Table 1.

Similarities and differences between MN and concept acquisition through meta-learning (CAML).

	MN	CAML
Pre-training goals	The goal is to give the network the ability to measure the similarity between two samples.	The goal is to make CAML learn the optimal initialization by updating on different tasks.
Need fine-tuning	MN can achieve classification tasks without fine-tuning, but the values generated after fine-tuning are more accurate.	Yes
Fine-tuning process	MN predicts new samples, and the error between the predicted value and the label is used to update the network parameters.	CAML updates parameters by one or a few steps of gradient descent with a few positive examples from the new task.