Deep learning methods achieve excellent performance in the fields of computer vision and natural language processing through end-to-end supervised training dependent on large scale labeled datasets. However, the existing methods are often targeted for single modal data, ignoring the inherent structure of the data with the lack of theoretical support. Therefore, the wavelet theory based deep convolution networks, the structured deep learning and the multi-modal deep learning are discussed in this paper to demonstrate the potential methods of the combination of deep learning techniques, wavelet theory and structure prediction, and the viable mechanism for extending to multi-modal data is explored as well.

With the rapid development of internet and other information technologies, the amount of visual data grows explosively. It is an essential research topic for efficient visual data mining in the big data era. In this paper, the technology of binary representation learning and its applications are reviewed. Binary representation shows efficiency in storage, transmission and matching, and it is successfully applied to several visual analysis tasks. The applications on visual search and recognition are summarized, and the future trend of binary representation learning is pointed out. From the aspect of methodology, the image-based hashing and the video-based hashing are illustrated, respectively. From the aspect of applications, the successful applications of binary representation learning on varying tasks are discussed including facial analysis, image categorization, image matching and visual tracking. Finally, the future trend of binary representation learning is introduced.

Based on the compression and recovery of high-dimensional data, the development process from the theory of Shannon sampling to sparse representation and compression perception and then to low-rank matrix problem is described. Then, the importance of low rank matrix relaxation and optimization problem is discussed. Subsequently, a detailed review of the existing methods is introduced from three aspects of low rank matrix minimization, decomposition, optimization and applications. Finally, some reasonable suggestions on the deficiencies of current research and the future research direction are put forward.

Face sketch synthesis refers to generating a sketch from an input face photo with some face sketch-photo pairs as the training set. An experimental study on the existing methods is nontrivial. The comprehensive review and the comparative study on representative face sketch synthesis methods are conducted. These methods are grouped into two categories: data-driven methods, also known as exemplar-based methods, and model-driven methods. Generally, data-driven face sketch synthesis consists of three sub-categories: subspace learning-based methods, sparse representation-based methods and Bayesian inference-based methods. Model-driven methods explicitly learn the mapping from face photos to face sketches. Some previously unknown conclusions are drawn as well.

The wide use of positioning technology makes the mining of the people movements easy and plenty of trajectory data are recorded. How to efficiently handle these data for location prediction is a popular research topic as it is fundamental to location-based services(LBS). The existing methods focus either on long time (days or months) visit prediction(i.e. point of interest recommendation) or on real time location prediction(i.e. trajectory prediction). In this paper, the location prediction problem in weak real time conditions is discussed to predict users′ movement in next minutes or hours. A spatial-temporal long-short term memory model(ST-LSTM) combining spatial-temporal influence into LSTM model naturally is proposed to mitigate the data sparse problem. Furthermore, following the idea of generative adversarial network(GAN) for seq2seq learning, the ST-GAN model is proposed, and it takes the proposed ST-LSTM as the generator and the proposed spatial-temporal convolutional neural network(ST-CNN) as the discriminator. The minimax game of ST-GAN can produce more real enough data to train a better prediction model. The proposed ST-GAN is evaluated on a real world trajectory dataset and the results demonstrate the effectiveness of the proposed model.

Online object tracking is a fundamental problem in computer vision and it is crucial to application in numerous fileds such as guided missile, video surveillance and unmanned aerial vehicle. Despite many studies on visual tracking, there are still many challenges during the tracking process including illumination variation, rotation, scale change, deformation, occlusion and camera motion. To make a clear understanding of visual tracking, visual tracking algorithms are summarized in this paper. Firstly, the meaning and the related work are briefly introduced. Secondly, the typical algorithms are classified, summarized and analyzed from two aspects: traditional algorithms and deep learning algorithms. Finally, the problems and the prediction of the future of visual tracking are discussed.

With the rapid informationization of education, extensive data records from online education of students are accumulated, and it provides a good opportunity for both data-driven educational assessment and intelligent tutoring. However, existing models are hard to accurately analyze the characteristics of questions and the academic levels of students from the massive and sparse data with high noise. Meanwhile, it is difficult for these models to satisfy the personalized needs of students and teachers. In this paper, educational data mining studies on these problems are summarized. To improve the student academic level, these studies focus on modeling three objects in education (i.e., questions, students and teachers) and apply effective techniques, such as personalized recommendation methods, combined with the domain knowledge from education. Specifically, a question text embedding framework is presented for question analysis and question retrieval. Then, personalized recommendation methods on learning resources are illustrated based on the cognitive diagnosis of students. Moreover, the way of providing effective guidance and suggestions for teachers is showed. Some of these research achievements are applied to the online educational system “ZHIXUE” in iFlyTek. Finally, the possible research directions in the future are discussed.

Communication technology develops greatly, especially in source and channel coding, modulation mechanisms and ultra-wideband communications. From the aspect of signal processing, current technologies have already approached the Shannon capacity. Directions for future researches in traditional data and signal transmission-based communication industry become unclear. According to the recent state-of-the-art artificial intelligence technologies and their influences on the revolution of the communication industry, a new communication mechanism, namely semantic communications, is proposed. Compared with the traditional communication technology based on pattern transmission, the key point of semantic communications is idea-passing communication, and it can also be referred to as content transmission. In such a revolutionary communication mechanism, the transmission is conducted upon ideas instead of data through the construction of a certain knowledge library. The error tolerance of the channel can also be improved via the matching between the idea transmitter and receiver. It could be considered as the artificial intelligence induced communication with its true meaning since it is a brain-resembling communication mechanism. The introduced brain-resembling communication mechanism based on idea-passing significantly reduces the data amount to be transmitted, and it is an efficient way to address the challenge to the communication industry due to the arrival of big data era. In this paper, the tentative idea of semantic communications is presented and the fundamental elements, semantic coding and decoding, future research directions and major challenges in semantic communications are discussed.