DETAILED ACTION
1. Claims 1-20 are pending.
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 .
Information Disclosure Statement
2. The information disclosure statement (IDS) submitted on May 10, 2023 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
3. The information disclosure statement (IDS) submitted on June 23, 2023 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Double Patenting
4. The nonstatutory double patenting rejection is based on a judicially created doctrine grounded in public policy (a policy reflected in the statute) so as to prevent the unjustified or improper timewise extension of the “right to exclude” granted by a patent and to prevent possible harassment by multiple assignees. A nonstatutory double patenting rejection is appropriate where the conflicting claims are not identical, but at least one examined application claim is not patentably distinct from the reference claim(s) because the examined application claim is either anticipated by, or would have been obvious over, the reference claim(s). See, e.g., In re Berg, 140 F.3d 1428, 46 USPQ2d 1226 (Fed. Cir. 1998); In re Goodman, 11 F.3d 1046, 29 USPQ2d 2010 (Fed. Cir. 1993); In re Longi, 759 F.2d 887, 225 USPQ 645 (Fed. Cir. 1985); In re Van Ornum, 686 F.2d 937, 214 USPQ 761 (CCPA 1982); In re Vogel, 422 F.2d 438, 164 USPQ 619 (CCPA 1970); In re Thorington, 418 F.2d 528, 163 USPQ 644 (CCPA 1969).
A timely filed terminal disclaimer in compliance with 37 CFR 1.321(c) or 1.321(d) may be used to overcome an actual or provisional rejection based on nonstatutory double patenting provided the reference application or patent either is shown to be commonly owned with the examined application, or claims an invention made as a result of activities undertaken within the scope of a joint research agreement. See MPEP § 717.02 for applications subject to examination under the first inventor to file provisions of the AIA as explained in MPEP § 2159. See MPEP § 2146 et seq. for applications not subject to examination under the first inventor to file provisions of the AIA . A terminal disclaimer must be signed in compliance with 37 CFR 1.321(b).
The filing of a terminal disclaimer by itself is not a complete reply to a nonstatutory double patenting (NSDP) rejection. A complete reply requires that the terminal disclaimer be accompanied by a reply requesting reconsideration of the prior Office action. Even where the NSDP rejection is provisional the reply must be complete. See MPEP § 804, subsection I.B.1. For a reply to a non-final Office action, see 37 CFR 1.111(a). For a reply to final Office action, see 37 CFR 1.113(c). A request for reconsideration while not provided for in 37 CFR 1.113(c) may be filed after final for consideration. See MPEP §§ 706.07(e) and 714.13.
The USPTO Internet website contains terminal disclaimer forms which may be used. Please visit www.uspto.gov/ patent/patents-forms. The actual filing date of the application in which the form is filed determines what form (e.g., PTO/SB/25, PTO/SB/26, PTO/AIA /25, or PTO/AIA /26) should be used. A web-based eTerminal Disclaimer may be filled out completely online using web-screens. An eTerminal Disclaimer that meets all requirements is auto-processed and approved immediately upon submission. For more information about eTerminal Disclaimers, refer to www.uspto.gov/ patents/apply/applying-online/eterminal-disclaimer.
5. Claim 1 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 1 of US Patent No. 11681945 B2, in view of Zhang et al. US 20170300693 A1. Although the claims at issue are not identical, they are not patentably distinct from each other because the claims in the instant application are broader version of the allowed claims. Therefore, the claim of the patent anticipates the broader claims of the instant application. The differences are italicized below, and the similarities are highlighted.
Examined Application
Patent No. US 11681945 B2
Claim 1:
A method comprising:
receiving, at an edge node, a machine learning model from a cloud, wherein the cloud provides the machine learning model with its initial training, the machine learning model being used to optimize performance of at least the edge node;
monitoring resources being used at the edge node, wherein a threshold amount of resources is identified as being an amount of resources needed for normal operations of the edge node;
identifying an amount of spare resources that are available at the edge node, wherein the amount of spare resources corresponds to a difference between a current amount of resources being used at the edge node that is less than the amount of resources needed for normal operations of the edge node;
allocating the identified amount of spare resources for metered training the machine learning model of the edge node, wherein the metered training prioritizes the normal operation of the edge node over training the machine learning model; and
training the machine learning model of the edge node using the identified amount of spare resources.
Claim 1:
A computer-implemented method for training a fog node in a fog layer, the method comprising:
receiving a machine learning model for the fog node from a cloud, wherein the cloud provides the machine learning model with its initial training, the machine learning model being used to optimize performance of at least the fog node at the fog layer;
monitoring resources being used at the fog node, wherein a threshold amount of resources is identified as being an amount of resources needed for normal operations of the fog node;
identifying an amount of spare resources that are available, wherein the amount of spare resources corresponds to a difference between a current amount of resources being used that is less than the amount of resources needed for normal operations of the fog node;
allocating the identified amount of spare resources for metered training of the machine learning model of the fog node, wherein: the metered training prioritizes normal operation of the fog node over training the machine learning model; and
the metered training of the machine learning model is performed using a sampled data set that is based on the amount of spare resources available to the fog node; and
training the machine learning model of the fog node using the identified amount of spare resources, wherein the training of the machine learning model is performed so long as the amount of spare resources are available at the fog node.
