Online in-vehicle learning for machine learning models
A system receives identification of a machine learning model, stored by a vehicle, that is ready for training. The system determines, based on a configuration file associated with the model, whether batch-based or live training is to be used to train the model and collects data for training, defined by a configuration file associated with the model. The system calls a learning as a service vehicle process to load a kernel and configures the kernel based on configuration data defined in the configuration file. The system receives notification from the learning as a service process that the training is complete. Additionally, the system validates a model, responsive to the notification including indication that training was successful, using a vehicle model validation process to test the model with live data before deployment by background execution of the model and saves a copy of the model for deployment responsive to successful validation.
1 . A system comprising:
one or more processors of a vehicle configured to:
receive identification of a machine learning model, stored by the vehicle, that is ready for training;
determine, based on a configuration file associated with the model, whether batch-based or live training is to be used to train the model;
collect data for training, defined by the configuration file associated with the model, the collection of data including subscription to one or more data topics created by a vehicle data gathering process responsive to a request defined by the configuration file as to what data is to be gathered for live training;
call a learning as a service vehicle process to load a kernel and configure the kernel based on configuration data defined in the configuration file;
receive notification from the learning as a service vehicle process that the training is complete;
validate the model, responsive to the notification including indication that training was successful, using a vehicle model validation process to test the model with live data before deployment by background execution of the model; and
save a copy of the model for deployment responsive to validation of the model being successful.
2 . The system of claim 1 , wherein the configuration file is saved in vehicle memory.
3 . The system of claim 1 , wherein the collection of data includes accessing historical data saved onboard the vehicle and associated with the model as data previously used by the model when the model was executing.
4 . The system of claim 1 , wherein the configuration file defines that type of training is to be used to train the model.
5 . The system of claim 1 , wherein the training being complete includes premature termination of training based on results produced by the trained model being outside accuracy parameters defined by the configuration file.
6 . The system of claim 1 , wherein the training being complete includes premature termination of training based on results produced by the trained model being outside loss thresholds being defined by the configuration file.
7 . A method comprising:
receiving identification of a machine learning model, stored by a vehicle, that is ready for training;
determining, based on a configuration file associated with the model, whether batch-based or live training is to be used to train the model;
collecting data for training, defined by the configuration file associated with the model, the collection of data including subscription to one or more data topics created by a vehicle data gathering process responsive to a request defined by the configuration file as to what data is to be gathered for live training;
calling a learning as a service vehicle process to load a kernel and configuring the kernel based on configuration data defined in the configuration file;
receiving notification from the learning as a service vehicle process that the training is complete;
validating the model, responsive to the notification including indication that training was successful, using a vehicle model validation process to test the model with live data before deployment by background execution of the model; and
saving a copy of the model for deployment responsive to validation of the model being successful.
8 . The method of claim 7 , wherein the configuration file is saved in vehicle memory.
9 . The method of claim 7 , wherein the collection of data includes accessing historical data saved onboard the vehicle and associated with the model as data previously used by the model when the model was executing.
10 . The method of claim 7 , wherein the configuration file defines what type of training is to be used to train the model.
11 . The method of claim 7 , wherein the training being complete includes premature termination of training based on results produced by the trained model being outside accuracy parameters defined by the configuration file.
12 . The method of claim 7 , wherein the training being complete includes premature termination of training based on results produced by the trained model being outside loss thresholds being defined by the configuration file.
13 . A non-transitory computer-readable storage medium storing instructions that, when executed by one or more processors, cause the one or more processors to perform a method comprising:
receiving identification of a machine learning model, stored by a vehicle, that is ready for training;
determining, based on a configuration file associated with the model, whether batch-based or live training is to be used to train the model;
collecting data for training, defined by the configuration file associated with the model, the collection of data including subscription to one or more data topics created by a vehicle data gathering process responsive to a request defined by the configuration file as to what data is to be gathered for live training;
calling a learning as a service vehicle process to load a kernel and configuring the kernel based on configuration data defined in the configuration file;
receiving notification from the learning as a service vehicle process that the training is complete;
validating the model, responsive to the notification including indication that training was successful, using a vehicle model validation process to test the model with live data before deployment by background execution of the model; and
saving a copy of the model for deployment responsive to validation of the model being successful.
14 . The storage medium of claim 13 , wherein the configuration file is saved in vehicle memory.
15 . The storage medium of claim 13 , wherein the configuration file defines what type of training is to be used to train the model.
16 . The storage medium of claim 13 , wherein the training being complete includes at least one of premature termination of training based on results produced by the trained model being outside accuracy parameters defined by the configuration file or premature termination of training based on results produced by the trained model being outside loss thresholds being defined by the configuration file.
17 . The storage medium of claim 13 , wherein the collection of data includes accessing historical data saved onboard the vehicle and associated with the model as data previously used by the model when the model was executing.