Recurrent Neural Networks: From Simple to Gated Architectures

This textbook provides a comprehensive treatment of recurrent neural networks, covering analytical and design steps from scratch. It focuses on the basics and nuances of RNNs, providing a technical and principled approach for training and using coding and deep learning computational frameworks. The authors adopt adaptive non-convex optimization with dynamic constraints to leverage the power of RNNs in organizing learning and training processes. They also introduce strategic co-training of output layers and hidden layers to generate more efficient internal representations and accuracy performance.

Format: Paperback / softback
Length: 121 pages
Publication date: 05 January 2023
Publisher: Springer Nature Switzerland AG

This comprehensive textbook offers a concise yet thorough exploration of recurrent neural networks, providing analytical and design steps from scratch. It delves into the realm of general recurrent neural networks, employing principled training methodologies that facilitate the (generalized) backpropagation through time (BPTT). The author's primary focus lies in elucidating the fundamental principles and intricate details of recurrent neural networks, offering a technical and principled treatment that aims to leverage coding and deep learning computational frameworks, such as Python and Tensorflow-Keras.

The treatment of recurrent neural networks encompasses a holistic approach, ranging from simple to gated architectures, while incorporating the technical machinery of adaptive non-convex optimization with dynamic constraints to harness its systematic prowess in organizing learning and training processes. This enables the seamless flow of concepts and techniques that provide robust support for design and training choices. By adopting the authors' approach, readers gain the ability to strategically co-train output layers through supervised learning and hidden layers through unsupervised learning, thereby fostering the generation of more efficient internal representations and enhancing accuracy performance. As a result, this textbook empowers readers to craft tailored designs that incorporate proficient procedures for recurrent neural networks in their specific applications.

Weight: 232g
Dimension: 235 x 155 (mm)
ISBN-13: 9783030899318
Edition number: 1st ed. 2022