Fundamentals of Deep Learning (Fundamentals of Deep Learning) (AI column)

I recently finished reading the first edition of this book, which was published in English. 2017 year:) Author in 20 graduated from MIT, in 2016 Remedy started to apply AI in healthcare, and in 2020 Ambience Healthcare is developing an “ambient AI” platform to help doctors automate routine documentation. He is also known in the AI community as the author of one of the first popular deep learning books.

As for the book, I bought it for years. 5 Back, because I liked the accessibility of the presentation, I flipped through it, and then postponed, after reading half the book and could not finish. The second updated edition of the book was published in May. 2022 One year ago, I just finished reading it yesterday. In the first edition of the book, many of the initial moments and “state of the art” are well disclosed before the appearance of “Attention is all you need”, so on the one hand it is seriously outdated, and on the other hand it covers already classical topics from the basics of the mathematical apparatus to the construction and training of different types of neural networks. In the beginning, the necessary elements of linear algebra and probability theory are considered, after which the concept of neural networks and the error backpropagation algorithm are introduced. After that, the author introduces readers to classical architectures: fully connected networks and their learning issues. (gradient descent, the problem of retraining, etc.)Convolutional neural networks for image analysis, models for sequence analysis (Recurrent networks, natural language processing)and methods of reducing dimensionality (for example, autoencoders). Some chapters are devoted to generative models (including GAN/Variational Autoencoder)Interpretability models and even basic ideas of reinforcement learning are discussed.

The book is applied in nature: key ideas are supported by examples of Python code. First edition. (2017) The TensorFlow library was used for illustrations, and in the second edition 2022 Examples updated on PyTorch (Added chapter on network implementation in PyTorch). Example code openly The reader can not only study the theory, but also immediately try to implement the basic models of neural networks with their own hands.

In general, the thematic scope of the book is quite wide, allowing you to get a holistic view of deep learning: from basic mathematics to the main types of neural network architectures and their application in practice. In 2017 The book was well received by the audience and the general opinion was that Buduma was able to explain complex things in simple language. But if we talk about the relevance of information, even the second edition does not contain a review of cutting-edge technologies. 2020-of the years. On the other hand, having fundamental knowledge from the "Fundamentals of Deep Learning", the reader will be much easier to master and transformers from the materials of the Internet or specialized chapters of other books. For beginners, the second edition of the book 2022 A year might be useful.

#AI #ML #Data #Engineering #Software