-
Record 1 of
Title:Towards convolutional neural networks compression via global error reconstruction
Author(s):Lin, Shaohui(1,2); Ji, Rongrong(1,2); Guo, Xiaowei(3); Li, Xuelong(4)
Source: IJCAI International Joint Conference on Artificial Intelligence Volume: 2016-January Issue: DOI: Published: 2016
Abstract:In recent years, convolutional neural networks (CNNs) have achieved remarkable success in various applications such as image classification, object detection, object parsing and face alignment. Such CNN models are extremely powerful to deal with massive amounts of training data by using millions and billions of parameters. However, these models are typically deficient due to the heavy cost in model storage, which prohibits their usage on resource-limited applications like mobile or embedded devices. In this paper, we target at compressing CNN models to an extreme without significantly losing their discriminability. Our main idea is to explicitly model the output reconstruction error between the original and compressed CNNs, which error is minimized to pursuit a satisfactory rate-distortion after compression. In particular, a global error reconstruction method termed GER is presented, which firstly leverages an SVD-based low-rank approximation to coarsely compress the parameters in the fully connected layers in a layerwise manner. Subsequently, such layer-wise initial compressions are jointly optimized in a global perspective via back-propagation. The proposed GER method is evaluated on the ILSVRC2012 image classification benchmark, with implementations on two widely-adopted convolutional neural networks, i.e., the AlexNet and VGGNet-19. Comparing to several state-of-the-art and alternative methods of CNN compression, the proposed scheme has demonstrated the best rate-distortion performance on both networks.
Accession Number: 20165103146967
-
Record 2 of
Title:New -1-norm relaxations and optimizations for graph clustering
Author(s):Nie, Feiping(1); Wang, Hua(2); Deng, Cheng(3); Gao, Xinbo(3); Li, Xuelong(4); Huang, Heng(1)
Source: 30th AAAI Conference on Artificial Intelligence, AAAI 2016 Volume: Issue: DOI: Published: 2016
Abstract:In recent data mining research, the graph clustering methods, such as normalized cut and ratio cut, have been well studied and applied to solve many unsupervised learning applications. The original graph clustering methods are NP-hard problems. Traditional approaches used spectral relaxation to solve the graph clustering problems. The main disadvantage of these approaches is that the obtained spectral solutions could severely deviate from the true solution. To solve this problem, in this paper, we propose a new relaxation mechanism for graph clustering methods. Instead of minimizing the squared distances of clustering results, we use the 1-norm distance. More important, considering the normalized consistency, we also use the 1- norm for the normalized terms in the new graph clustering relaxations. Due to the sparse result from the 1-norm minimization, the solutions of our new relaxed graph clustering methods get discrete values with many zeros, which are close to the ideal solutions. Our new objectives are difficult to be optimized, because the minimization problem involves the ratio of nonsmooth terms. The existing sparse learning optimization algorithms cannot be applied to solve this problem. In this paper, we propose a new optimization algorithm to solve this difficult non-smooth ratio minimization problem. The extensive experiments have been performed on three two-way clustering and eight multi-way clustering benchmark data sets. All empirical results show that our new relaxation methods consistently enhance the normalized cut and ratio cut clustering results. ? Copyright 2016, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved.
Accession Number: 20165203195650
-
Record 3 of
Title:Pedestrian detection inspired by appearance constancy and shape symmetry
Author(s):Cao, Jiale(1); Pang, Yanwei(1); Li, Xuelong(2)
Source: Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition Volume: 2016-December Issue: DOI: 10.1109/CVPR.2016.147 Published: December 9, 2016
Abstract:The discrimination and simplicity of features are very important for effective and efficient pedestrian detection. However, most state-of-the-art methods are unable to achieve good tradeoff between accuracy and efficiency. Inspired by some simple inherent attributes of pedestrians (i.e., appearance constancy and shape symmetry), we propose two new types of non-neighboring features (NNF): side-inner difference features (SIDF) and symmetrical similarity features (SSF). SIDF can characterize the difference between the background and pedestrian and the difference between the pedestrian contour and its inner part. SSF can capture the symmetrical similarity of pedestrian shape. However, it's difficult for neighboring features to have such above characterization abilities. Finally, we propose to combine both non-neighboring and neighboring features for pedestrian detection. It's found that nonneighboring features can further decrease the average miss rate by 4.44%. Experimental results on INRIA and Caltech pedestrian datasets demonstrate the effectiveness and efficiency of the proposed method. Compared to the state-of the-art methods without using CNN, our method achieves the best detection performance on Caltech, outperforming the second best method (i.e., Checkerboards) by 1.63%. ? 2016 IEEE.
