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 3, 4, 8, 9 if written in independent form in view of claim 1, is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding claim 3 in view of claim 1,
Step 1:
Claim 3 in view of claim 1 recites a method, one of the four statutory categories of patentable subject matter.
Step 2A, Prong I:
Claim 3 further recites the limitations of:
“evaluating the result of the training provided by the target data providers” This limitation recites a mental process. A person can mentally evaluate the training result of a machine learning model neural network. A person can view a training result by a machine learning model and mentally assess or judge as a mental process of evaluating the result, thus the evaluating process is a mental process.
“determining a level of contribution of the target data providers based on a result of the evaluation” This limitation recites a mental process. A person can mentally determine a level of contribution based on the result of evaluation of the machine learning model result. A person can mentally compare various evaluated training results to determine a relative level of contribution according to such comparison, which constitutes a mental process. Thus, the determination of a level of contribution is a mental process.
“determining compensation corresponding to the target data providers, based on the level of contribution” This limitation recites a mental process. A person can mentally determine compensation based on the level contribution toward the data provider. For instance, a person can mentally determine if some data providers need more training resources such as better equipment based on observing that the training result from some providers does not contribute much in comparison to others because the model training is too slow due to lack of equipment. Such determination for compensation constitutes a mental process, and thus determining compensation is considered to be a mental process.
Step 2A, Prong II:
Claim 3 in view of claim 1 recites the following additional elements:
“a central server” This additional element is a high-level recitation of generic computer components used as a tool, and does not provide integration into a practical application.
“providing data provider information to a consumer” This additional element recites an additional element of an insignificant extra-solution activity as identified in MPEP 2106.05(g) of mere data gathering, and does not provide integration into a practical application. The limitation simply recites transmitting data information between data provider and a consumer for mere data gathering.
“receiving, from the consumer, data request information comprising an artificial neural network (ANN) model” This additional element recites an additional element of an insignificant extra-solution activity as identified in MPEP 2106.05(g) of mere data gathering, and does not provide integration into a practical application. The limitation simply recites transmitting and receiving data including the ANN model between data providers and consumer for mere data gathering.
“transmitting the data request information to target data providers” This additional element recites an additional element of an insignificant extra-solution activity as identified in MPEP 2106.05(g) of mere data gathering, and does not provide integration into a practical application. The limitation simply recites transmitting data information between data provider and consumer for mere data gathering
“receiving, from the target data providers, a result of training of an individually trained ANN model” This additional element recites an additional element of an insignificant extra-solution activity as identified in MPEP 2106.05(g) of mere data gathering, and does not provide integration into a practical application. The limitation simply recites transmitting and receiving data information between data provider and a consumer including a training result from black-box application of the ANN for mere data gathering.
“generating a common model based on the result of the training” This additional element recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application. The limitation merely recites a black box application of neural network machine learning model, in which a machine learning model is trained to produce a result for generating a common model, without reciting specific machine learning algorithm to generate or train the model or unconventional machine learning practice to train the machine learning model or improvement toward any computer elements.
“transmitting the common model to the consumer” This additional element recites an additional element of an insignificant extra-solution activity as identified in MPEP 2106.05(g) of mere data gathering, and does not provide integration into a practical application. The limitation simply recites transmitting and receiving data information between data provider and a consumer including the model obtained from merely applying and training the ANN for mere data gathering.
Step 2B:
When considered individually or in combination, the additional limitations and elements of claim 1 do not amount to significantly more than the judicial exception for the same reasons discussed above as to why the additional limitations do not integrate the abstract idea into a practical application. The additional elements outlined in Step 2A performing functions as designed simply accomplish execution of the abstract ideas.
Additional elements “a central server” are high-level recitations of generic computer components used as a tool and does not amount to significantly more than the judicial exception for the same reasons discussed above as to why the additional limitations do not integrate the abstract idea into a practical application.
The additional element “providing data provider information to a consumer” recites an additional element of a well-understood, routine, conventional activity as identified in MPEP 2106.05(d)(i) of receiving or transmitting data over a network and does not provide integration into a practical application or amount to significantly more than the judicial exception.
The additional element “receiving, from the consumer, data request information comprising an artificial neural network (ANN) model” recites an additional element of a well-understood, routine, conventional activity as identified in MPEP 2106.05(d)(i) of receiving or transmitting data over a network and does not provide integration into a practical application or amount to significantly more than the judicial exception.
The additional element “transmitting the data request information to target data providers” recites an additional element of a well-understood, routine, conventional activity as identified in MPEP 2106.05(d)(i) of receiving or transmitting data over a network and does not provide integration into a practical application or amount to significantly more than the judicial exception.
The additional element “receiving, from the target data providers, a result of training of an individually trained ANN model” recites an additional element of a well-understood, routine, conventional activity as identified in MPEP 2106.05(d)(i) of receiving or transmitting data over a network and does not provide integration into a practical application or amount to significantly more than the judicial exception.
The additional element “generating a common model based on the result of the training” recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not amount to significantly more than the judicial exception for the same reasons discussed above.
The additional element “transmitting the common model to the consumer” recites an additional element of a well-understood, routine, conventional activity as identified in MPEP 2106.05(d)(i) of receiving or transmitting data over a network and does not provide integration into a practical application or amount to significantly more than the judicial exception.
