Impact of Layers Selective Approximation on CNNs Reliability and Performance
Rubens Luiz Rech, Paolo Rech · 2020
In this paper, we evaluate the impact on reliability and performance of the selective approximation of Convolutional Neural Networks (CNNs) layers on NVIDIA mixed-precision architectures. We found that, even without affecting accuracy, the approximation from single to half precision of each layer has a different impact on both performance and output error.