Patent
US 10,742,516KNOWLEDGE AGGREGATION FOR GAN-BASED ANOMALY DETECTORS
Volodymyr Iashyn
US
Copyright © 2026 Matter42. All rights reserved.
Patent
Atlas literature
Patent
US 10,742,516Patent drawings and their descriptions. Click a drawing to enlarge it.
FIG. 1 B illustrates an example fog computing architecture; [0008]
FIG. 2A illustrates a diagram of an example network environment, such as a data center; [0009]
FIG. 3 illustrates an example environment for distributing machine learning to edges of a network for monitoring network devices in the network; [0011]
FIG. 4 illustrates a flow for training an example GAN model; [0012]
FIG. 5 shows a graph of values of model parameters for determining when to push a GAN model from a network edge device to an applicable destination according …
FIG. 6 illustrates a flow by which generators for different GAN models can be combined to generate a unified GAN model; [0014]
FIG. 7B illustrates another data graph of a generator combined to form virtual telemetry data for training a unified GAN model; [0016]
FIG. 8 illustrates a flowchart for an example method of distributing machine learning to a network edge; [0017]
FIG. 9 illustrates an example computing system; and [0018]
FIG. 10 illustrates an example network device.
Claims define the patent's legal scope. Independent claims stand alone; dependent claims (nested) narrow them. Click a claim to expand its dependents.
A method comprising: deploying a first generative adversarial network (GAN) model to a first network edge device and a second GAN model to a second network edge device, wherein a generator of the first GAN model is trained using real telemetry data of a first computing node and a generator of the second GAN model is trained using real telemetry data of a second computing node; receiving, from the first network edge device, the generator of the first GAN model; receiving, from the second network edge device, the generator of the second GAN model; training a unified generator of a unified GAN model using the generator of the first GAN model and the generator of the second GAN model; and deploying the unified GAN model to a third computing node for monitoring operation of the third computing node.
The method of claim 1, detecting anomalies in the operation of the third computing node using the unified GAN model.
The method of claim 1, wherein the first computing node is the first network edge device and the generator of the first GAN model is trained at the first computing node.
The method of claim 1, wherein the second computing node is the second network edge device and the generator of the second GAN model is trained at the second computing node.
The method of claim 1, wherein the generator of the first GAN model and the generator of the second GAN model are received at one or more remote computing devices from the first network edge device and the second network edge device, and the one or more remote computing devices are configured to train the unified generator of the unified GAN model remote from the first network edge device and the second network edge device.
The method of claim 1, wherein the generator of the first GAN model and the generator of the second GAN model form virtual telemetry data of the first computing node and the second computing node and the virtual telemetry data is provided to a discriminator of the unified GAN model to train the unified generator of the unified GAN model.
The method of claim 1, wherein the first computing node, the second computing node and the third computing node are a same computing node type.
The method of claim 1, wherein the generator of the first GAN model is pushed from the first network edge device in response to a first triggering event and the generator of the second GAN model is pushed from the second network edge device in response to a second triggering event.
The method of claim 1, wherein the first GAN model is further trained using additional real telemetry data of the first computing node to generate an updated g enerator of the first GAN model and the second GAN model is further trained using additional real telemetry data of the second computing node to generate an updated g enerator of the second G AN model, the method further comprising: receiving. from the first network edge device, the updated generator of the first GAN model; receiving, from the second network edge device, the updated generator of the second GAN model; and updating the unified generator of the unified GAN model using the updated generator of the first GAN model and the updated generator of the second GAN model to create an updated unified generator of an updated unified GAN model.
A system comprising: one or more processors; and at least one computer-readable storage medium having stored therein instructions which, when executed by the one or more processors, cause the one or more processors to perform operations comprising: deploying a first generative adversarial network (GAN) model to a first network edge device and a second GAN model to a second network edge device, wherein a generator of the first GAN model is trained using real telemetry data of a first computing node and a generator of the second GAN model is trained using real telemetry data of a second computing node; receiving, from the first network edge device, the generator of the first GAN model; receiving, from the second network edge device, the generator of the second GAN model; and training a unified generator of a unified GAN model using the generator of the first GAN model and the generator of the second GAN model.
