Back to results
US 2019/0171428 A1Published application
Automated model management methods
AssigneeBanjo, Inc.
InventorsDamien Patton, Tilmann Bruckhaus
Life of this patent
Estimating the term…
Abstract
The method for model management includes, within a testing platform: building candidate model(s), validating the candidate model(s), and selectively deploying the candidate model(s) into a production environment. The method can optionally include, within a production environment: detecting a class instance based on the raw data using deployed model(s) within the production environment, and transmitting a notification to an endpoint associated with the detected class.
Claims22
A computer-implemented method comprising:
- labeling raw data from a set of data sources;
- detecting a concept drift rate for a geographic region from the raw data;
- detecting an occurrence of an event class at a deployed event detection model based on features of the raw data;
- identifying labeled raw data associated with the event class;
- segmenting the identified labeled raw data into a training data set and a testing data set;
- training a seed model with the training data set generating an event class candidate model specific to the geographic region in response to detecting the concept drift rate by determining candidate model evaluation metric values using the testing data set;
- in response to the candidate model evaluation metric values satisfying a set of deployment conditions, deploying the event class candidate model;
- and
- detecting a subsequent occurrence of the event class based on subsequent raw data received from the set of data sources, using the event class candidate model.
6 more elements
(canceled)
(canceled)
A computer-implemented model, comprising:
- detecting a concept drift rate from a geographic region from raw data;
- identifying labeled data including labels pre-associated with an event class;
- segmenting the labeled data into a training data set and testing data set;
- training a seed model specific to the geographic region and in response to detecting the concept drift rate with the training data set to generate a candidate model for the event class;
- determining evaluation metrics for the candidate model using the testing data set;
- detecting an occurrence of the event class based on raw data received from a set of data sources using the candidate model;
- identifying other raw data geotagged with geographic locations within the geographic region;
- determining feature values from the other raw data;
- and
- detecting an occurrence of an event based on the feature values.
7 more elements
A computer system comprising:
- a processor;
- system memory coupled to the processor and storing instructions configured to cause the processor to:
- label raw data from a set of data sources;
- detect a concept drift rate for a geographic region from the raw data;
- detect an occurrence of an event class at a deployed event detection model based on features of the raw data;
- identify labeled raw data associated with the event class;
- segment the identified labeled raw data into a training data set and a testing data set;
- train a seed model with the training data set generating an event class candidate model specific to the geographic region in response to detecting the concept drift rate;
- determine candidate model evaluation metric values using the testing data set;
- in response to the candidate model evaluation metric values satisfying a set of deployment conditions, deploy the event class candidate model;
- and
- detect a subsequent occurrence of the event class based on subsequent raw data received from the set of data sources, using the event class candidate model.
9 more elements
A computer system comprising:
- a processor;
- system memory coupled to the processor and storing instructions configured to cause the processor to:
- identify labeled data including labels pre-associated with an event class;
- segment the labeled data into a training data set and a testing data set that is disjoint from the training data set;
- train a seed model with the training data set to generate a candidate model for the event class;
- determine evaluation metrics for the candidate model using the testing data set;
- detect an occurrence of the event within an event detection system class based on raw data received from a set of data sources using the candidate model;
- transmit a notification to an endpoint in response to detecting the occurrence of the event;
- identify other raw data geotagged with geographic locations within a geographic region;
- determine feature values from the other raw data;
- and
- detect an occurrence of an event based on the feature values.
9 more elements
Dependencies are inferred from the claim text.