6. Claim 1 of the reference patent recites all of the limitations of claim 1 of the instant application except “edge node” However, Zhang et al. teaches: [0054] Referring now to FIGS. 5A-5B, examples are illustrated of fog/edge node [...]. It would have been obvious to person of ordinary skill in the art before the effective filing date of the claimed invention equate fog nodes with edge nodes. Both fog and edge nodes operate at the edge of a network. Also as shown in FIG. 3A, any number of Edge/Fog/Root devices/nodes (hereinafter “edge devices”) may be in communication with server 150, such as via WAN 130 or another backbone network. Any number of additional nodes (e.g., sensors, actuators, etc.) may be connected to the edge devices and communicate therewith. As would be appreciated, the network may include any number of edge devices (e.g., a first through nth edge device), each of which having any number of attached nodes/devices in its local network (e.g., sensors, actuators, etc.), as discussed in Zhang ([0039]).
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.
7. Claims 1-2, 4-5, 11, 13, 16, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Satou US 20200101603 A1 in view of Ghibril et al. US 20190164087 A1.
Satou was cited in IDS filed on May 10, 2023.
8. With regard to claim 1, Satou teaches:
A method comprising:
receiving, at an edge node, a machine learning model from a cloud, wherein the cloud provides the machine learning model with its initial training, the machine learning model being used to optimize performance of at least the edge node ([0059] The following description will focus on second to fourth embodiments in which the aforementioned controller 1 according to the first embodiment is implemented as a part of a system in which a plurality of devices including a cloud server, a host computer, fog computers, and edge computers (such as a robot controller and the controller) are connected to each other via a wired/wireless network; [0061] In such a system, the controller 1 according to an embodiment of the present invention can be implemented on any of the cloud server 6, the fog computer 7, and the edge computer 8, so that data for use in machine learning can be shared among the plurality of devices via the network for distributed learning, the generated learning model can be collected in the fog computer 7 and the cloud server 6 for large-scale analysis, and further the generated learning model can be mutually reused; [0072] In this control system 500, the controller 1′ having the machine learning device 100 estimates the adjustment of the control command of the manipulator in the industrial robot 2′ using the learning result of the learning unit 110. Further, the control system 500 can be configured such that at least one controller 1′ learns the adjustment of the control command of the manipulator in each of the industrial robots 2 and 2′ common to all the controllers 1 and 1′ based on the state variable S and the determination data D obtained by each of the other plurality of controllers 1 and 1′ and all the controllers 1 and 1′ share the learning results. Therefore, the control system 500 can improve learning speed and reliability by using more diverse data sets (including the state variable S and the determination data D) as the input; [0074] A control system 500′ of the present embodiment comprises at least one machine learning device 100 (illustrated as an example implemented as a part of the fog computer 7 in FIG. 9) implemented as a part of a computer such as a cloud server, a host computer, and a fog computer, a plurality of controllers 1″, and the network 5 connecting these controllers 1″ and the computer to each other. Note that the hardware configuration of the computer is the same as the schematic hardware configuration of the controller 1′ illustrated in FIG. 7 such that the hardware components such as the CPU 311, the RAM 313, and the non-volatile memory 314 provided in a general computer are connected through the bus 320; [0075] In the control system 500′ having the aforementioned configuration, based on the state variable S and the determination data D obtained from each of the plurality of controllers 1″, the machine learning device 100 learns the adjustment of the control command of the manipulator in the industrial robot 2 common to all the controllers 1″, and then by using the learning result, can perform the adjustment of the control command of the manipulator in each industrial robots 2. According to the configuration of the control system 500′, when needed, the necessary number of controllers 1″ can be connected to the machine learning device 100 regardless of where and when each of the plurality of controllers 1″ exists.);
Although Satou teaches of a fog computer that contains a machine learning device that learns optimal adjustments to operate a robot controller. The fog computer is optimized using the knowledge from the machine learning models. However, Satou fails to explicitly teach of monitoring resources being used at the edge node, wherein a threshold amount of resources is identified as being an amount of resources needed for normal operations of the edge node; identifying an amount of spare resources that are available at the edge node, wherein the amount of spare resources corresponds to a difference between a current amount of resources being used at the edge node that is less than the amount of resources needed for normal operations of the edge node; allocating the identified amount of spare resources for metered training the machine learning model of the edge node, wherein the metered training prioritizes the normal operation of the edge node over training the machine learning model; and training the machine learning model of the edge node using the identified amount of spare resources.
Ghibril was cited in IDS filed on May 10, 2023.
