Systems and methods for supporting always-on applications featuring artificial intelligence models by populating parallel data domains
Systems and methods for a novel architecture to support always-on applications and/or models suffering drift in their results. The system may comprise one or more servers that are configured to track the historical behavior of incoming request data for a model and/or redirect the request as needed using parallel data domains. The one or more servers may maintain and update a catalog of potential data domains that partitions the historically received data. One partition may comprise data output from a current model. Another partition may comprise detected outliers in the data. In the case of drift, outliers, and/or anomalies in the incoming data, the system may return an error signal that causes data to be duplicated into a new data domain.
1. A system for supporting always-on applications featuring artificial intelligence models by populating parallel data domains, the system comprising:
one or more processors; and
one or more non-transitory, computer-readable media comprising instructions recorded thereon that when executed on the one or more processors causes operations comprising:
receiving, from a first data stream, a first data input;
processing the first data input in a first model to generate a first data output;
determining, based on statistical analysis, whether the first data output corresponds to a first data domain or a second data domain;
in response to determining that the first data output corresponds to the first data domain, generating a second data output by duplicating the first data output;
adding the second data output to the first data domain;
determining, based on adding the second data output, a first size of the first data domain;
comparing the first size to a first threshold size to determine whether the first size corresponds to the first threshold size;
in response to determining that the first size corresponds to the first threshold size, retrieving a second model, wherein the second model is trained on a dataset from the first data domain; and
redirecting data from the first data stream to the second model.
2. A method for supporting always-on applications featuring artificial intelligence models by populating parallel data domains, the method comprising:
receiving a first data output from a first model;
determining, based on statistical analysis, whether the first data output corresponds to a first data domain or a second data domain;
in response to determining that the first data output corresponds to the first data domain, generating a second data output by duplicating the first data output;
adding the second data output to the first data domain;
determining, based on adding the second data output, a first size of the first data domain;
comparing the first size to a first threshold size to determine whether the first size corresponds to the first threshold size; and
in response to determining that the first size corresponds to the first threshold size, determining to deploy a second model by redirecting data from a first data stream to the second model.
3. The method of claim 2 , wherein determining whether the first data output corresponds to the first data domain or the second data domain further comprises:
determining a mean for a dataset output by the first model;
determining a number of standard deviations of the first data output from the mean; and
comparing the number of standard deviations to a threshold number of standard deviation to determine whether the first data output corresponds to the first data domain.
4. The method of claim 2 , wherein determining whether the first data output corresponds to the first data domain or the second data domain further comprises:
determining a median for a dataset output by the first model;
determining a median absolute deviation of the first data output; and
comparing the median absolute deviation to a threshold median absolute deviation to determine whether the first data output corresponds to the first data domain.
5. The method of claim 2 , wherein determining whether the first data output corresponds to the first data domain or the second data domain further comprises:
determining a distance of the first data output from a nearest neighboring data in a dataset output by the first model; and
comparing the distance to a threshold distance to determine whether the first data output corresponds to the first data domain.
6. The method of claim 2 , wherein determining whether the first data output corresponds to the first data domain or the second data domain further comprises:
determining a region for the first data output in a dataset output by the first model;
determining a data point density of the region; and
comparing the data point density to a threshold data point density to determine whether the first data output corresponds to the first data domain.
7. The method of claim 2 , wherein generating the second data output by duplicating the first data output further comprises:
receiving a write operation from a first server to generate the second data output; and
logging, by the first server, the write operation.
8. The method of claim 2 , wherein adding the second data output to the first data domain further comprises:
determining a first table in the first data domain; and
storing the second data output in the first table.
9. The method of claim 2 , wherein adding the second data output to the first data domain further comprises:
determining a first pointer location for the first data domain; and
storing the second data output based on the first pointer location.
10. The method of claim 2 , wherein determining the first size of the first data domain further comprises:
determining floating-point data corresponding to the second data output; and
determining the first size based on the floating-point data.
11. The method of claim 2 , wherein comparing the first size to the first threshold size to determine whether the first size corresponds to the first threshold size further comprises:
determining a model type of the first model; and
determining the first threshold size based on the model type.
12. The method of claim 2 , wherein comparing the first size to the first threshold size to determine whether the first size corresponds to the first threshold size further comprises:
determining an application using outputs from the first model; and
determining the first threshold size based on the application.
13. The method of claim 2 , wherein comparing the first size to the first threshold size to determine whether the first size corresponds to the first threshold size further comprises:
determining a time period for use of the first model; and
determining the first threshold size based on the time period.
14. The method of claim 2 , wherein determining to deploy the second model further comprises:
retrieving a dataset from the first data domain; and
retraining the first model using the dataset.
15. The method of claim 2 , wherein determining to deploy the second model further comprises:
retrieving the second model; and
validating the second model.
16. The method of claim 2 , wherein determining to deploy the second model further comprises:
determining that the first data stream provided a first data input, wherein the first data input is processed by the first model to generate the first data output.
17. The method of claim 2 , wherein generating the second data output by duplicating the first data output further comprises:
receiving a read operation from a first server to generate the second data output; and
logging, by the first server, the read operation.
18. The method of claim 2 , wherein determining whether the first data output corresponds to the first data domain or the second data domain further comprises:
determining a value based on a dataset output by the first model; and
determining whether the value corresponds to the first data domain.
19. The method of claim 2 , wherein determining whether the first data output corresponds to the first data domain or the second data domain further comprises:
retrieving a first value for the first data domain and a second value for the second data domain; and
comparing the first value and the second value to the first data output.
20. The method of claim 2 , wherein determining the first size of the first data domain further comprises:
determining data corresponding to the second data output; and
determining the first size based on the data.