Random-Forest based ML model that offers faster training and inference times with smaller memory footprint.


RADE is based on two main observations:
1) A small (coarse-grained) RF is sufficient to classify the majority of the classification queries correctly.
2) In these fewer "harder" cases where our coarse-grained RF exhibits insufficient confidence in its result, we can obtain good results by turning over the classification decision to one of a collection of "expert" (fine-grained) RFs explicitly trained for that particular situation. Using these observations, we present, design and evaluate RADE - an RF-based classifier, which can support lighter and faster ML model migration and scaling-out in the clouds, as well as be employed for resource-constrained edge devices.



  • Active Research Areas