However, in analogous art, Ghibril teaches:
monitoring resources being used at the edge node, wherein a threshold amount of resources is identified as being an amount of resources needed for normal operations of the edge node ([0031] An edge device is a type of computing device 130 having resources to host at least one service component of an information service and also function as an access point to a network for providing the information service. The resources in the edge device can be either hardware or software that are configurable based on a command received from an external source (e.g., an orchestrator 140). Example edge devices may include micro data centers, edge routers, provider edge routers, aggregation routers, customer premise equipment (CPE), set-top boxes, cloudlets, fog nodes, wireless access points, wireless base stations, Long Term Evolution (LTE) protocol nodes such as an Evolved Node B, cable modems, DSL modems, optical termination points, reconfigurable optical add-drop multiplexer (ROADM), road side units, onboard computers, connected vehicles, satellite receivers, ground stations, digital subscriber line access multiplexer (DSLAM), switches, cable modem termination system (CMTS), broadband gateways, among others; [0038] FIG. 4 is a diagram of various types of intelligent entities 120 according to one embodiment. Example types of intelligent entities 120 include a Fault IE, Capacity IE, Performance IE, Security IE, Inventory IE, and Alarm IE. The intelligent entities 120 may communicate with various management components (e.g., fault management, capacity management, etc.) of the OSS management and orchestration 410 to perform different types of tasks. Particularly, the Fault IE may identify and determine solutions to issues for fault management. The Capacity IE may track workload or resources of computing devices 130 and predict when additional capacity should be allocated to support an increase in demand. The Performance IE may monitor performance metrics of computing devices 130 such as latency, memory usage, CPU usage, network bandwidth, etc. The Security IE protects the infrastructure 400 from unauthorized activity and may detect anomalies in the system. The Inventory IE manages inventory of the computing devices 130 or other components in the system. The Alarm IE generates and transmits alarms responsive to determining that a given event has occurred (e.g., commissioning or decommissioning of a computing device 130) or that a certain condition has been satisfied (e.g., resource usage has reached at least a threshold level of capacity).);
identifying an amount of spare resources that are available at the edge node, wherein the amount of spare resources corresponds to a difference between a current amount of resources being used at the edge node that is less than the amount of resources needed for normal operations of the edge node ([0038] The Capacity IE may track workload or resources of computing devices 130 and predict when additional capacity should be allocated to support an increase in demand; [0043] For instance, responsive to determining a negative impact or reduction in available resources, the Predictive Auto-scaling IE determines to commission additional resources or turn off existing virtual machines or computing devices 130 to release lower priority or unused resources; [0051] The resource tracker 712 and provisioning module 714 operate on the control plane of the network 180. The resource tracker 712 monitors the resources 718 of the computing device 130. The resource tracker 712 may track, for example, current and historical demand for the resources 718, assignments of service components to the resources 718, performance requirements of service components, or characteristics of the resources 718. Types of the characteristics may include compute characteristics (e.g., CPU type, number of CPUs, CPU speed or latency, etc.), storage characteristics (e.g., volatile or non-volatile memory, storage volume in gigabytes or terabytes, read and write latency, etc.), networking characteristics (e.g., number of interfaces and network speed), node geographical location (e.g., jurisdiction, country, or longitude and latitude coordinates of the computing devices), node connectivity (e.g., nearby computing devices, connection speed, etc.), and access connectivity (e.g., fiber connection, radiofrequency access, spectrum, bandwidth, cell identification, etc.), among other characteristics. The resource tracker 712 may provide the tracked resource information to an orchestrator 140 or the provisioning module 714.);
allocating the identified amount of spare resources for metered training the machine learning model of the edge node, wherein the metered training prioritizes the normal operation of the edge node over training the machine learning model ([0043] The first channel 600 transmits information 650 associated with the alarm to the Predictive Auto-scaling IE responsive to receiving the parameters from the Alarm RCA IE. The Predictive Auto-scaling IE may determine a reallocation of resources to account for the identified outage on the network component. In some embodiments, the Predictive Auto-scaling IE may use a machine learning model to predict an impact on resource utilization as result of the outage. Based on the prediction, the Predictive Auto-scaling IE can determine an appropriate action or no action. For instance, responsive to determining a negative impact or reduction in available resources, the Predictive Auto-scaling IE determines to commission additional resources or turn off existing virtual machines or computing devices 130 to release lower priority or unused resources. In some embodiments, the Predictive Auto-scaling IE may check with another IE before performing an action. For example, the Predictive Auto-scaling IE checks with a Social Media IE to determine whether resource capacity should be maintained for an upcoming social event; [0053] FIG. 8 is a flow chart illustrating a process 800 for performing predictions using intelligent entities, according to one embodiment. In an example use case, an intelligent element framework 110 chains a Traffic IE, User Activity IE, Resource Prediction IE, and Recommendation IE to perform energy management. The Traffic IE determines trends 810 in resource demand. The Traffic IE may use a model trained using one or more features to determine trends. For instance, the features indicate metrics associated with available resources or previous resource demand, e.g., a growth or decrease in demand using historical traffic data. Accordingly, the Traffic IE may learn to predict that similar or different trends may occur in the future given certain conditions. In some embodiments, the features are generated by at least one other intelligent entity 120.); and
training the machine learning model of the edge node using the identified amount of spare resources ([0054] The User Activity IE predicts user activity 820, e.g., using social media information or historical user movement. The User Activity IE may also use a model trained using one or more features to generate predictions of user activity. The model used by the User Activity IE may be different than a model used by the Traffic IE. Generally, intelligent entities 120 may use different models from each other, models trained with different training data, or models trained using different machine learning algorithms. The User Activity IE may predict user activity such as usage levels of computing devices 130, periods of time with relatively greater or less traffic on a network, locations to which users are likely to travel, or aggregate activity from a population of users.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Satou with the teachings of Ghibril of monitoring resources being used at the edge node, wherein a threshold amount of resources is identified as being an amount of resources needed for normal operations of the edge node; identifying an amount of spare resources that are available at the edge node, wherein the amount of spare resources corresponds to a difference between a current amount of resources being used at the edge node that is less than the amount of resources needed for normal operations of the edge node; allocating the identified amount of spare resources for metered training the machine learning model of the edge node, wherein the metered training prioritizes the normal operation of the edge node over training the machine learning model; and training the machine learning model of the edge node using the identified amount of spare resources. As described above, Satou teaches of a fog computer that contains a machine learning device that learns optimal adjustments to operate a robot controller. The fog computer is optimized using the knowledge from the machine learning models. Similarly, Ghibril teaches of using a machine learning model to commission additional resources ([0043]). Specifically, Ghibril teaches of modular approach of multiple intelligent entities that are able to distribute tasks to be executed by machine learning algorithms. These services can be used for robot control ([0018]). More importantly, this enables devices, such as fog devices, to be able to adapt to changes in resource allocation, as discussed in Ghibril ([0043]). Together, Satou and Ghibril teach of optimizing performance of an edge node in a cloud environment by identifying spare resources available and allocating the spare resources for training the machine learning model.
9. With regard to claim 2, Ghibril further teaches:
wherein the monitoring is performed by a system performance monitor that is installed on the edge node ([0046] FIG. 7 is a block diagram of a computing device 130, according to one embodiment. The computing device 130 includes, among others, a storage medium 700, one or more processors 702, one or more network interfaces 704, a storage controller 706, one or more hardware components 708, and a bus 710 connecting these components. Hardware components 708 may include, for example, sensors, antennas, GPUs, display devices, I/O interfaces, etc. The one or more processors 702 execute instructions stored in the storage medium 700. The one or more network interfaces 704 are configured to communicatively connect the computing device 130 over the network 180 to external systems 170, computing devices 130, orchestrators 140, or other components; [0031] An edge device is a type of computing device 130 having resources to host at least one service component of an information service and also function as an access point to a network for providing the information service. The resources in the edge device can be either hardware or software that are configurable based on a command received from an external source (e.g., an orchestrator 140). Example edge devices may include micro data centers, edge routers, provider edge routers, aggregation routers, customer premise equipment (CPE), set-top boxes, cloudlets, fog nodes, wireless access points, wireless base stations, Long Term Evolution (LTE) protocol nodes such as an Evolved Node B, cable modems, DSL modems, optical termination points, reconfigurable optical add-drop multiplexer (ROADM), road side units, onboard computers, connected vehicles, satellite receivers, ground stations, digital subscriber line access multiplexer (DSLAM), switches, cable modem termination system (CMTS), broadband gateways, among others; [0051] The resource tracker 712 and provisioning module 714 operate on the control plane of the network 180. The resource tracker 712 monitors the resources 718 of the computing device 130. The resource tracker 712 may track, for example, current and historical demand for the resources 718, assignments of service components to the resources 718, performance requirements of service components, or characteristics of the resources 718. Types of the characteristics may include compute characteristics (e.g., CPU type, number of CPUs, CPU speed or latency, etc.), storage characteristics (e.g., volatile or non-volatile memory, storage volume in gigabytes or terabytes, read and write latency, etc.), networking characteristics (e.g., number of interfaces and network speed), node geographical location (e.g., jurisdiction, country, or longitude and latitude coordinates of the computing devices), node connectivity (e.g., nearby computing devices, connection speed, etc.), and access connectivity (e.g., fiber connection, radiofrequency access, spectrum, bandwidth, cell identification, etc.), among other characteristics. The resource tracker 712 may provide the tracked resource information to an orchestrator 140 or the provisioning module 714.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Satou with the teachings of Ghibril wherein the monitoring is performed by a system performance monitor that is installed on the edge node. Together, Satou and Ghibril teach of optimizing performance of an edge node in a cloud environment by identifying spare resources available and allocating the spare resources for training the machine learning model. Additionally, it would be obvious to one of ordinary skill in the art that there is some type of monitor that monitors resource levels in order to optimize performance of an edge node in a cloud environment and allocating resources accordingly. Ghibril teaches of a resource tracker that monitors resources of the computing device, which can be an edge device. This helps the system track current and historical demand for resources and allow resources to be provisioned accordingly, as discussed in Ghibril ([0050]; [0051]).
10. With regard to claim 4, Ghibril further teaches:
wherein the sampled data set is based on the amount of spare resources available to the edge node ([0055] The Traffic IE and User Activity IE may provide their outputs, resource trends and user activity predictions, respectively, to the Resource Prediction IE. In particular, the outputs may be provided via the intelligent element framework 110 that chains the Traffic IE and User Activity IE to the Resource Prediction IE, e.g., using one or more channels. The Resource Prediction IE predicts resource demand 830 for a cell (e.g., a computing device 130) using the input from the other IEs.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Satou with the teachings of Ghibril wherein the sampled data set is based on the amount of spare resources available to the edge node. Together, Satou and Ghibril teach of optimizing performance of an edge node in a cloud environment by identifying spare resources available and allocating the spare resources for training the machine learning model. Additionally, it would be obvious to one of ordinary skill in the art that the sampled data set used to train the machine learning model includes data regarding the amount of spare resources available. This data is needed to optimize performance of an edge node in a cloud environment and to allocate resources accordingly. Ghibril teaches using the input from the IEs, such as resource trends and user activity predictions as input data in order to train a machine learning model. This helps the system track current and historical demand for resources and allow resources to be provisioned accordingly, as discussed in Ghibril ([0050]; [0051]; [0055]; [0056]).
