This invention aims to allow mobile applications based on Machine Learning to run in
different devices with different configurations, by optimizing the tradeoff between accuracy,
response time and storage use specifically to each device. It consists in allowing the developer to
train multiple models to solve the same task. A subset of trained models, with acceptable
accuracy, will then be made available for the application. Whenever the application is installed
on the device, it will pre-load each of the available models, initialize it with random parameters,
and run some inference steps on the device, in order to measure the computation time. With this
information, together with the full size and accuracy of the model , the application can select the
most accurate model with acceptable response time (according to the application), or even allow
the user to select a model.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Share Alike 4.0 License.
INC, HP, "ADAPTIVE DEPLOYMENT OF MACHINE LEARNING MODELS IN MOBILE DEVICES", Technical Disclosure Commons, (May 31, 2019)