1. Motivation

• Traditional approaches of neural networks have problem that need abundant amount of data. To overcome such problem, few-shot learning is proposed which target to perform one-shot classification like human.
• The problem of such approach is that it is easy to happend overfitting. To address such problem, this paper proposed prototypical networks. To model is based on the idea that there exists an embedding in which points cluster around a single prototype representation for each class. Specifically, class means are used as prototypes for each class.
  

2. Method

2.1 Model

• In equation (1), it explains how the proposed model calculates the mean vector of embedded support points.
  

• Therefore, the probability of x is a specific class k is like equation (2)
  

3. Result