4 ms·
I see, so the idea is that enough of the quantization errors are sort of averaged out across the dimensions of the vector space to still be useful?
by s_m_t 3y ago
I see, so the idea is that enough of the quantization errors are sort of averaged out across the dimensions of the vector space to still be useful?
- singularity2001 3y agoThe way I think about it is finally it will end in a binary feature vector similar to 20Questions (male or female, alive or dead ...) just with 100s of dimensions