Maximising accuracy with limited training data using Transfer Learning
Introduction
Transfer Learning is a technique in Deep Learning in which the model trained in one problem is reused partly in another problem. This technique can be a powerful tool to acquire the highest accuracy with very limited data, which otherwise would not be possible.
Transfer learning is demonstrated using pytorch and pretrained model called vgg16 in this colab file.
This lesson is part of tutorial, ‘Pytorch for Deep Learning and Computer Vision’ by Rayan Slim, Jad Slim, Amer Sharaf and Sarmad Tanveer. It is available in learning.oreilly.com
Learnings
- Pretrained model approach to implement transfer learning
- Various pretrained models in computer vision domain, viz. vgg16, vgg32, alexanet, etc.
- Implementation and practical example to demonstrate
References:
This lesson is part of tutorial, ‘Pytorch for Deep Learning and Computer Vision’ by Rayan Slim, Jad Slim, Amer Sharaf and Sarmad Tanveer. It is available in learning.oreilly.com