “the instructions that when executed cause the processor to train the local version of the ML model further cause the processor to train the local version of the ML model using a training data subset of a local dataset at the training node.”
Dirac discloses, “the instructions that when executed cause the processor to train the local version of the ML model further cause the processor to train the local version of the ML model using a training data subset of a local dataset at the first training node” (Pg.13, particularly paragraph 0097; EN: this denotes breaking up datasets into training and test data subsets).
Szeto and Dirac are analogous art because both involve distributed machine learning.
Before the effective filing date it would have been obvious to one skilled in the art of distributed machine learning to combine the work of Szeto and Dirac in order to use split training data sets.
The motivation for doing so would be to provide “an iterative procedure that may be used to improve the quality of prediction made by a machine learning model” (Dirac, Pg. 24, paragraph 0144) or in the case of Szeto, allow the system to have testing data when training algorithms in order to properly test their algorithms on data different than the exact same training data.
Therefore before the effective filing date it would have been obvious to one skilled in the art of distributed machine learning to combine the work of Szeto and Dirac in order to use split training data sets.
As per claim 3, Dirac discloses, “wherein the instructions that when executed cause the processor to evaluate the local version of the ML model further cause the processor to evaluate the local version of the ML model using a validation data subset of the local dataset at the first training node” (Pg.13, particularly paragraph 0097; EN: this denotes breaking up datasets into training and test data subsets. Here the test data is the data used to validate that the system is being properly trained).
As per claim 4, Dirac discloses, “wherein the local dataset is divided into the training data and the validation data subsets prior to a local training iteration” (Pg.13, particularly paragraph 0097; EN: this denotes breaking up datasets into training and test data subsets. Here the test data is the data used to validate that the system is being properly trained).
As per claim 5, Szeto fails to explicitly disclose, “wherein the instructions that when executed cause the processor to train the local version of the ML model further cause the processor to train the local version of the ML models in batches.”
Dirac discloses, “wherein the instructions that when executed cause the processor to train the local version of the ML model further cause the processor to train the local version of the ML models in batches” (pg.19, particularly paragraph 0121; EN: this denotes breaking up the training data into multiple subsets for training (i.e. batches)).
Szeto and Dirac are analogous art because both involve distributed machine learning.
Before the effective filing date it would have been obvious to one skilled in the art of distributed machine learning to combine the work of Szeto and Dirac in order to use split training data sets.
The motivation for doing so would be to provide “In some embodiments, the machine learning model may support parallelized training of models, in which for example respective (and potentially overlapping) subsets of an input data set may be used to train a given model in parallel” (Dirac, Pg.19, paragraph 0121) or in the case of Szeto, allow the system to break the training data up into chunks/batches and train the system efficiently based upon the system’s ability/requirements.
Therefore before the effective filing date it would have been obvious to one skilled in the art of distributed machine learning to combine the work of Szeto and Dirac in order to use split training data sets.
As per claim 13, Szeto fails to explicitly disclose, “wherein the instructions that when executed cause the processor to train the local version of the ML model comprise instructions that when executed further cause the processor to train the local version of the ML model using a training data subset of data local to the training node.”
Dirac discloses, “wherein the instructions that when executed cause the processor to train the local version of the ML model comprise instructions that when executed further cause the processor to train the local version of the ML model using a training data subset of data local to the training node” (Pg.13, particularly paragraph 0097; EN: this denotes breaking up datasets into training and test data subsets).
Szeto and Dirac are analogous art because both involve distributed machine learning.
Before the effective filing date it would have been obvious to one skilled in the art of distributed machine learning to combine the work of Szeto and Dirac in order to use split training data sets.
The motivation for doing so would be to provide “an iterative procedure that may be used to improve the quality of prediction made by a machine learning model” (Dirac, Pg. 24, paragraph 0144) or in the case of Szeto, allow the system to have testing data when training algorithms in order to properly test their algorithms on data different than the exact same training data.
Therefore before the effective filing date it would have been obvious to one skilled in the art of distributed machine learning to combine the work of Szeto and Dirac in order to use split training data sets.
As per claim 14, Szeto discloses, “Wherein the memory unit includes instructions that when executed further cause the processor to calculate a … validation … value” (Pg.6, particularly paragraph 0050; EN: this denotes checking to see if the local model improves or worsens. Pg.14, particularly paragraph 0098; EN: this denotes breaking up of training data into training and validation sets).
Tran discloses, “local … loss value” (Pg.1388, particularly paragraph C2, section III; EN: this denotes the loss function of individual clients (i.e. local models)).
Dirac discloses, “a … validation … value using validation data subset of the data local to the training node” (Pg.13, particularly paragraph 0097; EN: this denotes breaking up datasets into training and test data subsets. Here the test data is the data used to validate that the system is being properly trained).