The system of claim 15, wherein the generator of the first GAN model and the generator of the second GAN model are received at one or more remote computing devices from the first network edge device and the second network edge device, and the one or more remote computing devices are configured to train the unified generator of the unified GAN model remote from the first network edge device and the second network edge device.
The system of claim 15, wherein the generator of the first GAN model is pushed from the first network edge device in response to a first triggering event and the generator of the second GAN model is pushed from the second network edge device in response to a second triggering event and the first triggering event corresponds to values of one or more model parameters of the first GAN model changing by a threshold amount during training of the first GAN model and wherein the second triggering event corresponds to values of one or more model parameters of the second GAN model changing by another threshold amount during training of the second GAN model.
A non-transitory computer-readable storage medium having stored therein instructions which, when executed by a processor, cause the processor to perform operations comprising: deploying a first generative adversarial network (GAN) model to a first network edge device and a second GAN model to a second network edge device, trained using real telemetry data of a first Internet second GAN model is trained using real telemetry first network edge device, the generator of the first network edge device, the generator of the second unified GAN model using the generator of the first GAN model; and deploying the unified GAN model during operation of the third I oT device. wherein a generator of the first GAN model is of Things (I oT) device and a generator of the data of a second I oT device; receiving, from the GAN model; receiving, from the second GAN model; training a unified generator of a GAN model and the generator of the second to a third T oT device for detecting anomalies
Measurements and analyses referenced in the patent, with their drawing references.
FIG. 5 shows a graph of values of model parameters for determining when to push a GAN model from a network edge device to an applicable destination according …
Related documents with shared materials, methods, properties, or citations.
SYSTEMS AND METHODS FOR A RADIO FREQUENCY TRANSMITTER WITH IMPROVED LINEARITY AND POWER OUT UTILIZING PRE-DISTORTION AND A GAN (GALLIUM NITRIDE) POWER AMPLIFIER DEVICE
ANSWER CLASSIFIER AND REPRESENTATION GENERATOR FOR QUESTION-ANSWERING SYSTEM USING GAN, AND COMPUTER PROGRAM FOR TRAINING THE REPRESENTATION GENERATOR
Patent
Atlas literature
Patent
US 10,742,516Patent drawings and their descriptions. Click a drawing to enlarge it.
FIG. 1 B illustrates an example fog computing architecture; [0008]
FIG. 2A illustrates a diagram of an example network environment, such as a data center; [0009]
FIG. 3 illustrates an example environment for distributing machine learning to edges of a network for monitoring network devices in the network; [0011]
FIG. 4 illustrates a flow for training an example GAN model; [0012]
FIG. 5 shows a graph of values of model parameters for determining when to push a GAN model from a network edge device to an applicable destination according …
FIG. 6 illustrates a flow by which generators for different GAN models can be combined to generate a unified GAN model; [0014]
FIG. 7B illustrates another data graph of a generator combined to form virtual telemetry data for training a unified GAN model; [0016]
FIG. 8 illustrates a flowchart for an example method of distributing machine learning to a network edge; [0017]
FIG. 9 illustrates an example computing system; and [0018]
FIG. 10 illustrates an example network device.
Claims define the patent's legal scope. Independent claims stand alone; dependent claims (nested) narrow them. Click a claim to expand its dependents.
A method comprising: deploying a first generative adversarial network (GAN) model to a first network edge device and a second GAN model to a second network edge device, wherein a generator of the first GAN model is trained using real telemetry data of a first computing node and a generator of the second GAN model is trained using real telemetry data of a second computing node; receiving, from the first network edge device, the generator of the first GAN model; receiving, from the second network edge device, the generator of the second GAN model; training a unified generator of a unified GAN model using the generator of the first GAN model and the generator of the second GAN model; and deploying the unified GAN model to a third computing node for monitoring operation of the third computing node.
The method of claim 1, detecting anomalies in the operation of the third computing node using the unified GAN model.
The method of claim 1, wherein the first computing node is the first network edge device and the generator of the first GAN model is trained at the first computing node.
The method of claim 1, wherein the second computing node is the second network edge device and the generator of the second GAN model is trained at the second computing node.