In conclusions from above for the elements considered as a mental process, elements reciting high-level recitation of generic computer components used as a tool, elements reciting a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), elements reciting an insignificant extra-solution activity as identified in MPEP 2106.05(g), and elements reciting a well-understood, routine, conventional activity as identified in MPEP 2106.05(d) are carried over and do not provide significantly more than the abstract idea. Looking at the limitations in combination and the claims as a whole does not change this conclusion and the claim is ineligible.
Therefore, additional limitations of claim 3 in view of claim 1 do not amount to significantly more than the judicial exception.
Thus, claim 3 recites abstract ideas with additional elements rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception.
Therefore, claim 3 in view of claim 1 if written in independent form is not patent eligible.
Regarding claim 4 depends on claim 1, thus the rejection of claim 1 is incorporated. Claim 4 if written in independent form in view of claim 1, is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Claim 4 recites:
“obtaining an amount of training data used for the training from each of the target data providers” This additional element recites an additional element of an insignificant extra-solution activity as identified in MPEP 2106.05(g) of mere data gathering, and does not provide integration into a practical application. The limitation further recites an additional element of a well-understood, routine, conventional activity as identified in MPEP 2106.05(d)(i) of receiving or transmitting data over a network and does not provide integration into a practical application or amount to significantly more than the judicial exception.
“determining reliability corresponding to each of the target data providers, based on the result of the training”. This limitation recites a mental process. A person can mentally determine reliability corresponding to each of the target data providers based on seeing the training result. Such determination involves mental comparison and evaluation between machine learning results from data providers, thus the process of determining reliability is considered to be a mental process.
“evaluating the result of the training based on the amount of training data and the reliability”. This limitation recites a mental process. A person can mentally evaluate the training result based on the amount of training data and the reliability. A person can mentally count the amount of training data such as count how many images is within the training data and determine the reliability, thus constitutes the evaluating as a mental process.
Thus, claim 4 recites abstract ideas with additional elements rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception. Therefore, claim 4 in view of claim 1 if written in independent form is not patent eligible.
Regarding claim 8 depends on claim 1, thus the rejection of claim 1 is incorporated. Claim 8 if written in independent form in view of claim 1, is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Claim 8 recites
“determining the target data providers based on the data request information”. This limitation recites a mental process. A person can mentally determine the target data providers based on the data request information. For example, a person can see if the data request information is a picture and determine the data provider as a library.
Thus, claim 8 recites abstract ideas with additional elements rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception. Therefore, claim 8 in view of claim 1 if written in independent form is not patent eligible.
Regarding claim 9 depends on claim 1, thus the rejection of claim 1 is incorporated. Claim 9 if written in independent form in view of claim 1, is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Claim 9 recites
“receiving, from a data provider, a number of pieces of training data corresponding to the data request information” This additional element recites an additional element of an insignificant extra-solution activity as identified in MPEP 2106.05(g) of mere data gathering, and does not provide integration into a practical application. The limitation further recites an additional element of a well-understood, routine, conventional activity as identified in MPEP 2106.05(d)(i) of receiving or transmitting data over a network and does not provide integration into a practical application or amount to significantly more than the judicial exception.
“determining, to be the target data providers, data providers of which the number of pieces of training data is greater than or equal to a predetermined threshold value”. This limitation recites a mental process. A person can mentally compare the count of training data as greater than or equal to a predetermined threshold value. Such comparison between numbers is considered as a mental process and can be performed within a human’s mind.
Thus, claim 9 recites abstract ideas with additional elements rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception. Therefore, claim 9 in view of claim 1 if written in independent form is not patent eligible.
Claims 14, 15 if written in independent form in view of claim 12, is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding claim 14 in view of claim 12,
Step 1:
Claim 14 in view of claim 12 recites a machine apparatus, one of the four statutory categories of patentable subject matter.
Step 2A, Prong I:
Claim 14 further recites the limitations of:
“evaluating the result of the training provided by the target data providers” This limitation recites a mental process. A person can mentally evaluate the training result of a machine learning model neural network. A person can view a training result by a machine learning model and mentally assess or judge as a mental process of evaluating the result, thus the evaluating process is a mental process.
“determining a level of contribution of the target data providers based on a result of the evaluation” This limitation recites a mental process. A person can mentally determine a level of contribution based on the result of evaluation of the machine learning model result. A person can mentally compare various evaluated training results to determine a relative level of contribution according to such comparison, which constitutes a mental process. Thus, the determination of a level of contribution is a mental process.
“determining compensation corresponding to the target data providers, based on the level of contribution” This limitation recites a mental process. A person can mentally determine compensation based on the level contribution toward the data provider. For instance, a person can mentally determine if some data providers need more training resources such as better equipment based on observing that the training result from some providers does not contribute much in comparison to others because the model training is too slow due to lack of equipment. Such determination for compensation constitutes a mental process, and thus determining compensation is considered to be a mental process.Step 2A, Prong II:
Claim 14 in view of claim 12 recites the following additional elements:
“A central server apparatus comprising: a transmitter; a receiver; and a processor” These additional elements are a high-level recitation of generic computer components used as a tool, and does not provide integration into a practical application.
