Entity

Time filter

Source Type


Guo Y.,Dalian University of Technology | Guo Y.,Postdoctoral Scientific Research Workstation | He R.,CAS Institute of Automation | Zheng W.-S.,Sun Yat Sen University | And 2 more authors.
Neurocomputing | Year: 2013

Spectral regression has been an efficient and powerful tool for face recognition. However, spectral regression is sensitive to the errors incurred by inaccurate annotation and occlusion. This paper studies robust spectral regression based discriminant subspace learning from correntropy and spatially smooth structure of facial subspace. First, we formulate the robust discriminant subspace learning problem as a maximum correntropy problem, which finds the most correlation solution between spectral targets and predictions. Second, total variation (TV) regularization is imposed on the correntropy objective to learn a spatially smooth face structure. Lastly, based on the additive form of half-quadratic optimization, we cast the maximum correntropy problem into a compound regularization model, which can be efficiently optimized via an accelerated proximal gradient algorithm. Compared with iteratively reweighted least squares based methods, the proposed method can not only improve recognition rates but also reduce computational cost. Experimental results on a couple of face recognition datasets demonstrate the robustness and effectiveness of our method against inaccurate annotation and occlusion. © 2013 Elsevier B.V. Source


Lu Z.,Postdoctoral Scientific Research Workstation
ICIC Express Letters, Part B: Applications | Year: 2016

The designers in recon_gurable computing _elds always require considerable knowledge in both software and hardware to build hybrid applications. The main chal- lenges and barriers that hamper the wide adoption of recon_gurable computing systems are the lack of high-level design and development tools. This paper presents a compila- tion framework for IR2HDL mapping for Field Programmable Gate Arrays. The modulo scheduling schemes with a constant initiation interval in most compilers schedule the iterations and generate pipelines by using a pipeline division method. In order to reduce the pipeline depth and increase throughput, a parallel loop scheduling scheme and im- proved modulo scheduling method are presented. The test case on selected kernels shows the method improves the delay stage and the scheme shows signi_cant performance im- provement. © 2016 ICIC International. Source


Wang X.-L.,China University of Petroleum - Beijing | Wang X.-L.,Postdoctoral Scientific Research Workstation | Shuai J.,China University of Petroleum - Beijing
Gongcheng Lixue/Engineering Mechanics | Year: 2011

River-crossing pipeline is threatened by flood which could induce the pipeline being eroded and exposed, moreover, floating in a large scale. Under the combined effects of dynamic water, buoyancy, gravity and resistance of bank soil, pipeline generally presents three-dimensional bending. A mechanical analysis model is built to investigate the mechanical behaviour of pipeline exposed in flood. Taking into account the nonlinear pipe-soil interaction, the axial force and material nonlinearity of pipeline steel, the dimensional bending curve and a deformation compatibility equation of pipeline are derived. And pipeline stress and strain are calculated with an iterative solving method. Based on the proposed analytical methodology, a computer program is developed and the pipeline deformation, stress and strain are analyzed through a series of examples. Each example is subsequently investigated with the finite element method and the calculation results of the proposed method are close to that of the finite element method. Finally, a safety assessment method for pipeline in flood is proposed according to limit state theory, by which the safety of a pipeline exposed in mountain torrent is assessed. Source


Li X.M.,PLA | Wang N.,PLA | Wang N.,Postdoctoral Scientific Research Workstation | Shi W.K.,PLA
Applied Mechanics and Materials | Year: 2014

In order to realize the zero-range-target simulation, the echo delay simulation method for the fixed radar and simulator is given. By this method, zero-range target simulation can be realized. Finally, it is verified by a computation example. © (2014) Trans Tech Publications, Switzerland. Source


Lu Z.,Postdoctoral Scientific Research Workstation
ICIC Express Letters | Year: 2016

Convolutional networks are computational models that are widely used in pattern recognition applications. Operations in these applications require convolution filtering at each pixel. Depending on the size of input image and convolution kernel, the two-dimensional convolution can require significant amounts of computation, thus suggesting a highly parallel implementation in hardware. This paper describes a loopbased computation and storage partition scheme for convolution filtering on Field Programmable Gate Arrays (FPGAs), which can guide the proper parameterization of a convolution filter bank on a given architecture. The case study on selected kernels shows the method can express the computation and storage tradeoffs involved in mapping a Convolutional Neural Networks (CNNs) on FPGAs. © 2016, ICIC Express Letters Office. All rights reserved. Source

Discover hidden collaborations