See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/388574425

Deep Learning: Foundations and Concepts: by Christopher M. Bishop and

Hugh Bishop, Springer Cham (2023). Hardback. ISBN 978-3031454677.

Article in Quantitative Finance · January 2025

DOI: 10.1080/14697688.2024.2436137

CITATIONS READS

0 180

3 authors, including:

Anastasis Kratsios Raeid Saqur

McMaster University University of Toronto
68 PUBLICATIONS 277 CITATIONS 16 PUBLICATIONS 30 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Raeid Saqur on 04 March 2025.

The user has requested enhancement of the downloaded file.

 Quantitative Finance

ISSN: (Print) (Online) Journal homepage: www.tandfonline.com/journals/rquf20

Deep Learning: Foundations and Concepts

by Christopher M. Bishop and Hugh Bishop, Springer Cham (2023). Hardback. ISBN
978-3031454677.

Blanka N. Horvath, Anastasis Kratsios & Raeid Saqur

To cite this article: Blanka N. Horvath, Anastasis Kratsios & Raeid Saqur (2024) Deep
Learning: Foundations and Concepts, Quantitative Finance, 24:12, 1725-1727, DOI:
10.1080/14697688.2024.2436137

To link to this article: https://doi.org/10.1080/14697688.2024.2436137

Published online: 31 Jan 2025.

Submit your article to this journal

Article views: 426

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=rquf20
Quantitative Finance, 2024
Vol. 24, No. 12, 1725–1727, https://doi.org/10.1080/14697688.2024.2436137

Book review

© 2023, Springer Cham

Deep Learning: Foundations and Concepts, by Christopher M. Today, much of the discussion around deep learning is dom-
Bishop and Hugh Bishop, Springer Cham (2023). Hardback. inated by their use for generative modelling and this book
ISBN 978-3031454677. delivers a comprehensive, historically rounded modern per-
spective on this on that account as well: Not only do they
Deep Learning by Christopher M. Bishop and Hugh Bishop list the different types of generative models available today
is an exceptional introductory resource for those looking to (VAEs and GANs) as classical Deep Learning Books† did,
understand the current state of deep learning. but the presentation of these is laid out in a way that prepares
Broad Topic: As the title says, ‘Deep Learning’ by Bishop the reader for the most-recent development of Generative AI
and Bishop focuses on the principles and applications of deep using Diffusion models and Large Language Models, which
neural networks, and the book delivers what the title promises dominate the modern generative AI landscape. For example,
in a clever, concise and self-contained manner that promotes the authors of this book review are thrilled with the presen-
profound understanding. tation that Bishop and Bishop chose by presenting the devel-
To lay the groundwork for understanding these topics, the opment Variational Autoencoders, which even make connec-
book starts with foundational concepts in probability the- tions between stochastic filtering and modern generative AI
ory and statistics, before guiding readers through classical as they are actively researching these connections from a dif-
models like Multilayer Perceptrons (MLPs),and other popu- ferent perspective. As a comparison, traditional‡ accounts of
lar network architectures like Convolutional Neural Networks generative models spent an overwhelming proportion of their
(CNNs) and Recurrent Neural Networks (RNNs), and delving focus on GANs.
into practical aspects like training and optimization tech-
niques for deep learning models. In doing so, there is an
† For example Deep Learning by Goodfellow et al.
apparent emphasis on developing the reader’s intuition about ‡ If such a thing as ‘traditional’ can even be said about an area that
how latent representations are learned by these models. is as new and dynamic as Generative Modelling.
1726 Book review

