3 ms·
Am I wrong or that article gets many things wrong. These chips are not designed to do any learning, but run pre-trained models (much like FPGA's few years back
by dzhiurgis 9y ago
Am I wrong or that article gets many things wrong. These chips are not designed to do any learning, but run pre-trained models (much like FPGA's few years back or Google's TPU).
The actual heavy lifting (model training) still happens and the """cloud""".
- AceJohnny2 9y agoYou are wrong. For one thing, the learning of your face, and learning over time as your face changes (glasses, hairstyle, beard) happens fully locally.