DL-gleaning: An approach for Improving inference speed and accuracy

HyunYong Lee and Byung-Tak Lee

Feb 06, 2017 (modified: Feb 06, 2017) ICLR 2017 workshop submission readers: everyone
  • Abstract: Improving inference speed and accuracy is one of the main objectives of current deep learning-related research. In this paper, we introduce our approach using middle output layer for this purpose. From the feasibility study using Inception-v4, we found that our approach has potential to reduce the average inference time while increasing the inference accuracy.
  • TL;DR: An approach exploiting middle output layers to improve inference accuracy and speed
  • Conflicts: etri.re.kr
  • Keywords: Deep learning, Supervised Learning