11. With regard to claim 5, Ghibril further teaches
further comprising:
automatically terminating a current training session of the machine learning model when the spare resources are no longer available at the edge node ([0056] The Recommendation IE determines a recommendation 850 to turn on or off management components of the cell. For example, responsive to a prediction that resource demand and user activity is predicted to decrease during a given time period (e.g., the weekend), the Recommendation IE recommends to turn off at least a portion of the cell to preserve energy. In some embodiments, the Recommendation IE may determine other types of recommendations, for example, requesting an intervention to mitigate an identified fault or alert, commissioning new management components or reconfiguring existing management components, or triggering other artificial intelligence tasks.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Satou with the teachings of Ghibril of automatically terminating a current training session of the machine learning model when the spare resources are no longer available at the edge node. Together, Satou and Ghibril teach of optimizing performance of an edge node in a cloud environment by identifying spare resources available and allocating the spare resources for training the machine learning model. Moreover, one of ordinary skill in the art would recognize that optimization would include not exhausting resources if spare resources are no longer available. Ghibril teaches of terminating current activity in order to preserve energy ([0056]).
12. Regarding claim 11, it is rejected under the same reasoning as claim 1 above. Therefore, it is rejected under the same rationale.
13. Regarding claim 13, it is rejected under the same reasoning as claim 4 above. Therefore, it is rejected under the same rationale.
14. Regarding claim 16, it is rejected under the same reasoning as claim 1 above. Therefore, it is rejected under the same rationale.
15. With regard to claim 18, Ghibril further teaches:
wherein the edge node is further configured to monitor the edge node to identify when spare resources are no longer available ([0056] The Recommendation IE determines a recommendation 850 to turn on or off management components of the cell. For example, responsive to a prediction that resource demand and user activity is predicted to decrease during a given time period (e.g., the weekend), the Recommendation IE recommends to turn off at least a portion of the cell to preserve energy. In some embodiments, the Recommendation IE may determine other types of recommendations, for example, requesting an intervention to mitigate an identified fault or alert, commissioning new management components or reconfiguring existing management components, or triggering other artificial intelligence tasks.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Satou with the teachings of Ghibril of wherein the edge node is further configured to monitor the edge node to identify when spare resources are no longer available. Together, Satou and Ghibril teach of optimizing performance of an edge node in a cloud environment by identifying spare resources available and allocating the spare resources for training the machine learning model. Moreover, one of ordinary skill in the art would recognize that optimization would include not exhausting resources if spare resources are no longer available. Ghibril teaches of terminating current activity in order to preserve energy ([0056]).
16. Claims 3, 12, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Satou US 20200101603 A1 and Ghibril et al. US 20190164087 A1, as applied in claim 1, in further view of Dai et al. US 20190147540 A1.
17. With regard to claim 3, Satou further teaches:
wherein the training of the machine learning model is performed using a sampled data set ([0072] In this control system 500, the controller 1′ having the machine learning device 100 estimates the adjustment of the control command of the manipulator in the industrial robot 2′ using the learning result of the learning unit 110. Further, the control system 500 can be configured such that at least one controller 1′ learns the adjustment of the control command of the manipulator in each of the industrial robots 2 and 2′ common to all the controllers 1 and 1′ based on the state variable S and the determination data D obtained by each of the other plurality of controllers 1 and 1′ and all the controllers 1 and 1′ share the learning results. Therefore, the control system 500 can improve learning speed and reliability by using more diverse data sets (including the state variable S and the determination data D) as the input.).
Satou teaches of training the machine learning model using input data and data sets. Similarly, Dai teaches:
[0011] In some embodiments, the user type determination model is trained and obtained by: acquiring an initial user type determination model and a predetermined first sample data set, wherein each piece of sample data in the first sample data set includes at least one personal attribute characteristic of a user and a user type of the user under the preset attribute; using the at least one personal attribute characteristic of the user in each piece of sample data in the first sample data set as input data, and the user type of the user under the preset attribute in the sample data as corresponding output data to train the initial user type determination model using a machine learning method; and defining the trained initial user type determination model as the pre-trained user type determination model.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Satou with the teachings of Dai wherein the training of the machine learning model is performed using a sampled data set. Dai reasonably shows that input data and sample data are analogous. Sample data is used as input data in order to train a machine learning model.
18. With regard to claim 12, Satou further teaches:
wherein the execution of the computer-readable instructions further cause the edge node to control a sampling of data being used to train the machine learning model ([0072] In this control system 500, the controller 1′ having the machine learning device 100 estimates the adjustment of the control command of the manipulator in the industrial robot 2′ using the learning result of the learning unit 110. Further, the control system 500 can be configured such that at least one controller 1′ learns the adjustment of the control command of the manipulator in each of the industrial robots 2 and 2′ common to all the controllers 1 and 1′ based on the state variable S and the determination data D obtained by each of the other plurality of controllers 1 and 1′ and all the controllers 1 and 1′ share the learning results. Therefore, the control system 500 can improve learning speed and reliability by using more diverse data sets (including the state variable S and the determination data D) as the input.).
