3 ms·
The information is embedded in the weights of various layers in the network. Trying to remove that information by editing weights would be like trying to alter
by mr_toad 4y ago
The information is embedded in the weights of various layers in the network. Trying to remove that information by editing weights would be like trying to alter someone’s memory by tinkering with synapses.
The only way to be completely sure of removing information would be to re-train the model without that data.