author: @himanshustwts

Let’s say you’ve been to Arijit’s concert and you enjoyed his songs and piano play.🎹

Now, you’re very keen to learn and play piano. However, you’ve already played instruments like violin or guitar before so you’ll not start from zero to learn piano. You already understand the rhythm, timing, and basics of music theory, so it becomes easier and faster for you to learn from your previous experiences.

Now, imagine if we could similarly teach a machine — letting it use the knowledge it already has to learn something new quickly!!

In this blog, we’ll explore the foundations and implementation of Transfer Learning from very scratch (building intuitions) covering the following topics:

• WHY Transfer Learning?
• Practical scenarios
• The Architecture behind Transfer Learning
• Mathematics behind Transfer Learning

• Perspective 1: Representation Learning
• Perspective 2: Bayesian Approach

• Hands-on Implementation

WHY TRANSFER LEARNING?

Let’s understand Transfer Learning in greater detail.

It is a machine learning method where a model developed for a first task is reused as the starting point for a model on a second task.

If we train a model to classify birds and cats we use the same model modified a little bit in the last layer and then use a new model to classify bees and dogs. So, it’s a very popular approach in deep learning that allows rapid generation of new models.