Transfer Learning with TensorFlowJS

In practice, I believe that in most cases rather than creating models from scratch you will create models which already trained and solve a problem that is close to yours. This technique is called Transfer Learning. As you may already know, one big issue of training models from scratch is that we need to collect and label a huge amount of data and it's pretty time-consuming work that may be not affordable for your project. Also, it is computationally very expensive to train a neural network on millions of images and it may require weeks of training on multiple GPUs. The mental workflow of Transfer Learning is depicted in figure 1.