Document worth reading: “Few-shot Learning: A Survey”

The quest of `can machines assume’ and `can machines do what human do’ are quests that drive the occasion of artificial intelligence. Although present artificial intelligence succeeds in a number of information intensive functions, it nonetheless lacks the ability of learning from restricted exemplars and fast generalizing to new duties. To kind out this disadvantage, one has to point out to machine learning, which helps the scientific analysis of artificial intelligence. Particularly, a machine learning disadvantage often known as Few-Shot Learning (FSL) targets at this case. It can rapidly generalize to new duties of restricted supervised experience by turning to prior data, which mimics human’s ability to build up data from few examples by the use of generalization and analogy. It has been seen as a test-bed for precise artificial intelligence, a way to chop again laborious information gathering and computationally costly teaching, and antidote for unusual circumstances learning. With in depth works on FSL rising, we give a whole survey for it. We first give the formal definition for FSL. Then we degree out the core issues with FSL, which turns the problem from ‘how one can treatment FSL’ to ‘how one can deal with the core factors’. Accordingly, current works from the supply of FSL to the newest revealed ones are categorized in a unified taxonomy, with thorough dialogue of the professionals and cons for numerous lessons. Finally, we envision doable future directions for FSL in the case of disadvantage setup, strategies, functions and precept, hoping to produce insights to every newcomers and expert researchers. Few-shot Learning: A Survey