Pipelined Training with Stale Weights in Deep Convolutional Neural Networks

Table 2

Inference accuracy for simulated pipelined training.

CNN	Nonpipelined (%)	4-Stage (%)	6-Stage (%)	8-Stage (%)	10-Stage (%)
LeNet-5	99.00	98.64	98.62	98.61	98.47
AlexNet	82.51	78.47	78.32	78.47	—
VGG-16	91.36	90.53	88.96	83.73	79.85
ResNet-20	91.50	90.05	88.00	83.01	—