XNOR_Net

This is a convolution neural network of binarized activations and weights.

Details

$I, W$ is the real-value input and weights.

We let $I_b \approx \alpha \sign{I}$, where $\alpha = \frac{1}{n}{||I||_{\el_1}}$

write by dnvtmf

References

• Binarized Neural Networks: Training Deep Neural Networks with Weights and Activations Constrained to +1 or -1, Matthieu Courbariaux, Itay Hubara, Daniel Soudry, Ran El-Yaniv, Yoshua Bengio
• Matthieu Courbariaux, Itay Hubara, Daniel Soudry, Ran El-Yaniv, Yoshua Bengio:XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks