XNOR-NetΒΆ
$$ w \in \mathbb{R} $$
$$ \hat{w} = w \in \mathbb{R} $$
$$ \hat{w}_q = \mbox{sign}(w)\alpha $$
$$ \alpha = \frac{1}{n}||\hat{W}||_{\ell_2} $$
Experiment:
alexnet
Reference:
[1] Binarynet: Training deep neural networks with weights and activations constrained to +1 or -1