Automated Deep Learning Using Neural Network Intelligence: Develop and Design PyTorch and TensorFlow Models Using Python

This book teaches how to optimize, develop, and design PyTorch and TensorFlow models using the Microsoft Neural Network Intelligence (NNI) toolkit. It covers topics such as hyper-parameter optimization, architecture search, model compression, and feature engineering, with practical examples and techniques for automated deep learning model development.

Format: Paperback / softback
Length: 384 pages
Publication date: 21 June 2022
Publisher: APress

This comprehensive book delves into the realm of optimizing, developing, and designing PyTorch and TensorFlow models using the Microsoft Neural Network Intelligence (NNI) toolkit. It serves as a valuable resource for practitioners seeking to leverage automated deep learning approaches to solve complex problems.

The book begins by introducing the basics of NNI toolkit usage and providing methods for addressing hyper-parameter optimization tasks. It explains the black-box function maximization problem and guides you through the process of preparing a TensorFlow or PyTorch model for hyper-parameter tuning, launching an experiment, and interpreting the results.

The subsequent chapters delve into optimization tuners and the search algorithms upon which they are based, including Evolution search, Annealing search, and the Bayesian Optimization approach. The book also covers the Neural Architecture Search, which enables the development of deep learning models from scratch. It presents multi-trial and one-shot searching approaches for automatic neural network design, along with practical techniques for constructing a search space and launching an architecture search using state-of-the-art exploration strategies such as Efficient Neural Architecture Search (ENAS) and Differential Architectural Search (DARTS).

Furthermore, the book emphasizes the importance of model compression and feature engineering methods in automated deep learning. It discusses techniques for reducing model size, enhancing model performance, and developing efficient feature representations. Additionally, the book explores performance techniques that enable the creation of large-scale distributive training platforms using NNI.

By the end of this book, readers will possess a comprehensive understanding of the toolkit of automated deep learning methods. The practical examples and techniques presented within will empower practitioners to apply these methods to their own projects, enabling them to optimize, develop, and design PyTorch and TensorFlow models for a wide range of applications.

Weight: 764g
Dimension: 177 x 254 x 26 (mm)
ISBN-13: 9781484281482
Edition number: 1st ed.