“wherein the transmitter is configured to providing data provider information to a consumer” This additional element recites an additional element of an insignificant extra-solution activity as identified in MPEP 2106.05(g) of mere data gathering, and does not provide integration into a practical application. The limitation simply recites transmitting data information between data provider and a consumer for mere data gathering.
“wherein the receiver is configured to receive, from the consumer, data request information comprising an artificial neural network (ANN) model” This additional element recites an additional element of an insignificant extra-solution activity as identified in MPEP 2106.05(g) of mere data gathering, and does not provide integration into a practical application. The limitation simply recites transmitting and receiving data including the ANN model between data providers and consumer for mere data gathering.
“wherein the transmitter is configured to transmit the data request information to target data providers” This additional element recites an additional element of an insignificant extra-solution activity as identified in MPEP 2106.05(g) of mere data gathering, and does not provide integration into a practical application. The limitation simply recites transmitting data information between data provider and consumer for mere data gathering.
“wherein the receiver is configured to receive, from the target data providers, a result of training of an individually trained ANN model” This additional element recites an additional element of an insignificant extra-solution activity as identified in MPEP 2106.05(g) of mere data gathering, and does not provide integration into a practical application. The limitation simply recites transmitting and receiving data information between data provider and a consumer including a training result from black-box application of the ANN for mere data gathering.
“wherein the processor is configured to generate a common model based on the result of the training” This additional element recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application. The limitation merely recites a black box application of neural network machine learning model, in which a machine learning model is trained to produce a result for generating a common model, without reciting specific machine learning algorithm to generate or train the model or unconventional machine learning practice to train the machine learning model or improvement toward any computer elements.
“wherein the transmitter is configured to transmit the common model to the consumer” This additional element recites an additional element of an insignificant extra-solution activity as identified in MPEP 2106.05(g) of mere data gathering, and does not provide integration into a practical application. The limitation simply recites transmitting and receiving data information between data provider and a consumer including the model obtained from merely applying and training the ANN for mere data gathering.
Step 2B:
When considered individually or in combination, the additional limitations and elements of claim 1 do not amount to significantly more than the judicial exception for the same reasons discussed above as to why the additional limitations do not integrate the abstract idea into a practical application. The additional elements outlined in Step 2A performing functions as designed simply accomplish execution of the abstract ideas.
Additional elements “A central server apparatus comprising: a transmitter; a receiver; and a processor” are high-level recitations of generic computer components used as a tool and does not amount to significantly more than the judicial exception for the same reasons discussed above as to why the additional limitations do not integrate the abstract idea into a practical application.
The additional element “wherein the transmitter is configured to providing data provider information to a consumer” recites an additional element of a well-understood, routine, conventional activity as identified in MPEP 2106.05(d)(i) of receiving or transmitting data over a network and does not provide integration into a practical application or amount to significantly more than the judicial exception.
The additional element “wherein the receiver is configured to receive, from the consumer, data request information comprising an artificial neural network (ANN) model” recites an additional element of a well-understood, routine, conventional activity as identified in MPEP 2106.05(d)(i) of receiving or transmitting data over a network and does not provide integration into a practical application or amount to significantly more than the judicial exception.
The additional element “wherein the transmitter is configured to transmit the data request information to target data providers” recites an additional element of a well-understood, routine, conventional activity as identified in MPEP 2106.05(d)(i) of receiving or transmitting data over a network and does not provide integration into a practical application or amount to significantly more than the judicial exception.
The additional element “wherein the receiver is configured to receive, from the target data providers, a result of training of an individually trained ANN model” recites an additional element of a well-understood, routine, conventional activity as identified in MPEP 2106.05(d)(i) of receiving or transmitting data over a network and does not provide integration into a practical application or amount to significantly more than the judicial exception.
The additional element “wherein the processor is configured to generate a common model based on the result of the training” recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not amount to significantly more than the judicial exception for the same reasons discussed above.
The additional element “wherein the transmitter is configured to transmit the common model to the consumer” recites an additional element of a well-understood, routine, conventional activity as identified MPEP 2106.05(d)(i) of receiving or transmitting data over a network and does not provide integration into a practical application or amount to significantly more than the judicial exception.
In conclusions from above for the elements considered as a mental process, elements reciting high-level recitation of generic computer components used as a tool, elements reciting a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), elements reciting an insignificant extra-solution activity as identified in MPEP 2106.05(g), and elements reciting a well-understood, routine, conventional activity as identified in MPEP 2106.05(d) are carried over and do not provide significantly more than the abstract idea. Looking at the limitations in combination and the claims as a whole does not change this conclusion and the claim is ineligible.
Therefore, additional limitations of claim 14 in view of claim 12 do not amount to significantly more than the judicial exception.
Thus, claim 14 recites abstract ideas with additional elements rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception.
Therefore, claim 14 in view of claim 12 if written in independent form is not patent eligible.
Regarding claim 15 depends on claim 12, thus the rejection of claim 12 is incorporated. Claim 15 if written in independent form in view of claim 12, is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Claim 15 is rejected under the same rationale as claim 4. The applicant is directed to the rejection of claim 4 above, because the claim recites similar limitations and processing steps.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claims 1-2, 5-8, 11-13 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Pushkin et.al (US 12061956 B1).
