Deep neural networks are very powerful, but they usually require huge corpora for training. These are usually not available so pre-trained networks can be used. How do social scientists reconcile the potential bias that pre-trained networks can introduce while maintaining validity?
Deep neural networks are very powerful, but they usually require huge corpora for training. These are usually not available so pre-trained networks can be used. How do social scientists reconcile the potential bias that pre-trained networks can introduce while maintaining validity?