Hybrid Fine-Tuning Strategy for Few-Shot Classification

Algorithm 1

Pseudocode for the AFT module.

	Input: the validation dataset: , the pretrained model
	Output: hybrid fine-tuning strategy represented by adaptive epoch:
	Hyper-parameters: the total number of pseudofine-tuning tasks , the maximum number of epochs , the number of epochs contained in a chunk , the number of nodes contained in a chunk , and the starting chunk number .
	for: do
	= RandomTaskSample ; #get train and test sets for task
	= FSLDA ; #initialize the model by FSLDA
	= Evaluate ;
	= /; #number of chunks
	= Backpropagation ; #train the model by epochs
	for each chunk : do
	for each node : do
	= Backpropagation ;
	= Evaluate ;
	end for
	= ; #the average accuracy of chunk
	if% or  =  then
	; #adaptive epochs
	if then; #optimal hybrid epoch
	break;
	end if
	end for
	end for
	Set  = , . #number of tasks needing finetuning
	if then ; #get the quantile of
	else .