Generally, transfer learning means transferring pre-trained machine learning model representations to another problem. In recent years, this is becoming a popular means of applying DL models to a problem, especially in image processing and recognition, as it enables training a DL model with comparatively little data.
The following diagram shows two models:
- An architecture for a standard DL model (a)
- An architecture for a transfer-learning DL model (b):
As shown in the figure of an architecture for a standard DL model, a fully trained neural net takes input values in an initial layer and then sequentially feeds this information forward with necessary transformation until the second-to-last layer (which is also known as the bottleneck layer) has constructed a high-level representation of the input that can more easily be transformed...