The method of claim 1, wherein the generator of the first GAN model and the generator of the second GAN model are received at one or more remote computing devices from the first network edge device and the second network edge device, and the one or more remote computing devices are configured to train the unified generator of the unified GAN model remote from the first network edge device and the second network edge device.
The method of claim 1, wherein the generator of the first GAN model and the generator of the second GAN model form virtual telemetry data of the first computing node and the second computing node and the virtual telemetry data is provided to a discriminator of the unified GAN model to train the unified generator of the unified GAN model.
The method of claim 1, wherein the first computing node, the second computing node and the third computing node are a same computing node type.
The method of claim 1, wherein the generator of the first GAN model is pushed from the first network edge device in response to a first triggering event and the generator of the second GAN model is pushed from the second network edge device in response to a second triggering event.
The method of claim 1, wherein the first GAN model is further trained using additional real telemetry data of the first computing node to generate an updated g enerator of the first GAN model and the second GAN model is further trained using additional real telemetry data of the second computing node to generate an updated g enerator of the second G AN model, the method further comprising: receiving. from the first network edge device, the updated generator of the first GAN model; receiving, from the second network edge device, the updated generator of the second GAN model; and updating the unified generator of the unified GAN model using the updated generator of the first GAN model and the updated generator of the second GAN model to create an updated unified generator of an updated unified GAN model.
A system comprising: one or more processors; and at least one computer-readable storage medium having stored therein instructions which, when executed by the one or more processors, cause the one or more processors to perform operations comprising: deploying a first generative adversarial network (GAN) model to a first network edge device and a second GAN model to a second network edge device, wherein a generator of the first GAN model is trained using real telemetry data of a first computing node and a generator of the second GAN model is trained using real telemetry data of a second computing node; receiving, from the first network edge device, the generator of the first GAN model; receiving, from the second network edge device, the generator of the second GAN model; and training a unified generator of a unified GAN model using the generator of the first GAN model and the generator of the second GAN model.
The system of claim 15, wherein the generator of the first GAN model and the generator of the second GAN model are received at one or more remote computing devices from the first network edge device and the second network edge device, and the one or more remote computing devices are configured to train the unified generator of the unified GAN model remote from the first network edge device and the second network edge device.
The system of claim 15, wherein the generator of the first GAN model is pushed from the first network edge device in response to a first triggering event and the generator of the second GAN model is pushed from the second network edge device in response to a second triggering event and the first triggering event corresponds to values of one or more model parameters of the first GAN model changing by a threshold amount during training of the first GAN model and wherein the second triggering event corresponds to values of one or more model parameters of the second GAN model changing by another threshold amount during training of the second GAN model.
A non-transitory computer-readable storage medium having stored therein instructions which, when executed by a processor, cause the processor to perform operations comprising: deploying a first generative adversarial network (GAN) model to a first network edge device and a second GAN model to a second network edge device, trained using real telemetry data of a first Internet second GAN model is trained using real telemetry first network edge device, the generator of the first network edge device, the generator of the second unified GAN model using the generator of the first GAN model; and deploying the unified GAN model during operation of the third I oT device. wherein a generator of the first GAN model is of Things (I oT) device and a generator of the data of a second I oT device; receiving, from the GAN model; receiving, from the second GAN model; training a unified generator of a GAN model and the generator of the second to a third T oT device for detecting anomalies
Measurements and analyses referenced in the patent, with their drawing references.
FIG. 5 shows a graph of values of model parameters for determining when to push a GAN model from a network edge device to an applicable destination according …
Related documents with shared materials, methods, properties, or citations.
SYSTEMS AND METHODS FOR A RADIO FREQUENCY TRANSMITTER WITH IMPROVED LINEARITY AND POWER OUT UTILIZING PRE-DISTORTION AND A GAN (GALLIUM NITRIDE) POWER AMPLIFIER DEVICE
ANSWER CLASSIFIER AND REPRESENTATION GENERATOR FOR QUESTION-ANSWERING SYSTEM USING GAN, AND COMPUTER PROGRAM FOR TRAINING THE REPRESENTATION GENERATOR
Patent
Atlas literature
Patent
US 10,742,516Patent drawings and their descriptions. Click a drawing to enlarge it.
