Architecture, Generative Model, and Deep Reinforcement Learning for IoT Applications: Deep Learning Perspective

Abstract

AbstractDeep learning is a subclass of machine learning. In the last few years, this has come to prominence with the core availability of GPUs for computing. There are many applications in which deep learning is using such as a self-driving car in industry that generated caption for images automatically. The deep learning algorithm consists of artificial neural network (ANN) with a number of hidden layers. It is used for nonlinear with supervised and unsupervised learning. Deep learning models provide better results and improvement as compared to traditional machine learning in many areas such as language translation, speech recognition, stock market, healthcare, etc. Today’s most of the deep learning frameworks are open source and freely available for use. Deep learning frameworks provide building blocks for validation of deep learning neural networks, training, and designing with the interface of high-level programming. Some of the most used frameworks are MXNet, PyTorch, TensorFlow, etc.On another side, the Internet of Things (or Internet Connected Things) has attracted many applications and sectors for producing results efficiently, and these devices have made people’s life longer to live. These devices are generating a lot of data (called big data) which can be used for further prediction/forecast by modern deep learning models. In deep learning, complex problems can be solved with the help of modern tools. Differently, each framework is built for a different purpose. These frameworks are often evolving and getting better very rapidly. There are many criteria to choose a framework. In the chapter, we will discuss different types of framework, pros and cons of every framework, architecture, and different criteria to choose the better framework which will be useful for Internet of Things-based applications.KeywordsInternet of things-based applicationsDeep learningMachine learningDeep learning frameworkArtificial neural network