Regarding claim 1,
Pushkin teaches “An operating method of a central server, the operating method comprising” (Column 1 lines 62-66 “FIG. 1 illustrates embodiments of systems that support a federated learning service. As illustrated, a provider network 100 includes a model training and hosting service 110. This service 110 allows for a user provided ML model to be trained and then hosted by the provider network 100” Pushkin discloses a provider network comprising a model training and hosting service and a federated learning service configured to coordinate the training and deployment of an ML model using a plurality of training clients. Accordingly, the provider network, including the federated learning service, corresponds to the claimed central server, and its operation in coordinating model training and deployment corresponds to the operating method of a central server as recited in the claim.)
Pushkin teaches “providing data provider information to a consumer” (Figure 1, and Column 4 lines 51-56 “a requesting edge device 140 requests that a ML model be trained using federated learning and subsequently deployed. The request may include one or more of: an identifier of an initial model (e.g., a path to a model), an initial model, an identifier of one or more training clients (e.g., a list of training clients)”, Column 2 lines 14-29 “In particular, a remote deployment service (or module) 122 provides for this ability. The ability to remove a model deployment focuses the remote deployment service 122 on managing a population of connected edge device (clients) ... client registration and ML model lifecycle tracking (e.g., clients can be registered with the federated learning service 120 and have their capabilities and availability advertised) ... edge device characteristics (e.g., power supply, processor type, latency, storage, location—address, operating system, current state, SDK, etc.)”, and Column 3 lines 19-24 “Users may interact with a provider network 100 across one or more intermediate networks 106 (e.g., the internet) via one or more interface(s), such as through use of application programming interface (API) calls, via a console implemented as a website or application, etc.” Pushkin discloses a provider network comprising a model training and hosting service and a federated learning service configured to coordinate the training and deployment of an ML model using a plurality of training clients. Accordingly, the provider network, including the federated learning service, corresponds to the claimed central server, and its operation in coordinating model training and deployment corresponds to the operating method of a central server as recited in the claim.)
Pushkin teaches “receiving, from the consumer, data request information comprising an artificial neural network (ANN) model” (Column 9 lines 57-63 “At 602, a request to develop and deploy a ML model using a federated learning service is received. As noted above, the request may include information such as an identifier an initial model, identifiers of training clients, a location of training data (e.g., locally stored on the training clients or accessible to the federated learning service to distribute), etc.”, and Column 5 lines 56-57 “Different local data may address different layers of the model.” Pushkin discloses a workflow in which a requesting edge device submits a request to a federated learning service to develop and deploy an ML model, wherein the request may include or identify an initial ML model having layers and adjustable weights, and wherein the federated learning service provides the initial ML model to participating edge devices for training using their respective local data. Accordingly, the initial ML model corresponds to the claimed ANN model, and the request including or identifying the initial ML model corresponds to the claimed data request information, thereby teaching receiving, from the consumer, data request information comprising an ANN model as claimed.)
Pushkin teaches “transmitting the data request information to target data providers” (Column 9-10, lines 64-67/1-9 “At 603, the development of a candidate ML model is caused to occur. This development may include several acts. At 604, the training of a “fine-tuned” ML model in each of a plurality of clients (e.g., edge devices) is caused to occur. For example, the training orchestrator 304/404 of the remote training service (module) 124 causes one or more edge devices ... to each train a fine-tuned model from an initial model ... The initial model candidate is provided to the plurality of clients in some embodiments at 606. Then each of the plurality clients is caused to train the initial model using local data”. Pushkin discloses a workflow in which a training orchestrator of a remote training service provides an initial ML model to a plurality of training clients (edge devices) for training using their respective local data. Accordingly, the training orchestrator corresponds to the claimed central server functionality, the training clients (edge devices) correspond to the claimed target data providers, and the initial ML model provided to the training clients (edge devices) corresponds to the claimed data request information, thereby teaching the transmitting of the data request information to the target data providers as claimed.)
Pushkin teaches “receiving, from the target data providers, a result of training of an individually trained ANN model” (Column 10 lines 12-17 “At 610, the differences between the initial model and the plurality of “fine-tuned” models are determined or received. In some embodiments, the training clients determine the differences. In other embodiments, the training orchestrator 304/404 of the remote training service (module) 124 determines the differences”, and Column 5 lines 58-65 “the remote training service (or module) 124 provides for an ability to train a given model using edge devices and/or fleets. In this example, edge devices are used. Within, or available to, the remote training service (or module) 124 are a production model 302 (the model to be deployed), a training orchestrator 304 that coordinates training of a “fine-tuned” model from an common initial model by one or more training clients (edge devices) 320(A)-(N)”. Pushkin discloses a workflow in which a plurality of training clients (edge devices) individually train an initial ML model using their respective local data to generate fine-tuned models and provide differences between the initial model and the fine-tuned models to a remote training service. The training clients (edge devices) correspond to the claimed target data providers, the fine-tuned models correspond to the claimed individually trained ANN models, and the differences provided to the remote training service correspond to the claimed results of training, thereby teaching the receiving, from the target data providers, a result of training of an individually trained ANN model as claimed.)