FIG. 1 B illustrates an example fog computing architecture; [0008]
FIG. 2A illustrates a diagram of an example network environment, such as a data center; [0009]
FIG. 3 illustrates an example environment for distributing machine learning to edges of a network for monitoring network devices in the network; [0011]
FIG. 4 illustrates a flow for training an example GAN model; [0012]
FIG. 5 shows a graph of values of model parameters for determining when to push a GAN model from a network edge device to an applicable destination according …
FIG. 6 illustrates a flow by which generators for different GAN models can be combined to generate a unified GAN model; [0014]
FIG. 7B illustrates another data graph of a generator combined to form virtual telemetry data for training a unified GAN model; [0016]
FIG. 8 illustrates a flowchart for an example method of distributing machine learning to a network edge; [0017]
FIG. 9 illustrates an example computing system; and [0018]
FIG. 10 illustrates an example network device.
Claims define the patent's legal scope. Independent claims stand alone; dependent claims (nested) narrow them. Click a claim to expand its dependents.
A method comprising: deploying a first generative adversarial network (GAN) model to a first network edge device and a second GAN model to a second network edge device, wherein a generator of the first GAN model is trained using real telemetry data of a first computing node and a generator of the second GAN model is trained using real telemetry data of a second computing node; receiving, from the first network edge device, the generator of the first GAN model; receiving, from the second network edge device, the generator of the second GAN model; training a unified generator of a unified GAN model using the generator of the first GAN model and the generator of the second GAN model; and deploying the unified GAN model to a third computing node for monitoring operation of the third computing node.
The method of claim 1, detecting anomalies in the operation of the third computing node using the unified GAN model.
The method of claim 1, wherein the first computing node is the first network edge device and the generator of the first GAN model is trained at the first computing node.
The method of claim 1, wherein the second computing node is the second network edge device and the generator of the second GAN model is trained at the second computing node.
The method of claim 1, wherein the generator of the first GAN model and the generator of the second GAN model are received at one or more remote computing devices from the first network edge device and the second network edge device, and the one or more remote computing devices are configured to train the unified generator of the unified GAN model remote from the first network edge device and the second network edge device.
The method of claim 1, wherein the generator of the first GAN model and the generator of the second GAN model form virtual telemetry data of the first computing node and the second computing node and the virtual telemetry data is provided to a discriminator of the unified GAN model to train the unified generator of the unified GAN model.
The method of claim 1, wherein the first computing node, the second computing node and the third computing node are a same computing node type.
The method of claim 1, wherein the generator of the first GAN model is pushed from the first network edge device in response to a first triggering event and the generator of the second GAN model is pushed from the second network edge device in response to a second triggering event.
The method of claim 1, wherein the first GAN model is further trained using additional real telemetry data of the first computing node to generate an updated g enerator of the first GAN model and the second GAN model is further trained using additional real telemetry data of the second computing node to generate an updated g enerator of the second G AN model, the method further comprising: receiving. from the first network edge device, the updated generator of the first GAN model; receiving, from the second network edge device, the updated generator of the second GAN model; and updating the unified generator of the unified GAN model using the updated generator of the first GAN model and the updated generator of the second GAN model to create an updated unified generator of an updated unified GAN model.
A system comprising: one or more processors; and at least one computer-readable storage medium having stored therein instructions which, when executed by the one or more processors, cause the one or more processors to perform operations comprising: deploying a first generative adversarial network (GAN) model to a first network edge device and a second GAN model to a second network edge device, wherein a generator of the first GAN model is trained using real telemetry data of a first computing node and a generator of the second GAN model is trained using real telemetry data of a second computing node; receiving, from the first network edge device, the generator of the first GAN model; receiving, from the second network edge device, the generator of the second GAN model; and training a unified generator of a unified GAN model using the generator of the first GAN model and the generator of the second GAN model.
The system of claim 15, wherein the generator of the first GAN model and the generator of the second GAN model are received at one or more remote computing devices from the first network edge device and the second network edge device, and the one or more remote computing devices are configured to train the unified generator of the unified GAN model remote from the first network edge device and the second network edge device.
