Knowledge Interest-few shot learning
Zero-shot learning: If there is no sample of a certain category in the training set, you have to learn a powerful mapping to learn the characteristics of this new category, which requires horses to run without letting horses eat grass. ;
One-shot learning: There are samples for each category in the training set, but there are only one or a few samples. To learn a new category, horses are required to run and not to let horses eat more grass. ;
FEW-SHOT learning: training focused on a small number of samples can identify a type not seen before;
Meta Learning: Learn to Learn;
traditional learning: need massive training data, home to a prairie, horse horse you just eat.
support set: a very important small amount of prior knowledge
traning set: training set
query sample: query category
5way: Choose a few of 5
3shot: Choose 3