[ Prime ] Generative Deep Learning: Teaching Machines to Paint, Write, Compose, and Play Author David Foster – Andy-palmer.co.uk
Generative Modeling Is One Of The Hottest Topics In AI It S Now Possible To Teach A Machine To Excel At Human Endeavors Such As Painting, Writing, And Composing Music With This Practical Book, Machine Learning Engineers And Data Scientists Will Discover How To Re Create Some Of The Most Impressive Examples Of Generative Deep Learning Models, Such As Variational Autoencoders,generative Adversarial Networks GANs , Encoder Decoder Models And World Models Author David Foster Demonstrates The Inner Workings Of Each Technique, Starting With The Basics Of Deep Learning Before Advancing To Some Of The Most Cutting Edge Algorithms In The Field Through Tips And Tricks, You Ll Understand How To Make Your Models Learn Efficiently And Become Creative Discover How Variational Autoencoders Can Change Facial Expressions In Photos Build Practical GAN Examples From Scratch, Including CycleGAN For Style Transfer And MuseGAN For Music Generation Create Recurrent Generative Models For Text Generation And Learn How To Improve The Models Using Attention Understand How Generative Models Can Help Agents To Accomplish Tasks Within A Reinforcement Learning Setting Explore The Architecture Of The Transformer BERT, GPT And Image Generation Models Such As ProGAN And StyleGAN I have only read the first two chapters got it only yesterday but find it superb Chapter 2 sums up my own slow progress for the last 10 months on my own, where I ve gradually settled on TensorFlow and Keras, and the model described in the book, by observing their use in random papers and Stackoverflow discussions I can hardly imagine a concise, yet sufficiently informative, description of what I have found out over these past months I wish I d seen this book earlier If the rest of the book, describing material I don t know, is as good as the part I do know, then I m in for a real treat. Very entertaining reading Print quality could be better, but that s on , not on the author I guess Wow I had a lot of fun reading this book and now I am working with that I am an expert in AI and a creative person Truth be told, in the 1990s I began my IT career because I was fascinated by what programming can do when it comes to art.Now I have David Foster s book here and it told me that Deep Neural Networks are perfectly capable of generating artificial content by training on the real ones A good example is definitely generating music based on what a Neural Network learned from works of Johann Sebastian Bach You just come up with a fitting network architecture and train it.The good thing is The book is not restricted to just music Music is just one use case Natural Language Processing for generating texts, GANs and Autoencoders for Images, you name it.And the best thing I have to admit that the content of the book is complete and a good state of the art There is no dust on the pages Everything you learn is relevant today and will be for a while. The book starts great Fantastic examples It appeals to the reader s intuition and imagination I loved the beginning and it was very easy working side by side with Jupyter Notebook The examples are easy to follow and the code is pure Python with Keras At that point I was going to give the book five stars However, I was stuck at Autoencoders when the author suddenly started using his own code shortcuts, which was completely unexpected It took me a while to figure it out that the code was no longer Keras but the functions and objects developed by the author, and imported from the local python files The book does not explain any of this and the code becomes very obscure The author s models and utilities were clearly meant to simplify development of complex neural networks by the reader Unfortunately, the code is no longer intelligible as it hides the true Keras APIs These shortcuts are not really necessary and the code they replace would not add much to the size of the book In those circumstances, if you move away from the book and the author s Github repository, you will no longer be able to reproduce the models and their tests easily While it is expected of any practitioner to develop his or her own helper library, this is not suitable for the book which needs simplicity and clarity In all honesty, the book does not claim to train the reader in Keras at all, however, it uses Keras and asks the reader to install the software, and then explains the basics of model creation with Keras, only to leave it behind I d recommend to replace all obscure code with the simplest model creation, which can be found in any Keras example on the web As the author is quite responsive to the reviews and open for comments, I have increased my rating. Sehr hilfreich Arbeite mich langsam durch I m a ML engineer I know the ropes I ve spent hours surfing the web for explanations and sample implementations of deep learning and GAN methodology You might have too You might be wondering, why would I need this book when I have access to Medium And that s a good question I can answer that for you.If you ve been around for a while, you might know that Medium content can be repetitive, unoriginal, full of too much filler, full of code that can t be re used, or in the habit of explaining mathematics probability poorly The difference between this book and Medium is that this book doesn t do that It s got good introductions to each popular dataset, contains useful code, is highly readable and refreshing, and uses equations sparingly and effectively, without dumbing down the content too much I can skim the content easily It s top notch, contains a lot of fresh developments like World Models Ch 8 , and seems like an essential book for my ML library.My only critique is that at times, the book seemed to read like a children s bedtime story Telling stories is an excellent way to explain concepts, but I don t know if I need David Foster renaming GAN discriminators and generators Di and Gene , with an introduction on taking photographs of ganimals to understand what a GAN is That s not to say this book does the story bit poorly If you re new to all this, it may be really useful for you It was just a bit off putting for me.Regardless, I highly recommend it to anyone that has familiarity with basic stats notation and is comfortable with Python3. Paper is super thin and images seem to have printed by a cheap printer in economy mode.I m not usually picky about these things but it is very noticeable.Content wise is OK, I especially liked the explanation about mode collapse in GANs and justification of WGAN losses and why they work There are many analogies to explain concepts VAEs, GANs, etc and but I find some of them could be better. Bien These generative models are both simple and hard It is a simple idea with a lot of potential and possibility But it is very hard to make it work The book is well written and explain the basic idea of generative framework very well I have learned these models previously from other sources This book offers a great survey to the field and a way for me to refresh my memory.Only thing I wish there is details in the difficulty of training GAN and the explore the math behind it.