Comment by strangecasts 2 years ago Nitpick: it's not zero-shot but few-shot - it still needs a training set with prototypes to go off 1 comment strangecasts Reply cs702 2 years ago Yes, it needs prototypes (i.e., possible neighbors) -- but it doesn't need any kind of pretraining to learn to map samples to a feature space.It uses gzip instead of a pretrained model to map samples to the feature space.
cs702 2 years ago Yes, it needs prototypes (i.e., possible neighbors) -- but it doesn't need any kind of pretraining to learn to map samples to a feature space.It uses gzip instead of a pretrained model to map samples to the feature space.
Yes, it needs prototypes (i.e., possible neighbors) -- but it doesn't need any kind of pretraining to learn to map samples to a feature space.
It uses gzip instead of a pretrained model to map samples to the feature space.