Pushkin teaches “generating a common model based on the result of the training” (Column 10 lines 18-20 “An aggregate (e.g., a mean, median, etc.) for the differences is determined at 612. The aggregate is applied to the initial model to generate a model candidate at 614. For example, the weights of the initial model are changed to reflect the aggregate for the differences.” Pushkin discloses a workflow in which a training orchestrator aggregates differences between an initial ML model and individually fine-tuned models and applies the aggregated differences to the initial ML model to generate an updated model candidate. Accordingly, the differences obtained from training the initial ML model using local data by each edge devices correspond to the claimed results of training, the aggregation and application of the differences correspond to generating a model based on the training results, and the resulting model candidate corresponds to the claimed common model, thereby teaching the generating of a common model based on the result of the training as claimed.)
Pushkin teaches “transmitting the common model to the consumer” (Column 2 lines 41-44 “The remote deployment service 122 communicates with a user sign-in service 112 that provides information about users that may communicate with the edge devices. As such, ML models are deployed for properly authenticated users”, Column 9 lines 25-29 “Upon a successful test (that is the trained ML model is acceptable), the remote deployment service (module) is alerted of an available production model at 507. The production model may also be stored for retrieval along with version information, etc.”, and Column 10 lines 31-34 “When the tested model candidate is good enough for production, it is approved for production (and the deployment service (module) notified) at 624. It can be deployed at 624.” Pushkin discloses a workflow in which a requesting edge device requests federated training and subsequent deployment of an ML model, and following training and testing, a remote deployment service deploys the resulting production model to edge devices for properly authenticated users; accordingly, a person of ordinary skill in the art would understand the requesting user to be an authenticated user for whom the resulting trained model is deployed, the production model corresponds to the claimed common model, and the requesting user corresponds to the claimed consumer, thereby teaching the transmitting of the common model to the consumer as claimed.)
Regarding claim 2 depends on claim 1, thus the rejection of claim 1 is incorporated.
Pushkin teaches “The operating method of claim 1, wherein the data provider comprises a federated learning module capable of communicating with the central server and is configured to train the ANN model using the federated learning module” (Column 1 lines 62-66 “FIG. 1 illustrates embodiments of systems that support a federated learning service. As illustrated, a provider network 100 includes a model training and hosting service 110. This service 110 allows for a user provided ML model to be trained and then hosted by the provider network 100”, and Column 5 lines 58-65 “the remote training service (or module) 124 provides for an ability to train a given model using edge devices and/or fleets. In this example, edge devices are used. Within, or available to, the remote training service (or module) 124 are a production model 302 (the model to be deployed), a training orchestrator 304 that coordinates training of a “fine-tuned” model from an common initial model by one or more training clients (edge devices) 320(A)-(N)”. Pushkin discloses a workflow in which a plurality of training clients (edge devices) participate in federated learning coordinated by a remote training service, wherein the training clients train a common initial ML model to generate respective fine-tuned models. The training clients (edge devices) correspond to the claimed data providers, their local training functionality corresponds to the claimed federated learning modules configured to train the ANN model, and their participation in coordinated training with the remote training service corresponds to communication with the central server, thereby teaching a data provider comprising a federated learning module capable of communicating with the central server and configured to train the ANN model using the federated learning module as claimed.)
Regarding claim 5 depends on claim 1, thus the rejection of claim 1 is incorporated.
Pushkin teaches “The operating method of claim 1, wherein the provider information comprises list information about a data provider and a data format provided by the data provider” (Column 4 lines 53-56 “The request may include one or more of: an identifier of an initial model (e.g., a path to a model), an initial model, an identifier of one or more training clients (e.g., a list of training clients)”, and Column 3 lines 24-28 “An API refers to an interface and/or communication protocol between a client and a server, such that if the client makes a request in a predefined format, the client should receive a response in a specific format or initiate a defined action”. Pushkin discloses a workflow in which a training request may include an identifier of one or more training clients (edge devices), including a list of training clients, and an API facilitates communication between clients and a server using predefined request and response formats. The list of training clients corresponds to the claimed list information about a data provider, and the predefined request and response formats correspond to the claimed data format, thereby teaching the provider information comprising list information about a data provider and a data format provided by the data provider as claimed.)
Regarding claim 6 depends on claim 1, thus the rejection of claim 1 is incorporated.
Pushkin teaches “The operating method of claim 1, wherein the transmitting of the data request information to the target data providers comprises transmitting the ANN model to the target data providers by encrypting the ANN model” (Column 5 lines 3-10 “The request may include one or more of: ... locations of encryption keys, or the keys themselves, to be used for data exchange and/or content within training or testing device”, and Column 7 lines 48-57 “The initial model 426 may be a model (or algorithm) that the requester (or someone on the requester's behalf) developed, a model (or algorithm) provided by the provider network 100 developed by an employee of the provider network 100, a model (or algorithm) provided by the provider network 100 developed by a third party, etc. In some embodiments, these components may be a part of an encrypted execution environment 321. As such, the remote training service 124 does not have access (without a key)”. Pushkin discloses a workflow in which a remote training service provides an initial ML model to training clients (edge devices) for federated training, wherein a training request may include encryption keys or locations of encryption keys for data exchange or content within the training devices, and the initial ML model may be part of an encrypted execution environment requiring a key for access. The initial ML model corresponds to the claimed ANN model, the training clients (edge devices) correspond to the claimed target data providers, and the provision of the initial ML model in conjunction with the disclosed encryption keys and encrypted execution environment corresponds to transmitting the ANN model to the target data providers by encrypting the ANN model as claimed.)
