Transfer learning (TL) is a research problem in machine learning (ML) that focuses on storing knowledge gained while solving one problem and applying it to a different but related problem. For example, knowledge gained while learning to recognize cars could apply when trying to recognize trucks.
We humans have a very special power that allows us to gain new skills faster as we learn more and more. We have the ability to use knowledge across different tasks. When we acquire knowledge when learning to do some task, we utilize this knowledge in some other related task in one way or other.
When you were a kid, you learned to ride a bicycle. After many failed attempts, you understood how to balance the bicycle and all the dynamics that are required to ride it. The knowledge you gained while learning to ride a bicycle, you made use of it when you started to learn to ride a bike. The balancing and control dynamics were very much related, so you were able to learn to ride a bike faster than you learned to ride the bicycle. Some simple examples of this transfer are,
- Know how to play and instrument -> Learn to play a song
- Know basic mathematics -> Learn advanced mathematics
- Know how to sketch and computers -> Learn to make animations In each of the above scenarios, we don’t learn everything from scratch when we attempt to learn new aspects or topics. We transfer and leverage our knowledge from what we have learnt in the past! Transfer Learning, the ability to use the knowledge from an existing model trained on some task to solve some other related task.
