Deep Learning, by Yann L., Yoshua B. & Geoffrey H. (2015) 引用次数：5716
Deep learning enables computational models that are composed of multiple processing layers to learn with multiple levels of abstraction, the representations of data. These methods have resulted in the improvement of the state-of-the-art in object detection, speech recognition, visual object recognition, and many other domains such as drug discovery and genomics dramatically.
TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems, by Martín A., Ashish A. B., Eugene B. C., et al. (2015)引用次数：2423
The system is flexible and can be used to express a variety of algorithms, that includes deep neural network models training as well as inference algorithms, and it has been used for conducting research and for the deployment of machine learning systems into production across more than a dozen areas of computer science and other fields, including the retrieval of information, speech recognition, robotics, computer vision, geographic information extraction, natural language processing, and computational drug discovery.
TensorFlow: a system for large-scale machine learning, by Martín A., Paul B., Jianmin C., Zhifeng C., Andy D. et al. (2016) 引用次数：2227
TensorFlow, an open-source project with its main focus on training and inference on deep neural networks. supports a variety of applications. Many services of Google in production make the use of TensorFlow and over time it has become widely used for research in the field of machine learning.
Deep learning in neural networks, by Juergen Schmidhuber (2015) 引用次数：2196
This historical survey has a compact summarization of relevant work, much of it from the previous millennium. Shallow as well as deep learners by the depth of their credit assignment paths are distinguished which are chains of possibly learnable, causal links between actions and effects.
Human-level control through deep reinforcement learning, by Volodymyr M., Koray K., David S., Andrei A. R., Joel V et al (2015)引用次数：2086
Here in order to develop a novel artificial agent, termed a deep Q-network, we make the use of recent advances in training deep neural networks that using end-to-end reinforcement learning can learn successful policies directly from high-dimensional sensory inputs. This agent was tested on the challenging domain of classic Atari 2600 games.
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks, by Shaoqing R., Kaiming H., Ross B. G. & Jian S. (2015) 引用次数：1421
In this work, you are introduced to a Region Proposal Network (RPN) that shares with the detection network, full-image Convolutional features, thus enabling nearly cost-free region proposals. A Region Proposal Network is a fully Convolutional network that at each position simultaneously predicts object bounds and objectness scores.
Long-term recurrent convolutional networks for visual recognition and description, by Jeff D., Lisa Anne H., Sergio G., Marcus R., Subhashini V. et al. (2015) 引用次数：1285
In contrast to current models which assume a fixed spatio-temporal receptive field or simple temporal averaging for sequential processing, recurrent Convolutional models are “doubly deep” in that they can be compositional in spatial and temporal “layers”.
MatConvNet: Convolutional Neural Networks for MATLAB, by Andrea Vedaldi & Karel Lenc (2015) 引用次数：1148
It as easy-to-use MATLAB functions exposes the building blocks of CNN's, providing routines for computing linear convolutions with filter banks, feature pooling, and many more. This document provides a great overview of Convolutional Neural Networks and how they have their implementation in MatConvNet and further also gives in the toolbox of each computational block the technical details of the same.
Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks, by Alec R., Luke M. & Soumith C. (2015) 引用次数：1054
In this work, the main focus is to help bridge the gap between the success of CNN's for supervised learning and unsupervised learning. Here, you are introduced to a class of CNN's called deep convolutional generative adversarial networks (DCGANs), that have certain architectural constraints, and demonstrate that they are a strong candidate for unsupervised learning.
U-Net: Convolutional Networks for Biomedical Image Segmentation, by Olaf R., Philipp F. &Thomas B. (2015) 引用次数：975
There is large consent that successful training of deep networks has a requirement of many annotated training samples. In this paper, you are presented a strategy in network and training that in order to more efficiently use the available annotated samples solely relies on the strong use of data augmentation.