Regarding claim 7 depends on claim 1, thus the rejection of claim 1 is incorporated.
Pushkin teaches “The operating method of claim l, wherein the target data providers are configured to determine training data corresponding to the ANN model in a database, based on the data request information, and configured to train the ANN model based on the determined training data” (Column 4 lines 51-59 “requesting edge device 140 requests that a ML model be trained using federated learning and subsequently deployed. The request may include one or more of: an identifier of an initial model (e.g., a path to a model), an initial model, an identifier of one or more training clients (e.g., a list of training clients), a location of training data ... a training data distribution method identifier (e.g., a path to a distribution method, an identifier of a selection of a provided distribution method, etc.)”, Column 2 lines 55-58 “In some embodiments, the remote training service (or module) 124 utilizes data stored on the edge devices themselves (and therefore is not exposed to the provider network in any fashion) for training and/or validation”, and Column 5 lines 51-57 “In some embodiments, the local data of the edge devices 320(A)-(N) addresses different aspects of the model to be generated. For example, a first edge device may use local data that is specific for finding cats, a second edge device may use local data that is specific for finding dogs, etc. that are for a larger animal classifier. Different local data may address different layers of the model”. Pushkin discloses a workflow in which a requesting edge device submits a training request identifying an initial ML model and a location of training data, and a remote training service coordinates training using training clients (edge devices), wherein data stored locally on the training clients is utilized for training and different training clients may use local data corresponding to different aspects of the initial ML model. The training clients (edge devices) correspond to the claimed target data providers, the training request identifying the initial ML model and training-data location corresponds to the claimed data request information, the locally stored training data corresponds to the claimed training data in a database, and the use of such locally stored data to train the initial ML model corresponds to training the ANN model based on the determined training data, thereby teaching the target data providers determining training data corresponding to the ANN model in a database based on the data request information and training the ANN model based on the determined training data as claimed.)
Regarding claim 8 depends on claim 1, thus the rejection of claim 1 is incorporated.
Pushkin teaches “The operating method of claim 1, further comprising: determining the target data providers based on the data request information” (Column 4 lines 51-56 “a requesting edge device 140 requests that a ML model be trained using federated learning and subsequently deployed. The request may include one or more of: an identifier of an initial model (e.g., a path to a model), an initial model, an identifier of one or more training clients (e.g., a list of training clients)”, and Column 6 lines 12-16 “In this illustration, only one training client (edge device) 320(A) is illustrated in detail. As shown, a training algorithm 322 trains an initial model 326 using the local training data 324”. Pushkin discloses a workflow in which a requesting edge device submits a request to train an initial ML model, wherein the request may identify one or more training clients (edge devices), including a list of training clients, and the identified training clients are used to train the initial ML model using their respective local training data. The request corresponds to the claimed data request information, the identified training clients (edge devices) correspond to the claimed target data providers, and selecting the training clients identified in the request for training corresponds to determining the target data providers based on the data request information, thereby teaching determining the target data providers based on the data request information as claimed.)
Regarding claim 11 depends on claim 1, thus the rejection of claim 1 is incorporated.
Pushkin teaches “A computer program stored in a non-transitory computer-readable medium, the computer program being configured to perform the operating method of claim 1 in combination with hardware” (Column 8 lines 49-61 “ Some or all of the operations (or other processes described herein, or variations, and/or combinations thereof) are performed under the control of one or more computer systems configured with executable instructions and are implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications) executing collectively on one or more processors, by hardware or combinations thereof. The code is stored on a computer-readable storage medium, for example, in the form of a computer program comprising instructions executable by one or more processors.” Pushkin discloses the federated learning system is carried out via one or more computer systems configured with executable instructions and are implemented as code of one or more computer programs executing collectively on one or more processors, by hardware, wherein the code is stored on a computer-readable storage medium.)
Regarding claim 12,
Pushkin teaches “A central server apparatus comprising: a transmitter; a receiver; and a processor” (Column 8 lines 52-58 “The computer-readable storage media reader can be connected with, or configured to receive, a computer-readable storage medium, representing remote, local, fixed, and/or removable storage devices as well as storage media for temporarily and/or more permanently containing, storing, transmitting, and retrieving computer-readable information”, and Column 14 lines 17-19 “System memory 1020 may store instructions and data accessible by processor(s)” Pushkin discloses the computer-readable storage media reader can be configured to receive information, storage media for transmitting computer-readable information, and a processor, which corresponds to the transmitter, the receiver and the processor, as claimed.)
The applicant is directed to the rejection of claim 1 above for further rejection of claim 12, because the claim recites similar limitations and processing steps.
Regarding claim 13 depends on claim 12, thus the rejection of claim 12 is incorporated. Claim 13 is rejected under the same rationale as claim 2. The applicant is directed to the rejection of claim 2 above, because the claim recites similar limitations and processing steps.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 3-4, 14-15 are rejected under 35 U.S.C. 103 as being unpatentable over Pushkin et.al (US 12061956 B1) in view of Reyes et.al (US 20240127114 A1)
Regarding claim 3 depends on claim 1, thus the rejection of claim 1 is incorporated.
