how to do transfer learning in practice? how to train only outer layer?
There quite a few more advanced courses offered by deeplearning.ai that will teach you how to do that. Browse through the courses/specializations when you have time.
does transfer learning useful in case of neural networks or we can use it for other algorithms as well ?
Hello @rahulpatil0012, welcome to our community!
Transfer Learning (TL) is useful and possible for not just Neural Network, but other modeling techniques such as Linear Regression, Bayesian Network and so on. I am sure you can name more other techniques if you go through the literatures or may be doing a search on Google Scholar.
We always link transfer learning to Neural Network, and particularly to Computer Vision models because there are many such trained models available online for others to do TL. Also, before doing a TL, you need to figure out the similarities between the inputs of the model you want to transfer, and the inputs of your current problem that you want to solve. Regarding to such matter, Computer Vision models are simple to overcome: because they take photos as inputs and as long as your current problem is also about photos, then that is enough for you to try out TL on Computer Vision models.
just as very simple example to transfer a certain style to a new image, I played around with couple of pictures in this repo with a training on a CPU in a reasonable time.
Of course much more advanced applications are possible, e.g. in Conputer Vision to the enhance label data base (e.g. by transfer from summer setting to winter), see also: