This is "a modern fully open-source textbook exploring why and how deep neural networks learn compact and information-dense representations of high-dimensional real-world data." It's a daunting read - and to be honest, I gave the first chapter a careful read but have only skimmed the rest (though it's now on my virtual bookshelf for future reference). I think people who are serious about understanding how we learn and how machines emulate this should at the very least read that first chapter. It contains what is to my understanding an accurate history of artificial intelligence (including the important early work in computation, cybernetics, and information theory). The aim of the book is to give us a framework that explains how neural networks work to generate intelligence. They authors write, "intelligence is not about collecting all informative data in advance and training a model to memorize their (low-dimensional) distribution. Instead, it requires computational mechanisms that can continuously improve current knowledge of the distribution and acquire new information when needed. A fundamental characteristic of any intelligent being or system - whether an animal, a human, a civilization, or the entire scientific community - is the ability to continuously improve or gain new information or knowledge on its own."
Today: Total: [Direct link] [Share]
