One of the exciting techniques in machine learning is transfer learning, where you transfer knowledge from one task to another similar task.

For example, 💡 knowledge learns during image classifications can be used for X-ray classification.

The intuition is that the pre-trained model already has known many low-level features in images like edges, curves, etc.

When should we use Transfer Learning?

Training any machine learning model requires vast amounts of data. Unfortunately, the problem you are working on might not have sufficient data.

You can leverage transfer learning and fine-tune your model by utilizing weights trained on some popular models like ImageNet.

We humans always use transfer learning; skills learn from one task can be used in another job. e.g., If we know how to ride a bicycle🚴‍♀️, we can transfer skills like balancing; while learning to ride a motorcycle 🏍.

Humans are good at abstracting conceptual knowledge irrespective of which task they learned. That's one of the things which separates us from machines.

Traditionally machine learning was very task-specific. leveraging transfer learning will be one step closer to AGI (Artificial General Intelligence) 🧠

Image Source