Satou teaches of training the machine learning model using input data and data sets. Similarly, Dai teaches:
[0011] In some embodiments, the user type determination model is trained and obtained by: acquiring an initial user type determination model and a predetermined first sample data set, wherein each piece of sample data in the first sample data set includes at least one personal attribute characteristic of a user and a user type of the user under the preset attribute; using the at least one personal attribute characteristic of the user in each piece of sample data in the first sample data set as input data, and the user type of the user under the preset attribute in the sample data as corresponding output data to train the initial user type determination model using a machine learning method; and defining the trained initial user type determination model as the pre-trained user type determination model.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Satou with the teachings of Dai wherein the execution of the computer-readable instructions further cause the edge node to control a sampling of data being used to train the machine learning model. Dai reasonably shows that input data and sample data are analogous. Sample data is used as input data in order to train a machine learning model.
19. Regarding claim 17, it is rejected under the same reasoning as claim 12 above. Therefore, it is rejected under the same rationale.
20. Claims 6-8 are rejected under 35 U.S.C. 103 as being unpatentable over Satou US 20200101603 A1 and Ghibril et al. US 20190164087 A1, as applied in claim 1, in further view of Prakash et al. US 20190138934 A1.
21. With regard to claim 6, Satou and Ghibril teach the method of claim 5 but fail to explicitly teach further comprising: automatically terminating future training sessions and completing current training sessions of the machine learning model when the identified amount of spare resources available at the edge node falls below a pre-determined threshold amount.
Prakash was cited in IDS filed on May 10, 2023.
However, in analogous art, Prakash teaches:
further comprising:
automatically terminating future training sessions and completing current training sessions of the machine learning model when the identified amount of spare resources available at the edge node falls below a pre-determined threshold amount ([0043] For edge-cloud ML or distributed learning, ML training is performed on a dataset to learn parameters of an underlying model β, where the dataset and computational tasks of the ML training process are distributed across a plurality of edge nodes 101, 201... By off-loading ML training tasks to individual edge nodes 101, 201, the ML training process may be accelerated and/or may provide a more efficient use of computational resources; [0044] When all selected edge nodes 101, 201 that were connected to a specific instance of the training process (model) β have disconnected, the instance of the training process (model) β may be terminated; [0048] The operational parameters of the edge compute nodes 101, 201 includes compute node capabilities and operational constraints or contexts. The compute node capabilities may include, for example, configuration information (e.g., a hardware platform make and model, hardware component types and arrangement within the hardware platform, associated peripheral and/or attached devices/systems, processor architecture, currently running operating systems and/or applications and/or their requirements, subscription data (e.g., data plan and permissions for network access), security levels or permissions (e.g., possible authentication and/or authorization required to access the edge compute node 101, 201), etc.); computational capacity (e.g., a total processor speed of one or more processors, a total number of VMs capable of being operated by the edge compute node 101, 201, a memory or storage size, an average computation time per workload, a reuse degree of computational resources, etc.); current or predicted computational load and/or computational resources (e.g., processor utilization or occupied processor resources, memory or storage utilization, etc.); current or predicted unoccupied computational resources (e.g., available or unused memory and/or processor resources, available VMs, etc.); network capabilities (e.g., link adaptation capabilities, configured and/or maximum transmit power, achievable data rate per channel usage, antenna configurations, supported radio technologies or functionalities of a device (e.g., whether a UE 101 supports Bluetooth/BLE; whether an (R)AN node 111 supports LTE-WLAN aggregation (LWA) and/or LTE/WLAN Radio Level Integration with IPsec Tunnel (LWIP), etc.), subscription information of particular UEs 101, etc.); energy budget (e.g., battery power budget); and/or other like capabilities; [0049] the operational contexts and/or constraints may be based on a pre-assessment of an operational state of the edge compute nodes 101, 102, which may be based on previously indicated operational contexts and/or constraints for different offloading opportunities. This may involve, for example, evaluating both computation and communication resources needed for different offloading opportunities. The threshold criteria or a desired level of reliability mentioned previously may be based on a certain amount or type of compute node capabilities (e.g., a certain processor speed) and/or a type of operational constraints under which the compute node is operating (e.g., a desired link quality, a desired surrounding temperature, a desired processor temperature, etc.).).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Satou and Ghibril with the teachings of Prakash of automatically terminating future training sessions and completing current training sessions of the machine learning model when the identified amount of spare resources available at the edge node falls below a pre-determined threshold amount. Together, Satou and Ghibril teach of optimizing performance of an edge node in a cloud environment by identifying spare resources available and allocating the spare resources for training the machine learning model. Similarly, Prakash teaches of machine learning training in a heterogeneous computing environment with nodes (Abstract; [0021]). Moreover, Prakesh teaches of terminating a future training session and finishing the current training sessions when resources fall below a pre-determined threshold. This ensures that certain levels of reliability are met, and that resources are allocated accordingly. This may involve, for example, evaluating both computation and communication resources needed for different offloading opportunities. The threshold criteria or a desired level of reliability mentioned previously may be based on a certain amount or type of compute node capabilities (e.g., a certain processor speed) and/or a type of operational constraints under which the compute node is operating (e.g., a desired link quality, a desired surrounding temperature, a desired processor temperature, etc.), as discussed in Prakesh ([0049]).