The system of claim 15, wherein the generator of the first GAN model is pushed from the first network edge device in response to a first triggering event and the generator of the second GAN model is pushed from the second network edge device in response to a second triggering event and the first triggering event corresponds to values of one or more model parameters of the first GAN model changing by a threshold amount during training of the first GAN model and wherein the second triggering event corresponds to values of one or more model parameters of the second GAN model changing by another threshold amount during training of the second GAN model.
A non-transitory computer-readable storage medium having stored therein instructions which, when executed by a processor, cause the processor to perform operations comprising: deploying a first generative adversarial network (GAN) model to a first network edge device and a second GAN model to a second network edge device, trained using real telemetry data of a first Internet second GAN model is trained using real telemetry first network edge device, the generator of the first network edge device, the generator of the second unified GAN model using the generator of the first GAN model; and deploying the unified GAN model during operation of the third I oT device. wherein a generator of the first GAN model is of Things (I oT) device and a generator of the data of a second I oT device; receiving, from the GAN model; receiving, from the second GAN model; training a unified generator of a GAN model and the generator of the second to a third T oT device for detecting anomalies
Measurements and analyses referenced in the patent, with their drawing references.
FIG. 5 shows a graph of values of model parameters for determining when to push a GAN model from a network edge device to an applicable destination according …
Related documents with shared materials, methods, properties, or citations.
SYSTEMS AND METHODS FOR A RADIO FREQUENCY TRANSMITTER WITH IMPROVED LINEARITY AND POWER OUT UTILIZING PRE-DISTORTION AND A GAN (GALLIUM NITRIDE) POWER AMPLIFIER DEVICE
ANSWER CLASSIFIER AND REPRESENTATION GENERATOR FOR QUESTION-ANSWERING SYSTEM USING GAN, AND COMPUTER PROGRAM FOR TRAINING THE REPRESENTATION GENERATOR
Patent
Atlas literature
Patent
US 10,742,516Patent drawings and their descriptions. Click a drawing to enlarge it.
FIG. 1 B illustrates an example fog computing architecture; [0008]
FIG. 2A illustrates a diagram of an example network environment, such as a data center; [0009]
FIG. 3 illustrates an example environment for distributing machine learning to edges of a network for monitoring network devices in the network; [0011]
FIG. 4 illustrates a flow for training an example GAN model; [0012]
FIG. 5 shows a graph of values of model parameters for determining when to push a GAN model from a network edge device to an applicable destination according …
FIG. 6 illustrates a flow by which generators for different GAN models can be combined to generate a unified GAN model; [0014]
FIG. 7B illustrates another data graph of a generator combined to form virtual telemetry data for training a unified GAN model; [0016]
FIG. 8 illustrates a flowchart for an example method of distributing machine learning to a network edge; [0017]
FIG. 9 illustrates an example computing system; and [0018]
FIG. 10 illustrates an example network device.
Claims define the patent's legal scope. Independent claims stand alone; dependent claims (nested) narrow them. Click a claim to expand its dependents.
A method comprising: deploying a first generative adversarial network (GAN) model to a first network edge device and a second GAN model to a second network edge device, wherein a generator of the first GAN model is trained using real telemetry data of a first computing node and a generator of the second GAN model is trained using real telemetry data of a second computing node; receiving, from the first network edge device, the generator of the first GAN model; receiving, from the second network edge device, the generator of the second GAN model; training a unified generator of a unified GAN model using the generator of the first GAN model and the generator of the second GAN model; and deploying the unified GAN model to a third computing node for monitoring operation of the third computing node.
The method of claim 1, detecting anomalies in the operation of the third computing node using the unified GAN model.
The method of claim 1, wherein the first computing node is the first network edge device and the generator of the first GAN model is trained at the first computing node.
The method of claim 1, wherein the second computing node is the second network edge device and the generator of the second GAN model is trained at the second computing node.
The method of claim 1, wherein the generator of the first GAN model and the generator of the second GAN model are received at one or more remote computing devices from the first network edge device and the second network edge device, and the one or more remote computing devices are configured to train the unified generator of the unified GAN model remote from the first network edge device and the second network edge device.
