BinaryFlex: On-the-Fly Kernel Generation in Binary Convolutional Networks

Anonymous

Nov 03, 2017 (modified: Nov 03, 2017) ICLR 2018 Conference Blind Submission readers: everyone Show Bibtex
  • Abstract: In this paper we present BinaryFlex, a binary neural network architecture that learns coefficients of predefined orthogonal binary basis instead of the conventional approach of learning binary parameters. We have demonstrated the feasibility of our approach for complex computer vision datasets such as ImageNet. Our architecture trained on ImageNet is able to achieve top-5 accuracy of 65.9% while being around 2x smaller than binary networks capable of achieving similar accuracy levels. By using deterministic basis our architecture offers a great deal of flexibility in memory footprint by allowing the application to decide whether to store the basis or to generate them on-the-fly when required.

Loading