Centralized management of distributed data sources
Aspects of the disclosure are directed to a central management plane (CMP) of one or more processors for regulating streams of data from each of a number of network nodes of a distributed network. The one or more processors can train and deploy machine learning models across the network nodes, and the CMP can generate policies for each network node. The generated policies specify how a network node is to transmit data to the platform for further training or retraining of the deployed machine learning models. The CMP generates the policies using metric data characterizing each network node and respective streams of input data, and are generated based on a number of objectives, including model output quality of the deployed models, and operational cost to transmit and process streams of data across the distributed network.
1. A system comprising:
a plurality of network nodes executing a machine learning model on a stream of input data; and
one or more processors configured to:
receive metric data at least partially characterizing the plurality of network nodes;
generate a policy for regulating the stream of input data according to one or more objectives based on the metric data, the one or more objectives comprising reducing latency of transmitting the stream of input data and increasing accuracy of the machine learning model; and
transmit the policy to the plurality of network nodes.
2. The system of claim 1 , wherein the policy comprises one or more rules for adjusting one or more of: a rate of the stream of input data transmitted by the plurality of network nodes, volume of the stream of input data transmitted by the plurality of network nodes, a time at which the stream of input data is transmitted by the plurality of network nodes, or types of data in the stream of input data transmitted by the plurality of network nodes.
3. The system of claim 1 , wherein the one or more processors are further configured to receive a regulated stream of input data after the policy is applied to the stream of input data.
4. The system of claim 1 , wherein, in generating the policy, the one or more processors are configured to process the metric data through a machine learning model trained to generate policies using labeled metric data.
5. The system of claim 4 , wherein the labeled metric data comprises features corresponding to the one or more objectives.
6. The system of claim 1 , wherein the one or more objectives further comprises one or more objectives for training the machine learning model.
7. The system of claim 1 , wherein the one or more processors are further configured to:
receive additional metric data at least partially characterizing performance of the plurality of network nodes; and
generate and send an additional policy for regulating the stream of input data based on the additional metric data.
8. The system of claim 1 , wherein the one or more processors are further configured to adjust a transmission of data in the stream of input data in response to the policy.
9. The system of claim 1 , wherein the metric data at least partially characterizing the plurality of network nodes comprises one or more of: physical characteristics of the plurality of network nodes or characteristics of the stream of input data, or characteristics of a data source for the stream of input data.
10. A computer-implemented method, comprising:
receiving, by one or more processors, metric data at least partially characterizing a plurality of network nodes executing a machine learning model on a stream of input data;
generating, by the one or more processors, a policy for regulating the stream of input data according to one or more objectives based on the metric data, the one or more objectives comprising reducing latency of transmitting the stream of input data and increasing accuracy of the machine learning model; and
transmitting, by the one or more processors, the policy to the plurality of network nodes.
11. The method of claim 10 , wherein the policy comprises one or more rules for adjusting one or more of: a rate of the stream of input data transmitted by the plurality of network nodes, volume of the stream of input data transmitted by the plurality of network nodes, a time at which the stream of input data is transmitted by the plurality of network nodes, or types of data in the stream of input data transmitted by the plurality of network nodes.
12. The method of claim 10 , further comprising receiving, by the one or more processors, a regulated stream of input data after the policy is applied to the stream of input data.
13. The method of claim 10 , wherein generating the policy comprises processing, by the one or more processors, the metric data through a machine learning model trained to generate policies using labeled metric data.
14. The method of claim 13 , wherein the labeled metric data comprises features corresponding to the one or more objectives.
15. The method of claim 10 , wherein the one or more objectives further comprises one or more objectives for training the machine learning model.
16. The method of claim 10 , further comprising:
receiving, by the one or more processors, additional metric data at least partially characterizing performance of the plurality of network nodes; and
generate and send an additional policy for regulating the stream of input data based on the additional metric data.
17. The method of claim 10 , further comprising adjusting, by the one or more processors, a transmission of data in the stream of input data in response to the policy.
18. The method of claim 10 , wherein the metric data at least partially characterizing the plurality of network nodes comprises one or more of: physical characteristics of the plurality of network nodes or characteristics of the stream of input data, or characteristics of a data source for the stream of input data.
19. One or more non-transitory computer-readable storage media encoding instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
receiving metric data at least partially characterizing a plurality of network nodes executing a machine learning model on a stream of input data;
generating a policy for regulating the stream of input data according to one or more objectives based on the metric data, the one or more objectives comprising reducing latency of transmitting the stream of input data and increasing accuracy of the machine learning model; and
transmitting the policy to the plurality of network nodes.
20. The one or more non-transitory computer-readable media of claim 19 , wherein the operations comprise receiving a regulated stream of input data after the policy is applied to the stream of input data.