This strategy uses some $\theta_0$ from some pre-trained models related to the current task as a starting point and uses $D_{train}$ to find better $\theta_0$.  

Modify Search Algorithm by Refining Existing Parameters

This strategy uses meta-learners learn a $\theta_0$ on similar tasks as the few-shot task, and the few-shot task will further refine $\theta_0$. For comparison, the strategy of refining existing parameters has a fixed $\theta_0$, while this strategy dynamically learn $\theta_0$ using meta learner.

Modify Search Algorithm by Refining Meta-Learned Parameters

This strategy uses meta-knowledge to learn an optimizer, which can directly output a step size and step direction.

Modify Search Algorithm by Learning the Optimizer

In terms of how searching strategy is affected by prior knowledge, searching algorithm motivated few-shot learning methods can be categorized as:

1)	Refining existing parameters

2)	Refining meta-learned parameters.

3)	Learning the optimizer


University of Notre Dame

Few-Shot Learning can be divided by what part that prior knowledge has been added to:

1)	Data: The training data is augmented by prior knowledge

2)	Model: The possible model space is constrained according to prior knowledge

3)	Algorithm: The search strategy in hypothesis model space has been improved.

Another classification of Few-Shot Learning

Wang, Y., Yao, Q., Kwok, J. T., & Ni, L. M. (2020). Generalizing from a few examples: A survey on few-shot learning. ACM computing surveys (csur), 53(3), 1-34.

Generalizing from a Few Examples: A Survey on Few-Shot Learning

Prior knowledge can be used to augment the training set, and below are several ways of achieving it:

1.	Transforming Samples from Training Set

2.	Transforming Samples from a Weakly Labeled or Unlabeled Data Set

3.	Transforming Samples from Similar Data Sets


Ways of Augmenting Data of Few-Shot Learning

1)	Multitask Learning

2)	Embedding Learning

3)	Learning with External Memory

4)	Generative Modeling


Learn Before

Related

Learn After