Training Deep Learning Models with Transfer Learning

One way to train a deep learning algorithm in MATLAB^® is through transfer learning.

In machine learning, transfer learning is the transfer of knowledge from one learned task to a new task. In the context of neural networks, it is transferring learned features of a pretrained network to a new problem.

The common practice in deep learning for such cases is to use a network that is trained on a large data set for a new problem. While the initial layers of the pretrained network can be fixed, the last few layers must be fine-tuned to learn the specific features of the new data set. Transfer learning usually results in faster training times than training a new convolutional neural network because you do not need to estimate all the parameters in the new network.