Accession Number: 20170403274876
-
Record 4 of
Title:Design of infrared signal processing system based on ZYNQ platform
Author(s):Bai, Zhuoyu(1,2); Leng, Haibing(1); Hu, Bingliang(1); Wang, Shuang(1)
Source: Proceedings of SPIE - The International Society for Optical Engineering Volume: 10157 Issue: DOI: 10.1117/12.2246949 Published: 2016
Abstract:A newly developed real-time infrared signal processing system based on the heterogeneous multi-processor system on chip (MPSoC) is proposed in this paper. The architecture, hardware configuration, image pre-processing algorithms used in the system and the experimental result are presented. Compared to the infrared signal processing system in being, Xilinx Zynq-7000 All Programmable SoC has been used in the proposed system which is more portable, integrated, and has excellent performance during its signal processing. ? 2016 SPIE.
Accession Number: 20170503310138
-
Record 5 of
Title:Video parsing via spatiotemporally analysis with images
Author(s):Li, Xuelong(1); Mou, Lichao(1); Lu, Xiaoqiang(1)
Source: Multimedia Tools and Applications Volume: 75 Issue: 19 DOI: 10.1007/s11042-015-2735-x Published: October 1, 2016
Abstract:Effective parsing of video through the spatial and temporal domains is vital to many computer vision problems because it is helpful to automatically label objects in video instead of manual fashion, which is tedious. Some literatures propose to parse the semantic information on individual 2D images or individual video frames, however, these approaches only take use of the spatial information, ignore the temporal continuity information and fail to consider the relevance of frames. On the other hand, some approaches which only consider the spatial information attempt to propagate labels in the temporal domain for parsing the semantic information of the whole video, yet the non-injective and non-surjective natures can cause the black hole effect. In this paper, inspirited by some annotated image datasets (e.g., Stanford Background Dataset, LabelMe, and SIFT-FLOW), we propose to transfer or propagate such labels from images to videos. The proposed approach consists of three main stages: I) the posterior category probability density function (PDF) is learned by an algorithm which combines frame relevance and label propagation from images. II) the prior contextual constraint PDF on the map of pixel categories through whole video is learned by the Markov Random Fields (MRF). III) finally, based on both learned PDFs, the final parsing results are yielded up to the maximum a posterior (MAP) process which is computed via a very efficient graph-cut based integer optimization algorithm. The experiments show that the black hole effect can be effectively handled by the proposed approach. ? 2015, Springer Science+Business Media New York.