The method of claim 1, wherein the generator of the first GAN model and the generator of the second GAN model form virtual telemetry data of the first computing node and the second computing node and the virtual telemetry data is provided to a discriminator of the unified GAN model to train the unified generator of the unified GAN model.
The method of claim 1, wherein the first computing node, the second computing node and the third computing node are a same computing node type.
The method of claim 1, wherein the generator of the first GAN model is pushed from the first network edge device in response to a first triggering event and the generator of the second GAN model is pushed from the second network edge device in response to a second triggering event.
The method of claim 1, wherein the first GAN model is further trained using additional real telemetry data of the first computing node to generate an updated g enerator of the first GAN model and the second GAN model is further trained using additional real telemetry data of the second computing node to generate an updated g enerator of the second G AN model, the method further comprising: receiving. from the first network edge device, the updated generator of the first GAN model; receiving, from the second network edge device, the updated generator of the second GAN model; and updating the unified generator of the unified GAN model using the updated generator of the first GAN model and the updated generator of the second GAN model to create an updated unified generator of an updated unified GAN model.
A system comprising: one or more processors; and at least one computer-readable storage medium having stored therein instructions which, when executed by the one or more processors, cause the one or more processors to perform operations comprising: deploying a first generative adversarial network (GAN) model to a first network edge device and a second GAN model to a second network edge device, wherein a generator of the first GAN model is trained using real telemetry data of a first computing node and a generator of the second GAN model is trained using real telemetry data of a second computing node; receiving, from the first network edge device, the generator of the first GAN model; receiving, from the second network edge device, the generator of the second GAN model; and training a unified generator of a unified GAN model using the generator of the first GAN model and the generator of the second GAN model.
The system of claim 15, wherein the generator of the first GAN model and the generator of the second GAN model are received at one or more remote computing devices from the first network edge device and the second network edge device, and the one or more remote computing devices are configured to train the unified generator of the unified GAN model remote from the first network edge device and the second network edge device.
The system of claim 15, wherein the generator of the first GAN model is pushed from the first network edge device in response to a first triggering event and the generator of the second GAN model is pushed from the second network edge device in response to a second triggering event and the first triggering event corresponds to values of one or more model parameters of the first GAN model changing by a threshold amount during training of the first GAN model and wherein the second triggering event corresponds to values of one or more model parameters of the second GAN model changing by another threshold amount during training of the second GAN model.
A non-transitory computer-readable storage medium having stored therein instructions which, when executed by a processor, cause the processor to perform operations comprising: deploying a first generative adversarial network (GAN) model to a first network edge device and a second GAN model to a second network edge device, trained using real telemetry data of a first Internet second GAN model is trained using real telemetry first network edge device, the generator of the first network edge device, the generator of the second unified GAN model using the generator of the first GAN model; and deploying the unified GAN model during operation of the third I oT device. wherein a generator of the first GAN model is of Things (I oT) device and a generator of the data of a second I oT device; receiving, from the GAN model; receiving, from the second GAN model; training a unified generator of a GAN model and the generator of the second to a third T oT device for detecting anomalies
Measurements and analyses referenced in the patent, with their drawing references.
FIG. 5 shows a graph of values of model parameters for determining when to push a GAN model from a network edge device to an applicable destination according …
Related documents with shared materials, methods, properties, or citations.
SYSTEMS AND METHODS FOR A RADIO FREQUENCY TRANSMITTER WITH IMPROVED LINEARITY AND POWER OUT UTILIZING PRE-DISTORTION AND A GAN (GALLIUM NITRIDE) POWER AMPLIFIER DEVICE
ANSWER CLASSIFIER AND REPRESENTATION GENERATOR FOR QUESTION-ANSWERING SYSTEM USING GAN, AND COMPUTER PROGRAM FOR TRAINING THE REPRESENTATION GENERATOR
FIG. 7B illustrates another data graph of a generator combined to form virtual telemetry data for training a unified GAN model; [0016]
FIG. 7B illustrates another data graph of a generator combined to form virtual telemetry data for training a unified GAN model; [0016]
FIG. 7B illustrates another data graph of a generator combined to form virtual telemetry data for training a unified GAN model; [0016]
FIG. 7B illustrates another data graph of a generator combined to form virtual telemetry data for training a unified GAN model; [0016]