Hello everyone. In this blog article, we will understand the concept of Transfer Learning. Deep Learning is awesome. It has given wings to computer vision and object detection/classification. However, we need a lot of data to train a good detection model which may not be readily available. In this article, I will demonstrate the use of transfer learning to train a model on less data and still getting good results. Let’s start with the article.
Human beings have a tendency to gain the knowledge from one task and apply it to another task. For instance, the knowledge gained from riding bicycle is used to ride Scooter.
Transfer learning is a process where the knowledge gained from one task is used to solve another task. In transfer learning, we use the pre-existing models created by someone else for their problem and these models were trained for several weeks, on various GPUs and went through painful hyper-parameter optimization. Thus by using transfer learning, we can do effective initialization of our models by using the weights of the pre-existing models and therefore, it can optimize the performance of the model. Transfer learning can be used where we have small dataset for our problem and we use the pre-existing model trained on relatively large dataset and thus, getting good results.

#deep-learning #data-science #computer-vision #machine-learning #transfer-learning