settings:huge labeled data for training are difficult to access
then we introduce the trans. learning
definition : adjust slightly the trained model to the new problem
bottleneck denotes the layers before last fully-connected layer, namely the features-extracted part
less effective but more time-conservative
网友评论