Probabilistic Deep Learning: With Python, Keras and TensorFlow Probability

All the explanations are very handable and in a friendly language. I don't think there is any issue that is not so well explained

I am so thankful to find this book. The authors did a great job explaining complex concepts from an unique, streamlined perspective. It’s interesting to see how one can set up maximum likelihood with non-constant variance using neural networks. Bayesian neural network layers can be introduced via flipout or dropout layers, for which the authors provide basic intuition and explanation of benefits. Another intriguing concept was normalizing flow, which appears to allow for modeling any complex multidimensional distributions using mathematical tricks and neural networks( while it is still unclear to me how to assess the quality of the models). Nonetheless, normalizing flow may have various application potentials other than image processing and manipulation.

As another reviewer points out, the first 40% is basic (and light) introduction to DL and Tensorflow. The second half is more useful, though too superficial. This is at best an initial introduction.What stands out is the terrible editing -- almost every page has some error. And I haven't yet found an online errata. There are also amusing, and uncaught, translation errors (the authors are German), such as "sinus" for "sine".I guess at this point the computer programming book industry is just a mill, and editors have been deleted.