3 ms·
You can probably apply the same techniques 'Deep neural networks are robust to weight binarization and other non-linear distortions' used to get to 0.68 bits /
by eru 3y ago
You can probably apply the same techniques 'Deep neural networks are robust to weight binarization and other non-linear distortions' used to get to 0.68 bits / weight to get your ternary weights below one bit; so you can claim they are still one-bit networks.