دانلود کتاب یادگیری عمیق تکاملی

Evolutionary deep learning, first described in this book, is a general categorization and grouping of a set of techniques that combine evolutionary methods with deep learning. The purpose of using these methods is to optimize a DL system from the collection of data to validation. EDL is not new; combining evolutionary methods with DL has gone by many cool names like Deep Neural Evolution, Evolutionary Neural AutoML, Neuroevolution, Evolutionary AI, and more.

EDL is the merger of 2 unique and often considered separate sub-fields of artificial intelligence. The combination of evolutionary computation with the application of deep learning for the purposes of automating and improving models. EC itself is a family of methods by which biological or natural processes are simulated to solve complex problems. That, in turn, can be applied on top of DL to automate and optimize solutions but has the potential to uncover new strategies and architectures.

The broad category of methods we will encompass under EDL is by no means new and itself has been around for 20 years. While much of that research has shown to be successful in auto-tuning DL models it has received secondary attention behind the AI hype of more cutting-edge hand-crafted examples. Where in many papers the authors discuss the extensive time taken to data/feature engineer and hyperparameter tune an innovative model.

However, for many now embracing DL the challenge of building robust and high-performance models is daunting and inflicted with many challenges. Challenges that require advanced and sophisticated knowledge of all the options and quirks of your chosen DL framework to understand when your model may just be incorrectly fitting. EDL as an AutoML solution is presented here to address most of the issues experienced or novice practitioners will face.

The purpose of EDL is currently to provide a better mechanism and toolset to provide automated machine learning (AutoML) to build DL solutions. Evolutionary methods are an excellent and relatively easy mechanism to provide a broad set of optimization tools that can be applied to DL. While there is potential that evolutionary techniques could automate the construction of more advanced AI, that is not the current intent of EDL or this book.

Instead, we will focus on building better-optimized networks using evolutionary techniques. Before we do that though, we will cover in-depth the operations and use of a wide variety of EC and EA to better get aquatinted with the basic concepts. Starting with a brief introduction to evolution and evolutionary processes in the next section.