As per claim 15, Szeto discloses, “wherein the instructions that when executed cause the processor to …. Local validation … values… comprises instructions that when executed cause the process or to… the local validation … values from each of the other training nodes ” (Pg.6, particularly paragraph 0050; EN: this denotes checking to see if the local model improves or worsens. Pg.14, particularly paragraph 0098; EN: this denotes breaking up of training data into training and validation sets). “from each of the other training nodes in addition to the local validation … value calculated by the training node” (Pg.11, particularly paragraph 0081; EN: this denotes allowing other private networks within the system to be the global model, which would make use of its own local model along with the rest).
Tran discloses, “average the local … loss values from each of the other training nodes” (Pg.1388, Particularly C2, Section III; EN: This denotes combining the loss functions of the clients to get a combined (i.e. global) loss function) The equation denotes this to be an average of the various local loss functions).
Claim Rejections - 35 USC § 103
Claims 8, 10, and 17-19 are rejected under 35 U.S.C. 103 as being unpatentable over Szeto et al (US 20180018590 A1) in view of Guo et al (US 20210099312 A1), Tran et al (“Federated Learning over Wireless Networks: Optimization Model Design and Analysis”) and Russinovich (US 20180227275 A1) and further in view of Lyu et al (“Towards Fair and Decentralized Privacy-Preserving Deep Learning with Blockchain”).
As per claim 8, Szeto discloses, “validation … value” (Pg.6, particularly paragraph 0050; EN: this denotes checking to see if the local model improves or worsens. Pg.14, particularly paragraph 0098; EN: this denotes breaking up of training data into training and validation sets).
Tran discloses, “local … loss value” (Pg.1388, particularly paragraph C2, section III; EN: this denotes the loss function of individual clients (i.e. local models)).
However, Szeto fails to explicitly disclose, “wherein the memory unit includes instructions that when executed further cause the processor to homomorphically encrypt the … value using a public key of a public and private key pair”
Lyu discloses, “wherein the memory unit includes instructions that when executed further cause the processor to homomorphically” (Pg.4, particularly C1, Section D: EN: this denotes the use of homomorphic encryption). “ encrypt the … value using a public key of a public and private key pair” (Pg.5, particularly C1, last paragraph; EN: this denotes the use of private keys and public keys to perform data transfer operations).
Szeto and Lyu are analogous art because both involve machine learning privacy.
Before the effective filing date it would have been obvious to one skilled in the art of privacy in machine learning to combine the work of Szeto and Lyu in order to use encryption to protect data.
The motivation for doing so would be to avoid “all parties encrypt[ing] their data using the same public key, [causing] each party not only be able to decrypt the aggregate, but also each individual’s values” (Lyu, Pg.4, C2, section D) or in the case of Szeto, allow the data to be transferred safely preserving the privacy of each individual local model.
Therefore before the effective filing date it would have been obvious to one skilled in the art of privacy in machine learning to combine the work of Szeto and Lyu in order to use encryption to protect data.
As per claim 10, Szeto discloses, “wherein the training node, the leader node, and additional node operate in a distributed swarm learning… network” (Pg.4, particularly paragraph 0029; EN: this denotes the fact that the private servers do not share their training data but only share parameters to the common model, which is the way this is described in the specification).
However, Szeto fails to explicitly disclose, “blockchain.”
Lyu discloses, “blockchain” (Pg.4, particularly C1, second paragraph and section B; EN: this denotes using Blockchain to preserve privacy in a machine learning network).
Szeto and Lyu are analogous art because both involve machine learning privacy.
Before the effective filing date it would have been obvious to one skilled in the art of privacy in machine learning to combine the work of Szeto and Lyu in order to blockchain to protect data.
The motivation for doing so would be to “establish a privacy-preserving collaborative deep framework based on private Blockchain to ensure both privacy and fairness” (Lyu, Pg.4, C1, Second paragraph) or in the case of Szeto, allow the system to use Blockchain to ensure privacy of data transferred by the system.
Therefore before the effective filing date it would have been obvious to one skilled in the art of privacy in machine learning to combine the work of Szeto and Lyu in order to blockchain to protect data.
As per claim 17, Szeto discloses, “Wherein the instructions that when executed cause the processor to receive local parameters, build the global version of the ML model, transmit the merged local parameters, and receive the local validation … values, further causes the processor to receive local parameters” (Pg.3, particularly paragraph 0019; EN: this denotes transmitting the features of the model to a non-private server to create an aggregated model. This non-private server will be the “leader node.”). “Built the global version of the ML model” (Pg.3, particularly paragraph 0019; EN: this denotes transmitting the features of the model to create an aggregated model). “Transmit the merged local parameters” (Pg.6, particularly paragraph 0050; EN: this denotes the global model/server returning parameters to the private servers (i.e. the local models) to improve their models).”and receive the local validation … values” (Pg.6, particularly paragraph 0050; EN: this denotes checking to see if the local model improves or worsens. Pg.14, particularly paragraph 0098; EN: this denotes breaking up of training data into training and validation sets).
Tran discloses, “local … loss values” (Pg.1388, particularly paragraph C2, section III; EN: this denotes the loss function of individual clients (i.e. local models)).
However, Szeto fails to explicitly disclose, “using a distributed blockchain ledger.”
Lyu discloses, “using a distributed blockchain ledger” (Pg.4, particularly C1, second paragraph and section B; EN: this denotes using Blockchain to preserve privacy in a machine learning network).
