Nikhil Ketkar - Deep Learning with Python: Learn Best Practices of Deep Learning Models with PyTorch

Any source code or other supplementary material referenced by the author in this book is available to readers on GitHub via the books product page, located at www.apress.com/978-1-4842-5363-2 . For more detailed information, please visit http://www.apress.com/source-code .

This book has been drafted with a unique approach. The second edition focuses on the practicality of the topics within deep learning that help the reader to embrace modern tools with the right mathematical foundations. The first edition focused on introducing a meaningful foundation for the subject, while limiting the depth of the practical implementations. While we explored a breadth of technical frameworks for deep learning (Theano, TensorFlow, Keras, and PyTorch), we limited the depth of the implementation details. The idea was to distill the mathematical foundations while focusing briefly on the practical tools used for implementation.

A lot has changed over the past three years. The deep learning fraternity is now stronger than ever, and the frameworks have evolved in size and adoption. Theano is now deprecated (ceased development); TensorFlow saw huge adoption in the industry and academia; and Keras became more popular among beginners and deep learning enthusiasts. However, PyTorch has emerged recently as a widely popular choice for academia as well as industry. The growing number of research publications that recently have used PyTorch over TensorFlow is a testament to its growth within deep learning.

On the same note, we felt the need to revise the book with a focus on engaging readers with hands-on exercises to aid a more meaningful understanding of the subject. In this book, we have struck the perfect balance, with mathematical foundations as well as hands-on exercises, to embrace practical implementation exclusively on PyTorch. Each exercise is supplemented with the required explanations of PyTorchs functionalities and required abstractions for programming complexities.

Part I serves as a brief introduction to machine learning, deep learning, and PyTorch. We explore the evolution of the field, from early rule-based systems to the present-day sophisticated algorithms, in an accelerated fashion.

Part II explores the essential deep learning building blocks. Chapter , we look at orchestrating all the building blocks discussed through so far, along with the performance metrics of deep learning models and the artifacts required to enable an improved means for trainingi.e., regularization, hyperparameter tuning, overfitting, underfitting, and model capacity. Finally, we leverage all this content to develop a deep neural network for a real-life dataset using PyTorch. In this exercise, we also explore additional PyTorch constructs that help in the orchestration of various deep learning building blocks.

Part III covers three important topics within deep learning. Chapter concludes the book by looking at some of the recent trends within deep learning. This chapter is only a cursory introduction and does not include any implementation details. The objective is to highlight some advances in the research and the possible next steps for advanced topics.

Overall, we have put in great efforts to write a structured, concise, exercise-rich book that balances the coverage between the mathematical foundations and the practical implementation.

I would like to thank my colleagues at Flipkart and Indix, and the technical reviewers, for their feedback and comments. I will also like to thank Charu Mudholkar for proofreading the book in its final stages.

Nikhil Ketkar

I would like to thank my beloved wife, Divya, for her constant support.

Jojo Moolayil