Few shot Learning & meta learning: small sample learning principle and Siamese network structure (I)

Meta Learning： The name is mysterious , In fact, it means to let the network learn how to learn , namely learn to learn.

Few-Shot Learning：Meta Learning A kind of , The two can also be regarded as the same thing . It is different from the traditional supervised learning of training neural networks through a large number of data sets , It is not to let the machine accurately recognize the image of the test set , Its goal is to let neural networks learn . That is, the purpose of training neural networks through a large amount of data is not to let the network learn wolves 、 Characteristics of animals such as elephants , And then accurately classify the animal images we haven't seen . Instead, let neural networks understand the similarities and differences of things , Learn to distinguish between different things , That is, when the network training is completed , Give any two images you haven't seen on the Internet , Let the network judge whether the contents of the two images are the same or different .

Supervised Learning and Few-Shot Learning The difference between ：

• <