A very interesting and well-grounded work, Gradient expectations provides a thorough overview and explanation of the structure and origins of prediction and predictive neural networks.
The book is structured into eight chapters, starting with an introduction to the basics of prediction and a definition of gradients. Chapter 2 covers looks at the properties of gradients as a formalization of the concepts of prediction, their temporal dimension, and their relatedness to the notions of goal and control in egocentric or ecocentric systems, where bias or objective truth take prevalence. Chapter 3 presents a very detailed account of how the processes of prediction translate in biological, chemical, and physiological processes. Chapter 4 explains different approaches and theories about linking energy with different prediction methods, aiming at optimal performance for networks with different levels of complexity.
Chapter 5 is dedicated to the relation between perception and information, and explains the concept, the mechanisms, and the formalization of predictive coding as a way to compress the neural data spaces by removing the inhibitory neurons based on evidence from the retina physiology. It also shows how this principle can be applied to machine learning in practice. Chapter 6 discusses the emergence of predictive networks, where it interprets the evolution of the human brain and also senses the structure and the functioning of the brain with hands-on examples of common human activities. Chapter 7 turns to artificial predictive networks and presents a historical account, from backpropagation to evolutionary algorithms that emulate direct or indirect genome encodings and achieve optimal performance for tasks like classification or the autonomous reaching of a goal. It also explains how evolving networks can be applied to deep learning, predictive coding contexts, and more, and recommends in what contexts they are respectively most useful. Finally, chapter 8 concludes the text with a summary of the main ideas about prediction and the related notion of expectations.
Presenting a very-well-structured methodology, this book will provide great value to students and scholars interested in the foundations of prediction, neural networks, their origins, formalization, and applications. Demonstrating and explaining the connection between networks and predictive processes with human biology, physiology, chemistry, brain functioning, and physics brings the reader to an in-depth understanding of the nature of prediction, as well as the foundation of artificial neural networks and their applications.