Preface
Deep learning uses multilayered neural networks trained with large data sets to
solve complex information processing tasks and has emerged as the most successful
paradigm in the field of machine learning. Over the last decade, deep learning has
revolutionized many domains including computer vision, speech recognition, and
natural language processing, and it is being used in a growing multitude of applica-
tions across healthcare, manufacturing, commerce, finance, scientific discovery, and
many other sectors. Recently, massive neural networks, known as large language
models and comprising of the order of a trillion learnable parameters, have been
found to exhibit the first indications of general artificial intelligence and are now
driving one of the biggest disruptions in the history of technology.
Goals of the book
This expanding impact has been accompanied by an explosion in the number
and breadth of research publications in machine learning, and the pace of innova-
tion continues to accelerate. For newcomers to the field, the challenge of getting
to grips with the key ideas, let alone catching up to the research frontier, can seem
daunting. Against this backdrop, Deep Learning: Foundations and Concepts aims
to provide newcomers to machine learning, as well as those already experienced in
the field, with a thorough understanding of both the foundational ideas that underpin
deep learning as well as the key concepts of modern deep learning architectures and
techniques. This material will equip the reader with a strong basis for future spe-
cialization. Due to the breadth and pace of change in the field, we have deliberately
avoided trying to create a comprehensive survey of the latest research. Instead, much
of the value of the book derives from a distillation of key ideas, and although the field
itself can be expected to continue its rapid advance, these foundations and concepts
are likely to stand the test of time. For example, large language models have been
evolving very rapidly at the time of writing, yet the underlying transformer archi-
tecture and attention mechanism have remained largely unchanged for the last five
years, while many core principles of machine learning have been known for decades.
v
