Transfer Learning
Explore transfer learning to understand how pretrained deep neural networks can be adapted for new but related image recognition tasks. This lesson shows how to use existing models like ResNet and MobileNet to speed up training and improve accuracy despite limited data, essential for deploying efficient deep learning apps on Android.
What is transfer learning?
Complex DL models require large datasets with millions of training data points. Suppose we want to build a DL model for object detection in images, but we have a small dataset of a few thousand images. Training a complex model on the available data might not be a good idea due to the insufficiency of the data. The model is likely to overfit the data and might provide poor generalization results on test data.
In this case, we resort to transfer learning, an ML method that allows us to use the knowledge gained from developing a model for a particular task to solve a new but similar task. In the context of DL, transfer learning stores the gained knowledge (network parameters) from training a large neural network and applies it to a different but related problem. A car recognition model, for instance, can be applied to recognize buses and ...