A state-of-the-art survey on deep learning theory and architectures.
Check out our new video:
Deep learning (also known as deep structured learning, hierarchical learning or deep machine learning) is a branch of machine learning based on a set of algorithms that attempt to model high-level abstractions in data. In a simple case, data might be an array of pixel values that describe an image. The goal of deep learning is to create models that learn to recognize complex patterns in data and can make predictions about new data.
Deep learning models are distinguished from other machine learning techniques by their ability to automatically extract features from raw data. For example, a deep learning model might be trained on a dataset of images that contains millions of pixels. The model would learn to identify features such as edges, corners, and textures; it would then learn how these features can be combined to form more complex patterns such as object shapes.
The term “deep” refers to the number of layers in the network—the more layers, the deeper the network. Deep learning networks are also often much larger than traditional machine learning models; they may have millions or billions of parameters.
The first deep learning networks were created in the 1980s, but they were not widely used until recent years due to the limited computational power available at that time. With the advent of powerful GPUs and new training techniques, deep learning has become very successful at tasks such as image classification, object detection, and speech recognition
What is Deep Learning?
Deep learning is a subfield of machine learning that is concerned with algorithms inspired by the structure and function of the brain called artificial neural networks. It has been used in image recognition, natural language processing, and robotics.
Theoretical Foundations of Deep Learning
Deep learning is a rapidly developing area of machine learning that is attracting a great deal of attention from both the research community and industry. Despite its recent popularity, deep learning still lacks a comprehensive theoretical foundation. In this paper, we provide a state-of-the-art survey of the theoretical foundations of deep learning. We begin by review the fundamental supervised learning problem and the generalization properties of machine learning algorithms. We then present the Minnesota gradient descent algorithm, which is the theoretical foundation of modern deep learning. We also discuss recent advances in unsupervised learning, reinforcement learning, and self-supervised learning. Finally, we survey some of the most popular deep learning architectures, including convolutional neural networks, recurrent neural networks, and autoencoders.
Deep Learning Architectures
The goal of this survey is to provide a comprehensive overview of Deep Learning (DL) from a computational perspective. We will first briefly introduce some key concepts in DL, before moving on to review the major DL architectures that have been proposed in the literature. We will also discuss recent advances in DL theory, which are important for understanding the computational properties of these architectures. Finally, we will conclude with some open problems and future directions for research.
Training Deep Neural Networks
Deep neural networks have shown immense success in a variety of tasks in recent years. Achieving this success requires training of very deep models with millions or even billions of parameters. However, train- ing such large models is notoriously difficult and often requires careful design choices and data preprocessing schemes that are painstakingly selected for each task. In this paper, we present a state-of-the-art survey on the theory and architectures for training deep neural networks. We first present the common principles for designing successful training algorithms, including a detailed discussion of gradient-based methods, normalization, initialization schemes, and regularization techniques. We then describe several scalable optimization methods that have been developed to tackle the problem of training extremely large models. Finally, we review recent advances in architectures andBlack- box methods that have enabled the success of deep learning.
Applications of Deep Learning
Deep learning has been applied to a wide range of tasks in different fields, such as computer vision, natural language processing, speech recognition, and so on. In this section, we will review some of the most representative applications of deep learning.
The survey provides an overview of the current state-of-the-art in deep learning theory and architectures. In particular, we focus on the representative types of deep neural networks (DNNs), their key components, and how they are interconnected. We also present some popular applications of DNNs and open issues for future research.
There are many excellent surveys and review articles on deep learning. We list a few of them here.
Deep Learning: ACritical Appraisal, by Yoshua Bengio et al., presents a comprehensive overview of deep learning from both a biological and computational perspective.
A Comprehensive Survey on Graph Neural Networks, by Zonghan Wu et al., surveys the field of graph neural networks, including various architectures and applications.
Deep Learning: An Introduction, by Geoffrey Hinton et al., is a concise introduction to deep learning, covering both the theoretical foundations and practical applications.
 I. Goodfellow, Y. Bengio, and A. Courville, “Deep Learning,” 2016.
 K. Simonyan and A. Zisserman, “Very Deep Convolutional Networks for Large-Scale Image Recognition,” in Proceedings of the 3rd International Conference on Learning Representations (ICLR), 2015.
 J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “ImageNet: A Large-Scale Hierarchical Image Database,” in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2009, pp. 248-255.
 O. Russakovsky, J. Deng, H. Su, J. Krause, S.-E. Bernstein, A.-Y. Ng, and L Fei-Fei , “ImageNet Large Scale Visual Recognition Challenge,” IJCV (), vol 115(3), pp 211-252, 2015
 NIST/SEMATECH e-Handbook of Statistical Methods,” NIST/SEMATECH e-Handbook of Statistical Methods”, https://www2.censusbystateofdelawarecountymedianhouseseholdsizesandaverages2006201020112012201320145 year estimatesalldelawarecounties-, Visit Date:7/15/2018
About the Author
Deep learning is a rapidly evolving area of machine learning that is enjoying a renaissance thanks to powerful new drugs and efficient training methods. In this state-of-the-art survey, we provide an overview of deep learning theory and architectures. We begin with a review of the linear algebra and probability theory that underlies deep learning, followed by a description of common architectures such as feedforward neural networks, convolutional neural networks, recurrent neural networks, and autoencoders. We then discuss recent advances in deep learning including generative models, deep reinforcement learning, and transfer learning. Finally, we survey a variety of applications where deep learning has been shown to be effective including computer vision, natural language processing, speech recognition, bioinformatics, and robotic control.
Keyword: A State-of-the-Art Survey on Deep Learning Theory and Architectures