Authors: Yiping Kang, Johann Hauswald, Cao Gao, Austin Rovinski, Trevor Mudge, Jason Mars, Lingjia Tang (UMich).
Understanding the paper
TL;DRs
This paper presents Neurosurgeon, a lightweight scheduler to automatically partition DNN computation between mobile devices and data centers at the granularity of neural network layers.
It doesn't require per-application profiling.
Overview
System Overview.
Dynamic DNN Partitioning
Analysis of the target DNN (use the prediction models to estimate)
Partition point selection (for best end-to-end latency or best mobile energy consumption)
Related Work
Previous research efforts focus on offloading computation from the mobile to the cloud.