Pushkin does not teach “evaluating the result of the training provided by the target data providers”. However, Reyes teaches or at least suggests this (paragraph 150 “During the federated learning iteration, each of the first node 220, the second node 230 and the third node 240 is configured to determine a respective training parameter indicative of a level of predictive uncertainty of the respective trained models during training.” Reyes discloses a plurality of nodes participating in federated learning determine respective training parameters indicative of predictive uncertainty of their trained models during training. The nodes correspond to the claimed target data providers, the respective trained models correspond to the claimed results of training, and determining predictive uncertainty of the trained models corresponds to evaluating the training results, thereby teaching or suggesting evaluating the results of training provided by the target data providers as claimed.)
Pushkin does not teach “determining a level of contribution of the target data providers based on a result of the evaluation” However, Reyes teaches or at least suggests this (paragraph 166 “In one or more embodiments, the central server 210 determines a respective aggregation weight for each of the first trained model 320, the second trained model 330,” and paragraph 167 “The respective aggregation weights are indicative of a contribution of the respective model in the aggregated trained model” Reyes discloses a central server determines respective aggregation weights for trained models, wherein the aggregation weights indicate the respective contributions of the trained models to an aggregated trained model. The trained models correspond to the claimed training results, and the respective aggregation weights correspond to the claimed levels of contribution of the target data providers, thereby teaching or suggesting determining a level of contribution of the target data providers based on the result of the evaluation as claimed.)
Pushkin does not teach “determining compensation corresponding to the target data providers, based on the level of contribution” However, Reyes teaches or at least suggests this (paragraph 198 “The respective aggregation weights enable determining a relative contribution of each of the first training dataset 228 and the second training dataset 238 in the aggregated trained model 350. Thus, it will be appreciated that the providers of the first training dataset 228 and the second training dataset 238 may be compensated based on the relative contribution of their training datasets to the global training” Reyes discloses a workflow in which respective aggregation weights determine the relative contributions of training datasets to an aggregated trained model, and providers of the training datasets may be compensated based on their relative contributions to global training. The providers of the training datasets correspond to the claimed target data providers, the relative contributions correspond to the claimed levels of contribution, and compensation based on those relative contributions corresponds to the claimed compensation, thereby teaching or suggesting determining compensation corresponding to the target data providers based on their levels of contribution as claimed.)
Before the effective filing date, it would have been obvious to a person ordinary skill in the art to combine the teaching of federated learning service in a provider network and training machine learning models using external devices by Pushkin with the teaching of determining relative contribution of local datasets within federated learning by Reyes. The motivation to do so is disclosed by Reyes (paragraph 4 “Despite the benefits provided by distributed machine learning, data privacy and data aggregation are raising concerns addressed in various resource-constrained domains. For example, the communication costs incurred when updating deep learning models in mobile devices is expensive for most users as their internet bandwidths are typically low. In addition, the data used during the training of models in mobile devices is privacy-sensitive, and operations of raw data outside the portable devices are susceptible to attacks”, paragraph 11 “The present technology enables improving training of machine learning models by penalizing the model uncertainty at the client level to improve the robustness of the aggregated model”, and paragraph 12 “Further, the present technology enables determining the relative contribution of each of the local datasets to the aggregated training model and by taking into consideration and penalizing the uncertainty of each client in the aggregated result. For example, this enables determining, via the contribution of a training dataset, how the training data provided by a given entity has impacted the final model, without sharing the training dataset or disclosing potentially sensitive information included therein.” Reyes discloses that considering and penalizing model uncertainty at the client level improves the robustness of an aggregated model and enables determination of the relative contribution of each local training dataset, including how training data provided by a particular entity impacts the final model. A person ordinary skill in the art would have been motivated to apply Reyes’s uncertainty-based evaluation and contribution determination to Pushkin’s existing federated-learning system to further improve the robustness of the aggregated model and evaluate the contribution of participating clients. Furthermore, a person of ordinary skill in the art would have recognized that the teachings of Pushkin can be combined with Reyes because both are directed to federated learning in which multiple client devices locally train models and contribute to an aggregated model.)
Regarding claim 4 depends on claim 1, thus the rejection of claim 1 is incorporated.
Reyes teaches or at least suggests “obtaining an amount of training data used for the training from each of the target data providers” (paragraph 99 “It will be appreciated that the nature of the first training dataset 228 and the number of training data is not limited and depends on the task at hand. The training dataset may comprise any kind of digital file which may be processed by a machine learning model as described herein to generate predictions”, and paragraph 103 “Non-limiting examples of training datasets include MNIST, CIFAR-10, Fashion-MNIST and LIDC. The MNIST dataset consists of 70,000 grayscale images (28×28 pixels in size) which are divided in 60,000 training and 10,000 test samples” Reyes discloses a workflow in which respective training datasets comprising amounts of training data are used for training machine learning models, wherein the number of training data depends on the task and the training datasets may comprise known quantities of training samples. The respective training datasets correspond to the claimed training data of the target data providers, and the numbers of training samples correspond to the claimed amounts of training data used for training, thereby teaching or suggesting obtaining an amount of training data used for training from each of the target data providers as claimed.)
