Elastic Regression-Tree Learning in a Heterogeneous Computing Environment

Abstract

Embedded intelligence demands sensor functions with minimal operator interference. While computing resources in close proximity to a device are sometimes limited, more capable computing units suffer from latency in long-range data transmission. Here, we demonstrate a workflow on optimizing sensor performance for individual samples, in which we utilize an elastic regression-tree method to distribute computing tasks between an embedded chipset and a full-fledged workstation. A reconciliation of computing resources of different nature enables an efficient data acquisition in a completely automated manner, crucial to the deployment of edge intelligence in a broad context.