By Simon Prince, laptop science Professor on the College of Alberta. To be printed by MIT Press, Dec 2023. The writer shares the related Jupyter notebooks on his web site, right here. Highly regarded, it received over 5,000 likes when the writer introduced the upcoming guide on LinkedIn. I pre-ordered my copy.

Abstract

An authoritative, accessible, and up-to-date therapy of deep studying that strikes a realistic center floor between principle and apply.

Deep studying is a fast-moving subject with sweeping relevance in at this time’s more and more digital world. Understanding Deep Studying offers an authoritative, accessible, and up-to-date therapy of the topic, overlaying all the important thing matters together with latest advances and cutting-edge ideas. Many deep studying texts are crowded with technical particulars that obscure fundamentals, however Simon Prince ruthlessly curates solely crucial concepts to offer a excessive density of essential data in an intuitive and digestible type. From machine studying fundamentals to superior fashions, every idea is offered in lay phrases after which detailed exactly in mathematical type and illustrated visually. The result’s a lucid, self-contained textbook appropriate for anybody with a fundamental background in utilized arithmetic.

• Up-to-date therapy of deep studying covers cutting-edge matters not present in current texts, similar to transformers and diffusion fashions
• Quick, centered chapters progress in complexity, easing college students into tough ideas
• Pragmatic strategy straddling principle and apply provides readers the extent of element required to implement naive variations of fashions
• Streamlined presentation separates essential concepts from background context and extraneous element
• Minimal mathematical stipulations, intensive illustrations, and apply issues make difficult materials extensively accessible
• Programming workouts supplied in accompanying Python Notebooks

Contents

Every subject beneath is a chapter.

• Introduction
• Supervised studying
• Shallow neural networks
• Deep neural networks
• Loss features
• Coaching fashions
• Gradients and initialization
• Measuring efficiency
• Regularization
• Convolutional networks
• Residual networks
• Transformers
• Graph neural networks
• Unsupervised studying
• Generative adversarial networks
• Normalizing flows
• Variational autoencoders
• Diffusion fashions
• Deep reinforcement studying
• Why does deep studying work?
• Deep studying and ethics

Get Your Copy

You may pre-order the guide on Amazon, right here. If as well as you might be on the lookout for state-of-the-art new developments (slightly than a strong, fashionable introduction on deep studying), be at liberty to take a look at my very own books. They characteristic upcoming developments and new open-source Python libraries, such because the highly effective full multivariate KS analysis metrics, and new strategies similar to NoGAN, not based mostly on neural networks, together with basic GAN (and how you can make it extra generic). NoGAN and its sisters run 1000x quicker and persistently ship higher and explainable outcomes, with auto-tuning. Thus, saving plenty of GPU and cloud time, and in consequence, a number of value financial savings. My books can be found right here.

To not miss future articles and entry members-only content material, sign-up to my free publication, right here.