Research Article
Collaborative Intelligence: Accelerating Deep Neural Network Inference via Device-Edge Synergy
Algorithm 1
Automated model pruning and partition algorithm.
| (1) Input: | | (2) N: number of layers in the DNN | | (3) {Lt | t = 1,…, N}: layers in the DNN | | (4) Agent (Lt): reinforcement learning agent predicting the output parameters and latency of executing Lt | | (5) B: current wireless network uplink bandwidth | | (6) E: current edge server load | | (7) H: hardware accelerator’s feedback | | (8) procedure THE FIRST STEP | | (9) for eacht in 1… Ndo | | (10) | | (11) | | (12) | | (13) Tt | | (14) end for | | (15) return Partition point Actionp | | (16) procedure THE SECOND STEP | | (17) for eacht in 1… Ndo if OptTarget min(Accuracy) then | | (18) return Pruning ratio Actiont | | (19) end if | | (20) end for | | (21) return NULL |
|