22. With regard to claim 7, Prakesh further teaches:
wherein the training of the machine learning model is performed so long as a pre-determined threshold amount of spare resources are available at the edge node ([0045] As an example, the load balancing policy may define the particular type or types of operational parameters (discussed infra) that should be collected by the MEC system 200. In another example, the load balancing policy may define criteria to be used by the MEC system 200 for determining threshold criteria or a desired level of reliability for selecting a particular edge compute node 101, 201 to perform computational tasks β. In this example, the threshold criteria may be based on a desired epoch time for computing a full gradient from obtained partial gradients from each edge compute node 101, 201. In another example, the load balancing policy may define criteria (e.g., load allocation criteria) to be used by the MEC system 200 for determining how to partition the training data into different datasets x.sub.1-x.sub.m.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Satou and Ghibril with the teachings of Prakash of wherein the training of the machine learning model is performed so long as a pre-determined threshold amount of spare resources are available at the edge node. Together, Satou and Ghibril teach of optimizing performance of an edge node in a cloud environment by identifying spare resources available and allocating the spare resources for training the machine learning model. Similarly, Prakash teaches of machine learning training in a heterogeneous computing environment with nodes (Abstract; [0021]). Moreover, Prakesh teaches of training machine learning models as long as spare resources are available. This ensures that certain levels of reliability are met, and that resources are allocated accordingly. This may involve, for example, evaluating both computation and communication resources needed for different offloading opportunities. The threshold criteria or a desired level of reliability mentioned previously may be based on a certain amount or type of compute node capabilities (e.g., a certain processor speed) and/or a type of operational constraints under which the compute node is operating (e.g., a desired link quality, a desired surrounding temperature, a desired processor temperature, etc.), as discussed in Prakesh ([0049]).
23. With regard to claim 8, Prakesh further teaches:
wherein the initial training of the machine learning model uses default values or past machine learning models of similar edge nodes stored within the cloud ([0004] Many forms of machine learning (ML), such as supervised learning, perform a training process on a relatively large dataset to estimate an underlying ML model. Linear regression is one type of supervised ML algorithm that is used for classification, stock market analysis, weather prediction, and the like. Gradient descent (GD) algorithms are often used in linear regression. Given a function defined by a set of parameters, a GD algorithm starts with an initial set of parameter values, and iteratively moves toward a set of parameter values that minimize the function. This iterative minimization is achieved by taking steps in the negative direction of the function gradient. Example use cases for GD algorithms include localization in wireless sensor networks and distributed path-planning for drones; [0021] The present disclosure is related to distributed machine learning (ML) in distributed heterogeneous computing environments, where computational resources of multiple edge compute nodes are utilized for collaborative learning for an underlying ML model. Distributed heterogeneous computing environments are computing environments where compute (processing) and storage resources are available at multiple edge compute nodes, with varying capabilities and operational constraints. Generally, an ML algorithm is a computer program that learns from experience with respect to some task and some performance measure, and an ML model may be any object or data structure created after an ML algorithm is trained with one or more training datasets. After training, an ML model may be used to make predictions on new datasets. Although the term “ML algorithm” refers to different concepts than the term “ML model,” these terms as discussed herein may be used interchangeably for the purposes of the present disclosure; [0099] Gradient descent (GD) is an optimization algorithm used to minimize a target function by iteratively moving in the direction of a steepest descent as defined by a negative of the gradient. An objective of GD in machine learning (ML) is to utilize a training dataset D in order to accurately estimate the unknown model β over one or more epochs r. In ML, GD is used to update the parameters of the unknown model β. Parameters refer to coefficients in linear regression and weights in a neural network. These objectives are realized in an iterative fashion by computing β.sup.(r) at the r-th epoch, and evaluating a gradient associated with the squared-error cost function defined by f (β.sup.(r))=∥Xβ.sup.(r)−Y∥.sup.2. The cost function indicates how accurate the model β is at making predictions for a given set of parameters. The cost function has a corresponding curve and corresponding gradients, where the slope of the cost function curve indicates how the parameters should be changed to make the model β more accurate. In other words, the model β is used to make predictions, and the cost function is used to update the parameters for the model β).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Satou and Ghibril with the teachings of Prakesh wherein the initial training of the machine learning model uses default values or past machine learning models of similar edge nodes stored within the cloud. Together, Satou and Ghibril teach of optimizing performance of an edge node in a cloud environment by identifying spare resources available and allocating the spare resources for training the machine learning model. Similarly, Prakash teaches of machine learning training in a heterogeneous computing environment with nodes (Abstract; [0021]). Moreover, Prakesh teaches multiple machine learning models that can be used to train a current machine learning model. This ensures that the model is able to achieve high accuracy with minimal iterations by adapting to changes in real time, as described by the process outlined in Prakesh ([0099]).