Accession Number: 20152801019554
-
Record 6 of
Title:Preparation method of Ce1?xZrxO2/tourmaline nanocomposite with high far-infrared emissivity and its mechanism
Author(s):Guo, Bin(1,2); Yang, Liqing(1); Li, Wenlong(1,2); Wang, Haojing(1); Zhang, Hong(1)
Source: Applied Physics A: Materials Science and Processing Volume: 122 Issue: 2 DOI: 10.1007/s00339-015-9586-1 Published: February 1, 2016
Abstract:Far-infrared functional nanocomposites were prepared by the coprecipitation method using natural tourmaline (XY3Z6Si6O18(BO3)3V3W, where X is Na+, Ca2+, K+, or vacancy; Y is Mg2+, Fe2+, Mn2+, Al3+, Fe3+, Mn3+, Cr3+, Li+, or Ti4+; Z is Al3+, Mg2+, Cr3+, or V3+; V is O2?, OH?; and W is O2?, OH?, or F?) powders, ammonium cerium(IV) nitrate and zirconium(IV) nitrate pentahydrate as raw materials. The reference sample tourmaline modified with ammonium cerium(IV) nitrate alone was also prepared by a similar precipitation route. The results of Fourier transform infrared spectroscopy show that Ce–Zr can further enhance the far-infrared emission properties of tourmaline than Ce alone. Through characterization by X-ray diffraction (XRD), transmission electron microscopy (TEM) and X-ray photoelectron spectroscopy (XPS), the mechanism by which Ce(–Zr) acts on the far-infrared emission property of tourmaline was systematically studied. The XPS spectra show that the Fe3+ ratio inside tourmaline powders after heat treatment can be raised by doping Ce and further raised after adding Zr. Moreover, it is showed that Ce3+ is dominant inside the samples, but its dominance is replaced by Ce4+ outside. In addition, XRD results indicate the formation of CeO2 and Ce1?xZrxO2 crystallites during the heat treatment, and further, TEM observations show they exist as nanoparticles on the surface of tourmaline powders. Based on these results, we attribute the improved far-infrared emission properties of Ce–Zr-doped tourmaline to the enhanced unit cell shrinkage of the tourmaline arisen from much more oxidation of Fe2+ (0.074?nm in radius) to Fe3+ (0.064?nm in radius) inside the tourmaline caused by Zr enhancing the redox shift between Ce4+ and Ce3+ via improving the oxygen mobility in the Ce–Zr crystal. ? 2016, Springer-Verlag Berlin Heidelberg.
Accession Number: 20160501873311
-
Record 7 of
Title:Low-penalty up to 16-QAM wavelength conversion in a low loss CMOS compatible spiral waveguide
Author(s):Da Ros, Francesco(1); Porto Da Silva, Edson(1); Zibar, Darko(1); Chu, Sai T.(2); Little, Brent E.(3); Morandotti, Roberto(4); Galili, Michael(1); Moss, David J.(5); Oxenlewe, Leif K.(1)
Source: 2016 Optical Fiber Communications Conference and Exhibition, OFC 2016 Volume: Issue: DOI: 10.1364/ofc.2016.tu2k.5 Published: August 9, 2016
Abstract:Wavelength conversion of 32-Gbaud QPSK and 10-Gbaud 16-QAM is demonstrated using a 50-cm long low loss spiral Hydex-glass waveguide. BER ? 2016 OSA.
Accession Number: 20163702799781
-
Record 8 of
Title:Wavelength conversion of QPSK and 16-QAM coherent signals in a CMOS compatible spiral waveguide
Author(s):Da Ros, Francesco(1); da Silva, Edson Porto(1); Zibar, Darko(1); Chu, Sai T.(2); Little, Brent E.(3); Morandotti, Roberto(4); Galili, Michael(1); Moss, David J.(5); Oxenl?we, Leif K.(1)
Source: Optics InfoBase Conference Papers Volume: Issue: DOI: Published: 2016
Abstract:We characterize a wavelength converter based on a 50-cm long low-loss spiral Hydex waveguide. A 10-nm FWM bandwidth is shown over which low OSNR penalty ( ? OSA 2016.
Accession Number: 20171403515669
-
Record 9 of
Title:Non-negative matrix factorization with sinkhorn distance
Author(s):Qian, Wei(1); Hong, Bin(1); Cai, Deng(1); He, Xiaofei(1); Li, Xuelong(2)
Source: IJCAI International Joint Conference on Artificial Intelligence Volume: 2016-January Issue: DOI: Published: 2016
Abstract:Non-negative Matrix Factorization (NMF) has received considerable attentions in various areas for its psychological and physiological interpretation of naturally occurring data whose representation may be parts-based in the human brain. Despite its good practical performance, one shortcoming of original NMF is that it ignores intrinsic structure of data set. On one hand, samples might be on a manifold and thus one may hope that geometric information can be exploited to improve NMF's performance. On the other hand, features might correlate with each other, thus conventional L2 distance can not well measure the distance between samples. Although some works have been proposed to solve these problems, rare connects them together. In this paper, we propose a novel method that exploits knowledge in both data manifold and features correlation. We adopt an approximation of Earth Mover's Distance (EMD) as metric and add a graph regularized term based on EMD to NMF. Furthermore, we propose an efficient multiplicative iteration algorithm to solve it. Our empirical study shows the encouraging results of the proposed algorithm comparing with other NMF methods.