Szeto and Lyu are analogous art because both involve machine learning privacy.
Before the effective filing date it would have been obvious to one skilled in the art of privacy in machine learning to combine the work of Szeto and Lyu in order to blockchain to protect data.
The motivation for doing so would be to “establish a privacy-preserving collaborative deep framework based on private Blockchain to ensure both privacy and fairness” (Lyu, Pg.4, C1, Second paragraph) or in the case of Szeto, allow the system to use Blockchain to ensure privacy of data transferred by the system.
Therefore before the effective filing date it would have been obvious to one skilled in the art of privacy in machine learning to combine the work of Szeto and Lyu in order to blockchain to protect data.
As per claim 18, Szeto fails to explicitly disclose, “wherein the memory unit includes instructions that when executed further causes the processor to request a key manager to generate an asymmetric key pair with which the local parameters are encrypted and with which the merged local parameters are decrypted.”
Lyu discloses, “wherein the memory unit includes instructions that when executed further causes the processor to request a key manager to generate an asymmetric key pair with which the local parameters are encrypted and with which the merged local parameters are decrypted” (Pg.5, particularly C2, first full paragraph; EN: this denotes the creation and use of kyes for local model parameters, including asymmetric keys).
Szeto and Lyu are analogous art because both involve machine learning privacy.
Before the effective filing date it would have been obvious to one skilled in the art of privacy in machine learning to combine the work of Szeto and Lyu in order to use encryption to protect data.
The motivation for doing so would be to avoid “all parties encrypt[ing] their data using the same public key, [causing] each party not only be able to decrypt the aggregate, but also each individual’s values” (Lyu, Pg.4, C2, section D) or in the case of Szeto, allow the data to be transferred safely preserving the privacy of each individual local model.
Therefore before the effective filing date it would have been obvious to one skilled in the art of privacy in machine learning to combine the work of Szeto and Lyu in order to use encryption to protect data.
As per claim 19, Szeto discloses, “local validation… values” (Pg.6, particularly paragraph 0050; EN: this denotes checking to see if the local model improves or worsens. Pg.14, particularly paragraph 0098; EN: this denotes breaking up of training data into training and validation sets).
Tran discloses, “Local … loss value” (Pg.1388, particularly paragraph C2, section III; EN: this denotes the loss function of individual clients (i.e. local models)).
“averaged local … loss value” (Pg.1388, Particularly C2, Section III; EN: This denotes combining the loss functions of the clients to get a combined (i.e. global) loss function) The equation denotes this to be an average of the various local loss functions).
Lyu discloses, “wherein the memory unit includes instructions that when executed cause the processor to request a key manager to generate another asymmetric key pair with which the local … values are encrypted and with which the …local validation … values are decrypted” (Pg.5, particularly C2, first full paragraph; EN: this denotes the creation and use of kyes for local model parameters, including asymmetric keys. When combined with Szeto, this would generate all the required keys to provide proper privacy/encryption as needed).
Claim Rejections - 35 USC § 103
Claim 21 is rejected under 35 U.S.C. 103 as being unpatentable over Szeto et al (US 20180018590 A1) in view of Guo et al (US 20210099312 A1), Tran et al (“Federated Learning over Wireless Networks: Optimization Model Design and Analysis”) and Russinovich (US 20180227275 A1) and further in view of Invidia et al (“An IoT-oriented Fast Prototyping Platform for BLE-based Star Topology Networks”).
Szeto discloses, “Wherein the instructions that when executed cause the processor to receive… from each training node to download information for the training process of each training node” (Pg.14, particularly paragraph 0099; EN: this denotes the various network connections the system can use to communicate and pass information as needed).
Szeto fails to explicitly disclose, “via a star topology, a uniform resource locator (URL).”
Invidia discloses, “via a star topology, a uniform resource locator (URL)” (Pg.139, particularly C1, second paragraph; EN: this denotes the use of star topology networks; Pg.145, particularly C1, second paragraph; EN: this denotes the use of URL with the network).
Szeto and Invidia are analogous art because both involve computer networks.
Before the effective filing date it would have been obvious to one skilled in the art of computer networks to combine the work of Szeto and Invidia in order to use star topology networks.
The motivation for doing so would be because “The topology uses the least amount of power compared to other architectures thanks to simple direct wireless connections” (Invidia, Pg.139, C2, second full paragraph) or in the case of Szeto, allow the system to use a lower power connection as needed among the nodes.
Therefore before the effective filing date it would have been obvious to one skilled in the art of computer networks to combine the work of Szeto and Invidia in order to use star topology networks.
Response to Arguments
Applicant's arguments with respect to claims 1-8 and 10-20 have been considered but are moot in view of the new ground(s) of rejection.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to BEN M RIFKIN whose telephone number is (571)272-9768. The examiner can normally be reached Monday-Friday 9 am - 5 pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Alexey Shmatov can be reached at (571) 270-3428. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/BEN M RIFKIN/Primary Examiner, Art Unit 2123
Prosecution Insights