Recently, the State Key Laboratory of Robotics (build upon Shenyang Institute of Automation, Chinese Academy of Sciences) made important progress on the theoretical research of deep learning. The paper Depth selection for deep ReLU nets in feature extraction and generalization (first author: Prof. Zhi Han, second author: Siquan Yu PhD) has just been published on the international top journal IEEE Trans. on Pattern Analysis and Machine Intelligence.

Deep learning is recognized to be capable of discovering deep features for representation learning and pattern recognition without requiring elegant feature engineering techniques by taking advantages of human ingenuity and prior knowledge. Thus it has triggered enormous research activities in machine learning and pattern recognition. One of the most important challenges of deep learning is to figure out relations between a feature and the depth of deep neural networks to reflect the necessity of depth. In this cooperated research, they present the adaptivity of features to depths and vice-verse via showing a depth-parameter trade-off in extracting both single feature and composite features. Based on these results, they prove that implementing the classical empirical risk minimization on deep nets can achieve the optimal generalization performance for numerous learning tasks. The theoretical results are verified by a series of numerical experiments including toy simulations and a real application of earthquake seismic intensity prediction.

For years, the vision group from the State Key Laboratory of Robotics focuses on the researches of robotic vision, artificial intelligence, pattern recognition, etc. The works on deep learning, complex illumination/weather processing, robotic vision representation, low rank matrix/tensor modeling were published on various famous international journals and top conferences, including IEEE Trans. on Image Processing, IEEE Trans. on Neural Networks and Learning Systems, IEEE Trans. on Multimedia,Pattern Recognition，IEEE CVPR, ICCV, AAAI, etc. Related theoretical achievements are applied to accomplishing several major national projects.

This research was partially supported by the National Natural Science Foundation of China, the Youth Innovation Promotion Association of Chinese Academy of Sciences and the Research Grant Council of Hong Kong.