The usual way of training a neural network consists of the following steps:
- Preparing and labeling a dataset
- Developing a neural network architecture
- Starting to train by initializing weights randomly
- Training the network and iterating the process once again until the desired result is achieved
- Saving the model
You would expect to execute the same steps again and again as you start working on a new problem with different data. Instead of training a new network from scratch with randomly initialized weights, you can reuse the structure and weights from another working model which was previously used by you or an open community. The process of using existing neural networks to solve a different problem is referred to as using a pre-trained network. The first network is going to be your pre-trained network. The second one is the network you...