Bishops’s Deep learning book does not stop there, but pro- foundations of deep learning, such as universal approximation
gresses to more advanced topics, including GNNs, transform- theorems. Together, these resources offer a comprehensive
ers, and some of the latest innovations, including Denoising view of the field, from practical applications to the theo-
Diffusion Models. retical reasons behind their effectiveness. For students and
Market Need: As deep learning technologies become practitioners alike, this book provides a solid foundation for
increasingly integrated across a wide array of fields, indus- exploring the rapidly evolving landscape of deep learning.
tries, and business applications, the demand for accessible, Content Review The book contains twenty main chap-
engaging, and comprehensive resources on the subject contin- ters. The chapters progression seems coherent and natural -
ues to grow. Thus, books that provide a solid understanding of building up from pre-requisite knowledge then on to increas-
deep learning concepts are highly sought after by researchers, ingly involved chapters. Chapters 1–3 provide background
students, and industry professionals alike who want to stay mathematical knowledge; Chapters 5–9 discuss key building
ahead of the curve in this rapidly evolving field. blocks and technical primitives of neural networks; and the
While a few landmark textbooks have served as such sta- remaining Chapters 10–20 discuss relevant AI/ML models
ple, go-to reference over the past years (Murphy 2012, Bengio and architectures including Transformers, CNNs, GANs, and
et al. 2017)—including the author’s earlier seminal book GNNs.
titled Pattern Recognition and Machine Learning (Bishop and Chapter 11, Structured Distributions, feels somewhat out of
Nasrabadi 2006), the fast pace and rapid evolution of this field order; it would have been more logical to place it before the
demands newer editions including updated techniques more GNN chapter (Chapter 13).
frequently to raise the knowledge baseline required for new RNNs (and variants like LSTMs) get surprisingly little
researchers, students and practitioners alike to stay atop the attention, appearing only a handful of times. It gets a brief
fast-moving technological paradigm. discussion as a sub-section (12.2.5) on the Transformers chap-
This book effectively addresses that need by delivering a ter. Although RNNs/LSTMs are becoming outdated, their
strong foundational understanding of key concepts such as importance as precursor technologies and many conceptual
probability, distributions, neural network architectures, and understanding for topics (like BPTT, vanishing gradients
information theory. It effectively bridges the gap to advanced etc.) perhaps warranted more attention, especially from an
topics like transformers and normalizing flows, equipping educational perspective.
readers with the tools necessary to grasp the state-of-the-art The pedagogical explanations, figures, examples, and exer-
methods that define modern AI/ML practice. cises are intuitive and meet the high standard expected from a
About the authors: Christopher Michael Bishop is a Bishop book.
British computer scientist and Microsoft Technical Fellow, Similar Books and Unique Value Proposition Deep
known for his work in AI and machine learning. He is also an Learning by Bengio et al. (2017) is perhaps the most directly
author of influential textbooks and has received multiple hon- comparable textbook compared to this. For a more mathemat-
ours, including fellowships in the Royal Society and Royal ically advanced perspective, Murphys Probabilistic Machine
Academy of Engineering. His son, Hugh Bishop, is an applied Learning (Murphy 2012) and the updated probabilistic series
AI Scientist at Wayve, holding a graduate degree in machine of books (Murphy 2022, 2023) are also notable alternatives.
learning from Cambridge. In the preface, the authors give While these similar books cover overlapping topics, this
a heartwarming account about how the book project devel- book is distinct in its coverage of more recent advance-
oped, to pass their time writing and thinking about AI together ments, such as transformers, normalizing flows, and diffusion
during the lockdown. models, which are not discussed in the older texts.
The clear explanations and thoughtful progression make Additionally, the book provides relevant examples to go
complex topics accessible to beginners while still providing alongside the discussions. These examples are woven into the
depth for more advanced readers. Throughout the text, the chapters through exercises and illustrations.
authors provide valuable historical context, offering insights Targeted Audience & Prequisite Knowledge This book is
into the evolution of deep learning tools. For example, they targeted toward upper-year undergraduate and early graduate
explore variational models with latent stochastic variables and students. As such, it assumes readers have a general founda-
connect them to stochastic filtering models, such as the deep tion of knowledge in mathematics and basic machine learning
Kalman filter. This approach ensures that readers learn how concepts, which is appropriate for students at this level.
these models work and understand their development over Reasons to Read Reading and thoroughly working through
time. The authors also emphasize the importance of under- this book will raise a readers baseline understanding of fun-
standing the underlying principles of these models, helping damental concepts, preparing them to tackle more specialized
readers build a strong conceptual framework. textbooks in AI/ML subdomains such as computer vision,
Motivation and Goals of the Book The authors’ motiva- natural language processing, or reinforcement learning.
tion seems to be providing an accessible pedagogical expla- Value to the Quantitative Finance Community This book
nation of essential techniques, aiming to make the content is an excellent read for anyone in the quantitative finance com-
understandable for a broad audience from various fields. munity interested in applying AI/ML techniques in their field.
Subsequently, this book’s primary goal is to equip readers However, it is worth noting that many examples focus on
with the knowledge needed to understand and deploy deep medical domain problems in computer vision. Readers will
learning models in practice. It is an excellent complement to need to draw their own parallels and adapt these techniques to
other works that focus on learning theory and the analytical their specific domains, including quantitative finance.
 Book review 1727