24. Claims 9, 14, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Satou US 20200101603 A1 and Ghibril et al. US 20190164087 A1, as applied in claim 1, in further view of Wang et al. US 11521090 B2.
25. With regard to claim 9, Satou and Ghibril teach the method of claim 1 but fail to explicitly teach wherein after being trained, the machine learning model is shared with at least one other edge node having a threshold similarity to the edge node.
However, in analogous art, Wang teaches:
wherein after being trained, the machine learning model is shared with at least one other edge node having a threshold similarity to the edge node (Col. 6, lines 38-64, FIGS. 5A-5B depict a method 500 for distributed machine learning, according to an exemplary embodiment. At 502, a model requester 306 (shown in FIGS. 3 and 4) generates a specification (such as input data format, number of output classes, etc.) of a machine learning model for which the model requester wants to accomplish a machine learning task. At 504, the model requester 306 performs one or more steps of preliminary training of the machine learning model, using its own local dataset. Then, at 506 the model requester 306 sends the specification to other edge nodes 304 (generally, any desired number of edge nodes such as 304a, 304b, 304c . . . ). In a brokered architecture as shown in FIG. 3, the model requester 306 sends the specification via the broker 302; otherwise, the model requester sends the specification directly to the other edge nodes; At 508, each edge node checks whether it possesses data that matches with the specification (e.g., if a node only has audio data, it cannot participate in a task for image classification), and at 510 each edge node checks whether it has enough computation and communication resources to run its part of a distributed learning task. At 512, the model requester receives replies from each of the edge nodes 304, based on the checking steps, regarding each edge node's ability to match the specification and to participate in the model requester's distributed learning task. Based on the replies, the model requester node 306 identifies the participating edge nodes.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Satou and Ghibril with the teachings of Wang wherein after being trained, the machine learning model is shared with at least one other edge node having a threshold similarity to the edge node. Together, Satou and Ghibril teach of optimizing performance of an edge node in a cloud environment by identifying spare resources available and allocating the spare resources for training the machine learning model. Similarly, Wang teaches of a cloud computing network with edge nodes. A node generates a specification of a machine learning model and trains the model by sending the model to other edge nodes. Participating edge nodes are those that have the ability to match the specification of the machine learning model, which is having a threshold of similarity. This allows the initial edge node to establish new parameters for the machine learning model by aggregating updates from participating edge nodes, as discussed in Wang (Abstract; Col. 1, lines 47-67).
26. Regarding claim 14, it is rejected under the same reasoning as claim 9 above. Therefore, it is rejected under the same rationale.
27. Regarding claim 19, it is rejected under the same reasoning as claim 9 above. Therefore, it is rejected under the same rationale.
28. Claims 10, 15, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Satou US 20200101603 A1 and Ghibril et al. US 20190164087 A1, as applied in claim 1, in further view of Malladi et al. US 10007513 B2.
29. With regard to claim 10, Satou, Ghibril, and Wang teach the method of claim 9 but fail to explicitly teach wherein the threshold similarity is based on at least a proximity of the edge node and the at least one other edge node.
However, in analogous art, Malladi teaches:
wherein the threshold similarity is based on at least a proximity of the edge node and the at least one other edge node (Col. 2, lines 47-56, Edge intelligence platform is a software-based solution based on fog computing concepts which extends data processing and analytics closer to the edge where the IIoT devices reside. Maintaining close proximity to the edge devices rather than sending all data to a distant centralized cloud, minimizes latency allowing for maximum performance, faster response times, and more effective maintenance and operational strategies. It also significantly reduces overall bandwidth requirements and the cost of managing widely distributed networks.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Satou, Ghibril, and Wang with the teachings of Malladi wherein the threshold similarity is based on at least a proximity of the edge node and the at least one other edge node. Together, Satou and Ghibril teach of optimizing performance of an edge node in a cloud environment by identifying spare resources available and allocating the spare resources for training the machine learning model. Similarly, Wang teaches of an edge node that generates a specification of a machine learning model and trains the model by sending the model to other edge nodes that have similar characteristics to establish new parameters for the machine learning model by aggregating updates from participating edge nodes. Similarly, Malladi teaches a fog computing concept that extends data closer to the edge where the IIoT devices reside. By maintaining close proximity to the edge devices rather than sending all data to a distant centralized cloud, minimizes latency allowing for maximum performance, faster response times, and more effective maintenance and operational strategies. It also significantly reduces overall bandwidth requirements and the cost of managing widely distributed networks, as discussed in Malladi (Col. 2, lines 47-56).
30. Regarding claim 15, it is rejected under the same reasoning as claim 10 above. Therefore, it is rejected under the same rationale.
31. Regarding claim 20, it is rejected under the same reasoning as claim 10 above. Therefore, it is rejected under the same rationale.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to AN-AN N NGUYEN whose telephone number is (571)272-6147. The examiner can normally be reached Monday-Friday 8:00-5:00 ET.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, AIMEE LI can be reached at (571) 272-4169. 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.
/AN-AN NGOC NGUYEN/Examiner, Art Unit 2195
/Aimee Li/Supervisory Patent Examiner, Art Unit 2195
Prosecution Insights