Reyes teaches or at least suggests “determining reliability corresponding to each of the target data providers, based on the result of the training” (paragraph 150 “During the federated learning iteration, each of the first node 220, the second node 230 and the third node 240 is configured to determine a respective training parameter indicative of a level of predictive uncertainty of the respective trained models during training” Reyes discloses nodes participating in federated learning determine respective training parameters indicative of predictive uncertainty of their trained models during training. The nodes correspond to the claimed target data providers, the respective trained models correspond to the claimed results of training, and the predictive uncertainty parameters correspond to the claimed reliability associated with the training results, thereby teaching or suggesting determining reliability corresponding to each of the target data providers based on the result of the training as claimed.)
Reyes teaches or at least suggests “evaluating the result of the training based on the amount of training data and the reliability” (paragraph 150 “During the federated learning iteration, each of the first node 220, the second node 230 and the third node 240 is configured to determine a respective training parameter indicative of a level of predictive uncertainty of the respective trained models during training.” Reyes discloses that during federated learning, respective training datasets comprising amounts of training data are used to train models at each node, and each participating nodes determine training parameters indicative of predictive uncertainty of their respective trained models during training. The respective amounts of training data correspond to the claimed amounts of training data used for training, and the predictive uncertainty parameters correspond to the claimed reliability, wherein the trained models are generated using the respective training datasets and evaluated based on their predictive uncertainty, thereby suggesting evaluating the results of training based on the amounts of training data and the reliability as claimed.)
The motivation to combine the teaching of Pushkin and Reyes is similar to the motivation as disclosed in claim 3 above.
Regarding claim 14 depends on claim 12, thus the rejection of claim 12 is incorporated. Claim 14 is rejected under the same rationale as claim 3. The applicant is directed to the rejection of claim 3 above, because the claim recites similar limitations and processing steps.
Regarding claim 15 depends on claim 12, thus the rejection of claim 12 is incorporated. Claim 15 is rejected under the same rationale as claim 4. The applicant is directed to the rejection of claim 4 above, because the claim recites similar limitations and processing steps.
Claims 9 is rejected under 35 U.S.C. 103 as being unpatentable over Pushkin et.al (US 12061956 B1) in view of Singhal et.al (US 20220398500 A1)
Regarding claim 9 depends on claim 8, thus the rejection of claim 8 is incorporated.
Pushkin does not teach “The operating method of claim 8, wherein the determining of the target data providers comprises: receiving, from a data provider, a number of pieces of training data corresponding to the data request information”. However, Singhal teaches or at least suggests this (paragraph 88 “In some implementations, the system additionally generates a weighting factor to be associated with the update to the set of global parameters based on a size of the local data maintained at the client computing device. For example, the system can determine the value of the weighting factor in proportion to a number of training examples generated from the query dataset”. Singhal discloses a federated learning system in which the system determines a weighting factor based on the size of local data maintained at a client computing device, wherein the size of the local data may be represented by a number of training examples. The client computing device corresponds to the claimed data provider, and the number of training examples corresponds to the claimed number of prices of training data, thereby teaches or at least suggests the receiving a number of pieces of training data corresponding to the data request information from a data provider, as claimed.)
Pushkin does not teach “determining, to be the target data providers, data providers of which the number of pieces of training data is greater than or equal to a predetermined threshold value”. However, Singhal teaches or at least suggests this (paragraph 96 “In some implementations, instead of combining all of the parameter update data that has been received, the system combines only the parameter update data received from certain client computing devices that each maintain more than a threshold amount of local data. This can stabilize the training.” Singhal discloses that the system may use parameter updates from client computing devices that each maintain more than a threshold amount of local data. The client computing devices correspond to the claimed data providers, the amount of local data corresponds to the claimed number of pieces of training data, and selecting clients having more than the threshold amount of local data corresponds to determining as target data providers those data providers having the number of pieces of training data is greater than or equal to a predetermined threshold value, thus Singhal teaches or at least suggests the limitation.)
Before the effective filing date, it would have been obvious to a person ordinary skill in the art to combine the teaching of federated learning service in a provider network and training machine learning models using external devices by Pushkin with the teaching of determining amount of local data and threshold by Singhal. The motivation to do so is referred to in Singhal’s disclosure (paragraph 96 “In some implementations, instead of combining all of the parameter update data that has been received, the system combines only the parameter update data received from certain client computing devices that each maintain more than a threshold amount of local data. This can stabilize the training.” Singhal discloses that using updates from clients maintaining more than a threshold amount of local data can stabilize the training. Furthermore, both Singhal and Pushkin use client devices having local training data to participate in federated learning. A person of ordinary skill in the art would have recognized that Singhal’s client-selected based on threshold of training data can be applied to Pushkin’s federated learning system because they both recite teaching of federated learning and the combination can further improve and stabilize the training process, thereby improving the deployed model by Singhal.)
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DUY TU DIEP whose telephone number is (703)756-1738. The examiner can normally be reached M-F 8-4:30.
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.
/DUY T DIEP/Examiner, Art Unit 2123
/ALEXEY SHMATOV/Supervisory Patent Examiner, Art Unit 2123