References for advanced data representation. Blankas work is evidenced to have

influenced the financial industry, with numerous contributions imple-
Bengio, Y., Goodfellow, I. and Courville, A., Deep Learning, volume mented by leading organizations. It has also been recognized with
1, 2017 (MIT Press: Cambridge, MA). prestigious grants and honours, including being the inaugural recip-
Bishop, C.M. and Nasrabadi, N.M., Pattern Recognition and ient of Risk Magazines Rising Star Award in 2020 and the London
Machine Learning, volume 4, 2006 (Springer: New York). Mathematical Society’s Emmy Noether Fellowship 2024–2025.
Murphy, K.P., Machine Learning: A Probabilistic Perspective, 2012
(MIT Press: Cambridge, MA). Anastasis Kratsios is an Assistant Professor at McMaster Univer-
Murphy, K.P., Probabilistic Machine Learning: An Introduction, sity, affiliated with the Vector Institute in Toronto. His research
2022 (MIT Press: Cambridge, MA). focuses on the mathematical foundations of geometric deep learning
Murphy, K.P., Probabilistic Machine Learning: Advanced Topics, and the design of custom universal deep learning models optimized
2023 (MIT Press: Cambridge, MA). for problem-specific geometries. His work has appeared in numer-
ous top machine learning venues, from JMLR to NeurIPS. Anastasis
Blanka N. Horvath completed his postdocs at ETH Zrich in the mathematical finance
University of Oxford, Oxford, UK group and in the computer science and probability groups at the
Anastasis Kratsios Universitt Basel.
McMaster University, Hamilton, Canada
Raeid Saqur is a final year PhD candidate in Computer Science
Raeid Saqur and a lecturer for the Natural Language Computing (CS401/2511)
University of Toronto, Toronto, Canada course at the University of Toronto and Vector Institute for AI. He
© 2025 The Author(s) is cross-affiliated with Princeton University as a Fulbright Scholar,
collaborating with the CS NLP group. Raeid’s educational back-
Blanka N. Horvath is an Associate Professor at the University ground spans engineering (BSc.), applied computing (MSc.), and
of Oxford and researcher at the Oxford Man Institute, and a core an MBA from the Rotman School of Management. His research
member of the DataSig group affiliated with the Alan Turing Insti- interests range from multi-modal, grounded representation learning,
tute. With a career spanning across different countries, she has held language compositionality to long-horizon action planning for AI
tenure-track faculty positions at the Technical University of Munich agents by adaptive reinforcement learning, and human-like cogni-
and Kings College London, as well as postdoctoral appointments at tive biases in large language models. Raeid made his mark at IBM as
Imperial College London and ETH Zurich. Blanka’s research bridges their youngest Enterprise Thought Leader, and in the entrepreneurial
foundational theory and industry application. Her expertise encom- space. His industry innovations are recognized by the prestigious
passes stochastic volatility modelling, with pioneering contributions Peter F. Drucker Effective Executive Scholarship and numerous
to rough volatility models, and extends to cutting-edge developments leadership awards. Beyond his scholarly work, Raeid has mentored
in generative models, emphasizing the use of rough path signatures over 300 students in AI and NLP.

View publication stats