Brick-by-Brick Approach to Modern Deep Learning Architectures

Erscheinungsjahr 2027
ISBN: 978-3-032-33067-3
Verlag: Springer Nature Switzerland AG

Buch, Englisch, 330 Seiten, Format (B × H): 155 mm x 235 mm

The textbook uses a unique "brick-by-brick" methodology to build understanding incrementally, starting from attention mechanisms and progressing systematically through the complete Transformer architecture. Each concept is explained with mathematical precision, intuitive visualizations, practical code examples, and chapter-end learning objectives that help readers consolidate key concepts. Every chapter also includes fully worked exercises and detailed solutions, enabling self-assessment and reinforcing both theoretical understanding and practical skills.

Perfect for master’s students and professionals entering deep learning, this book bridges the gap between superficial tutorials and dense research papers. Through a series of hands-on Python labs, readers do not simply learn about Transformers, they progressively build the key components of the architecture, train and evaluate their models, and gain the practical skills needed to work confidently with modern Transformer-based systems.

Zielgruppe

Upper undergraduate

Autoren/Hrsg.

Ros, Frédéric

Fachgebiete

• Technische Wissenschaften Elektronik | Nachrichtentechnik Nachrichten- und Kommunikationstechnik Signalverarbeitung
• Technische Wissenschaften Elektronik | Nachrichtentechnik Elektronik
• Mathematik | Informatik EDV | Informatik Informatik Natürliche Sprachen & Maschinelle Übersetzung
• Mathematik | Informatik Mathematik Mathematische Analysis Integralrechnungen- und -gleichungen

Inhaltsverzeichnis

Introduction.- Introduction To Transformers.- Self-Attention and the Query–Key–Value Mechanism.- Positional Encoding, Feed-Forward Networks, And Residual Connections.- The Transformer Encoder.- The Decoder and Autoregressive Generation.- Global Overview and Synthesis of Transformers.- Practical Labs.- Conclusion.

Über Autor(innen)

Frédéric Ros holds a Ph.D. from ENGREF (AgroParisTech, Paris) and an Habilitation à Diriger des Recherches (HDR) from the University of Orléans. He began his career at INRAE in image analysis for robotics and artificial systems before leading computer vision and artificial intelligence activities within major international technology companies, including Schlumberger and GEMALTO (now THALES). His expertise spans machine vision, artificial intelligence, data analysis, fuzzy systems, machine learning, digital watermarking, and evolutionary algorithms, with a long-standing interest in adaptive and self-tuning systems for complex industrial environments. He has also supervised several doctoral and research projects at the interface between academia and industry.

Since 2008, he has been Professor at the University of Orléans and an associate researcher at the PRISME Laboratory, where his research focuses on pattern recognition, unsupervised learning, knowledge integration, and pseudo-labeling. Alongside his research and teaching activities, he has led a technology park for 17 years and contributed extensively to the creation and development of a digital startup incubator, fostering innovation, technology transfer, and entrepreneurship.