Key Laboratory of Intelligent Information Processing of Jilin Universities

Changchun, China

Key Laboratory of Intelligent Information Processing of Jilin Universities

Changchun, China
SEARCH FILTERS
Time filter
Source Type

Sun T.,Northeast Normal University | Sun T.,Key Laboratory of Intelligent Information Processing of Jilin Universities | Peng Y.,Northeast Normal University | Yang F.,Northeast Normal University | And 5 more authors.
ICIC Express Letters | Year: 2017

Traditional text classifiers often rely on many human-designed features, which is complicated and inconvenient in practical applications. This paper proposes a novel neural network based text classification method that uses a self-adaptive mechanism to integrate various pre-trained word embedding. Firstly, the proposed method independently constructs feature vector from various word embedding with rich semantic information. Then, in order to distinguish the contribution of multiple word embedding in classification task and revise the influence of different embedding on the final text representation, we use Highway network to adjust the weight of feature vectors automatically. Finally, we combine these feature vectors and form the final text representation to classify. The experiment has been conducted on sentiment and news classification datasets; the results show that the method achieves better results with several benchmarks, and gets the similar performance to the state-of-the-art methods on English news classification task. The further contrast experiment also proves that the method has the ability of adaptively integrating multiple word embedding. © 2017.


Sun H.-G.,Key Laboratory of Intelligent Information Processing of Jilin Universities | Sun H.-G.,Southern Polytechnic State University | Yang X.-Q.,Northeast Normal University | Sun T.-L.,Northeast Normal University | Hung C.-C.,Southern Polytechnic State University
ICIC Express Letters | Year: 2015

In this paper, we propose an adaptive haze removal method based on dark channel prior. The dark channel value is compensated for estimating the transmission; the atmospheric light will be re-estimated through the analysis of the characteristics of images. Images are usually degraded with haze due to the object absorption and atmospheric scattering. The experiments show that the images recovered by the proposed method are better than other existing methods and recovered images are closer to real images. Meanwhile, the Mean Square Error (MSE) is used to verify the proposed approach for the images recovered which are taken from the FRIDA database. Through the comparison of Inverse Discrete Cosine Transforms (IDCT) of haze images and haze-free images, the statistics of the comparison shows that the haze images consume more computation time. To improve the time complexity, the Graphics Processing Unit (GPU) was used to increase the computing efficiency. © 2015, ICIC International.


Sun T.,Northeast Normal University | Sun T.,Key Laboratory of Intelligent Information Processing of Jilin Universities | Guan W.,PLA Air Force Aviation University | Yang F.,Northeast Normal University | And 7 more authors.
ICIC Express Letters | Year: 2016

Subjective statement screening is a fundamental part of sentiment analysis of product reviews. Its aim is to extract the subjective texts from product reviews and filter objective texts that do not contain any sentimental information. The existing screening methods are mainly based on only one feature. However, single feature is insufficient to mark all subjective or objective statements. We propose a feature combination method based on the N-pos feature, the word feature and the syntactic dependency feature for subjective statement screening. The experiments on the real datasets show that the proposed feature combination method significantly improves the accuracy of subjective statement screening and is really feasible and effective. © 2016, ICIC Express Letters Office. All rights reserved.


Li Y.W.,Northeast Normal University | Li Y.W.,Key Laboratory of Intelligent Information Processing of Jilin Universities | Gao W.Z.,Audi AG | Zhang H.J.,Northeast Normal University | And 5 more authors.
Applied Mechanics and Materials | Year: 2014

To avoid injury with the coexistence of human and mobile robot in the circumstance, a vision-based system for mobile robot/human localization is designed and implemented. The details of the system are described in this paper. © (2014) Trans Tech Publications, Switzerland.


Sun H.,Northeast Normal University | Sun H.,Key Laboratory of Intelligent Information Processing of Jilin Universities | Yang Y.,Northeast Normal University | Sun T.,Northeast Normal University | And 5 more authors.
ICIC Express Letters | Year: 2016

Sentiment analysis is an important task in the field of natural language processing. This paper has focus on opinion target extraction of sentiment extraction in the primary task of sentiment analysis. First, the features are extracted from preprocessed results. Then, the results of feature extraction with the corresponding template will be input to the CRFs (Conditional Random Fields) module for training and recognition. In the process of feature extraction, three features are applied on the basis of existing features by analyzing syntactic structure. To select template with better performance, many comparison experiments have carried on a variety of templates with different sizes. Experimental results show our method is more effective than the baseline which can exactly extract opinion targets. © 2016 ISSN.


Zhang H.,Northeast Normal University | Zhang H.,Key Laboratory of Intelligent Information Processing of Jilin Universities | Ma Y.,Northeast Normal University | Ma Y.,Key Laboratory of Intelligent Information Processing of Jilin Universities | And 8 more authors.
Mathematical Problems in Engineering | Year: 2013

Multiresolution hierarchy based on features (FMRH) has been applied in the field of terrain modeling and obtained significant results in real engineering. However, it is difficult to schedule multiresolution data in FMRH from external memory. This paper proposed new multiscale feature model and related strategies to cluster spatial data blocks and solve the scheduling problems of FMRH using spatial neighborhood. In the model, the nodes with similar error in the different layers should be in one cluster. On this basis, a space index algorithm for each cluster guided by Hilbert curve is proposed. It ensures that multi-resolution terrain data can be loaded without traversing the whole FMRH; therefore, the efficiency of data scheduling is improved. Moreover, a spatial closeness theorem of cluster is put forward and is also proved. It guarantees that the union of data blocks composites a whole terrain without any data loss. Finally, experiments have been carried out on many different large scale data sets, and the results demonstrate that the schedule time is shortened and the efficiency of I/O operation is apparently improved, which is important in real engineering. © 2013 Huijie Zhang et al.


Zhang H.,Northeast Normal University | Zhang H.,Key Laboratory of Intelligent Information Processing of Jilin Universities | He X.,Northeast Normal University | He X.,Key Laboratory of Intelligent Information Processing of Jilin Universities | And 6 more authors.
Yi Qi Yi Biao Xue Bao/Chinese Journal of Scientific Instrument | Year: 2014

In order to meet the requirement of terrain dynamic modeling, a real-time modeling method for variable terrain is proposed. An adaptive multi-resolution model based on the connotative hierarchical structure is established, combined with physical-crater model. According to the different explosion position, the real-time terrain deformation effects can be rendered. To keep the details of physical-crater, the grid model with full resolution is adopted to build the carter model. According to the different incident direction and terrain features, an algorithm about deformation region extension is presented. Furthermore, the minimum projection area of crater is determined, and then the active points on the connotative hierarchical structure are defined. After changing the state of active points, the seamless fusion between the multi-resolution terrain and the crater model is realized. The above methods are applied to many benchmark data with different landform. The result, the deformation terrain model can realize space continuity. The dynamic terrain modeling approach meets the requirements of virtual environment simulation. ©, 2014, Science Press. All right reserved.

Loading Key Laboratory of Intelligent Information Processing of Jilin Universities collaborators
Loading Key Laboratory of Intelligent Information Processing of Jilin Universities collaborators