Zero shot, one shot and few shot

Catalog

One 、Zero-shot learning （ Zero sample learning , abbreviation ZSL）

Two 、One-shot learning

3、 ... and 、Few-shot learning

---

For the division of these concepts , Mainly from the test volume category , From the perspective of the training set category and the number of samples corresponding to the category .

One 、Zero-shot learning （ Zero sample learning , abbreviation ZSL）

Task definition ：

Use training set data to train the model , It enables the model to classify the objects of the test set , But there is no intersection between training set categories and test set categories ; During this period, you need to use the description of the category , To establish the connection between the training set and the test set , Thus making the model effective .

ZSL It is hoped that our model can classify categories it has never seen , Let the machine have reasoning ability , Realize true intelligence . Zero of them （Zero-shot） It refers to the category objects to be classified , Don't study once .

such as ： Suppose our model has been able to identify horses , Tigers and pandas , Now we need this model to recognize zebras , Then we need to tell the model , What kind of object is a zebra , But you can't directly let the model see the zebra . So the information that the model needs to know is the horse sample 、 Samples of tigers 、 Samples of pandas and labels of samples , And descriptions of the first three animals and zebras .
 

Two 、One-shot learning

If the training is concentrated , There is only one sample of different categories , Then become One-shot learning.

One-shot learning Belong to Few-shot learning A special case of .

3、 ... and 、Few-shot learning

If the training is concentrated , There are only a few samples in different categories , Then become Few-shot learning.

It is to give a small number of samples of the category to be predicted by the model , Then let the model predict the category by looking at other samples of the category . such as ： Show children a picture of a panda , Then when children go to the zoo and see pictures of pandas , It can be identified as a panda .

Recommended reading small sample learning summary ：