Accession Number: 20165103147046
-
Record 10 of
Title:Mode-order-invariant beam splitter on silicon-on-insulator waveguide
Author(s):Liao, Jianwen(1); Wang, Guoxi(1); Zhang, Wenfu(2)
Source: IEEE International Conference on Group IV Photonics GFP Volume: 2016-November Issue: DOI: 10.1109/GROUP4.2016.7739134 Published: November 8, 2016
Abstract:We present a mode splitter which is able to split the TE0&TE1 modes without changing the mode order. High coupling efficiency (>-2 dB), low insertion loss ( ? 2016 IEEE.
Accession Number: 20165003114281
-
Record 11 of
Title:Infrared small target and background separation via column-wise weighted robust principal component analysis
Author(s):Dai, Yimian(1); Wu, Yiquan(1,2,3,4); Song, Yu(1)
Source: Infrared Physics and Technology Volume: 77 Issue: DOI: 10.1016/j.infrared.2016.06.021 Published: July 1, 2016
Abstract:When facing extremely complex infrared background, due to the defect of l1 norm based sparsity measure, the state-of-the-art infrared patch-image (IPI) model would be in a dilemma where either the dim targets are over-shrinked in the separation or the strong cloud edges remains in the target image. In order to suppress the strong edges while preserving the dim targets, a weighted infrared patch-image (WIPI) model is proposed, incorporating structural prior information into the process of infrared small target and background separation. Instead of adopting a global weight, we allocate adaptive weight to each column of the target patch-image according to its patch structure. Then the proposed WIPI model is converted to a column-wise weighted robust principal component analysis (CWRPCA) problem. In addition, a target unlikelihood coefficient is designed based on the steering kernel, serving as the adaptive weight for each column. Finally, in order to solve the CWPRCA problem, a solution algorithm is developed based on Alternating Direction Method (ADM). Detailed experiment results demonstrate that the proposed method has a significant improvement over the other nine classical or state-of-the-art methods in terms of subjective visual quality, quantitative evaluation indexes and convergence rate. ? 2016 Elsevier B.V.
Accession Number: 20162702569229
-
Record 12 of
Title:Hierarchical learning of large-margin metrics for large-scale image classification
Author(s):Lei, Hao(1,2); Mei, Kuizhi(2); Xin, Jingmin(2); Dong, Peixiang(2); Fan, Jianping(3)
Source: Neurocomputing Volume: 208 Issue: DOI: 10.1016/j.neucom.2016.01.100 Published: October 5, 2016
Abstract:Large-scale image classification is a challenging task and has recently attracted active research interests. In this paper, a new algorithm is developed to achieve more effective implementation of large-scale image classification by hierarchical learning of large-margin metrics (HLMMs). A hierarchical visual tree is seamlessly integrated with metric learning to learn a set of node-specific/category-specific large-margin metrics. First, a hierarchical visual tree is learned to characterize the inter-category visual correlations effectively and organize large numbers of image categories in a coarse-to-fine fashion. Second, a new algorithm is developed to support hierarchical learning of large-margin metrics by training nearest class mean (NCM) classifiers over our hierarchical visual tree. In addition, we also consider dimensionality reduction as a regularizer for high-dimensional data in our large-margin metric learning. Two top-down approaches are developed for supporting hierarchical learning of large-margin metrics. We focus on learning more discriminative metrics for NCM node classifiers to identify the visually similar sub-nodes (visually similar image categories) under the same parent node over our hierarchical visual tree. A mini-batch stochastic gradient descend method is used to optimize our HLMMs learning algorithm. The experimental results on ImageNet Large Scale Visual Recognition Challenge 2010 dataset (ILSVRC2010) have demonstrated that our HLMMs learning algorithm is very promising for supporting large-scale image classification. ? 2016 Elsevier B.V.
Accession Number: 20163702807173
