Fujitsu Research and Development Center

Beijing, China

Fujitsu Research and Development Center

Beijing, China

Time filter

Source Type

Li F.,Intelligent Systems Technology, Inc. | Han C.,Intelligent Systems Technology, Inc. | Huang M.,Intelligent Systems Technology, Inc. | Zhu X.,Intelligent Systems Technology, Inc. | And 3 more authors.
Coling 2010 - 23rd International Conference on Computational Linguistics, Proceedings of the Conference | Year: 2010

In this paper, we focus on object feature based review summarization. Different from most of previous work with linguistic rules or statistical methods, we formulate the review mining task as a joint structure tagging problem. We propose a new machine learning framework based on Conditional Random Fields (CRFs). It can employ rich features to jointly extract positive opinions, negative opinions and object features for review sentences. The linguistic structure can be naturally integrated into model representation. Besides linear- chain structure, we also investigate conjunction structure and syntactic tree structure in this framework. Through extensive experiments on movie review and product review data sets, we show that structure-aware models outperform many state-of-the-art approaches to review mining.


Yan M.,Fujitsu Research and Development Center | Tao Z.,Fujitsu Research and Development Center | Yan W.,Fujitsu Research and Development Center | Li L.,Fujitsu Research and Development Center | And 2 more authors.
IEEE Photonics Technology Letters | Year: 2012

The inter-carrier interference (ICI) in the no-guard-interval orthogonal frequency-division-multiplexing (NGI-OFDM) system causes severe performance degradation when 2-samples/symbol (Sa/sym) digital signal processing (DSP) is used at the receiver. To enable narrow bandwidth devices and low sampling rate DSP, a novel ICI suppression (ICIS) method is proposed and experimentally demonstrated in this letter. The proposed method improves the performance of the 4-Gbaud 25-carrier NGI-OFDM system by up to 4 dB, and 2-Sa/sym DSP with ICIS achieves comparable performance as the conventional 8-Sa/sym DSP without ICIS, with moderately increased complexity. © 2006 IEEE.


Sun B.,Hunan University | Li S.,Hunan University | Sun J.,Fujitsu Research and Development Center
IEEE Transactions on Image Processing | Year: 2014

Currently, most electrophotographic printers use halftoning technique to print continuous tone images, so scanned images obtained from such hard copies are usually corrupted by screen like artifacts. In this paper, a new model of scanned halftone image is proposed to consider both printing distortions and halftone patterns. Based on this model, an adaptive filtering based descreening method is proposed to recover high quality contone images from the scanned images. Image redundancy based denoising algorithm is first adopted to reduce printing noise and attenuate distortions. Then, screen frequency of the scanned image and local gradient features are used for adaptive filtering. Basic contone estimate is obtained by filtering the denoised scanned image with an anisotropic Gaussian kernel, whose parameters are automatically adjusted with the screen frequency and local gradient information. Finally, an edge-preserving filter is used to further enhance the sharpness of edges to recover a high quality contone image. Experiments on real scanned images demonstrate that the proposed method can recover high quality contone images from the scanned images. Compared with the state-of-the-art methods, the proposed method produces very sharp edges and much cleaner smooth regions. © 2014 IEEE.


Wu J.,Fujitsu Research and Development Center
IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, PIMRC | Year: 2012

In this paper, we newly design a CSI-RS for virtualized antenna system when LTE and LTE-A are efficiently coexisting in Rel-10 and beyond. To this end, we newly insert a small amount of CSI-RS into the locations which is close to the existing CRS positions for legacy LTE as enough as possible. The newly designed CSI-RS modulated by space time coded signal efficiently works well for LTE-A UE channel estimation when the number of antennas involved in LTE-A system is larger than four which are normally used for LTE system as maximal. In such a legacy system with eight physical transmit antennas, eight antenna ports with eight orthogonal CSI-RSs are exploited for LTE-A system while virtualized four antenna ports are for LTE system. The benefit of the proposed solution is to significantly reduce the CSI-RS overhead, compared to the overhead employed in the existing Rel-10 system. © 2012 IEEE.


Miao Q.,Fujitsu Research and Development Center | Meng Y.,Fujitsu Research and Development Center | Sun J.,Fujitsu Research and Development Center
Proceedings of 2016 IEEE International Conference on Cloud Computing and Big Data Analysis, ICCCBDA 2016 | Year: 2016

With the successful adoption of Web 2.0 applications, consumers are increasingly voicing their opinions and concerns, and providing suggestions in online review communities. Consumers who significantly influence the opinions and decisions of others through communicative acts are called opinion leaders. These leaders usually capture the most representative opinions and play a crucial role in electronic word-of-mouth. Since opinion leaders have different expertise and interests, they are typically topic sensitive. In this paper, we discuss the challenges of identifying topic-sensitive opinion leaders in online review communities, and present a two-stage approach. In Stage I, we derive opinion leaders' interest from their tags. In Stage II, we rank leaders according to not only the link structure of networks but also opinion leaders' personal expertise and interests. A preliminary experiment on real-world data indicates the proposed approach is effective. © 2016 IEEE.


Wang W.,Fujitsu Research and Development Center | Wang X.,Fujitsu Research and Development Center
2015 IEEE 82nd Vehicular Technology Conference, VTC Fall 2015 - Proceedings | Year: 2015

In this paper, three schemes based on the proportional fairness scheduling with power control are proposed for the virtual full-duplex (VFD) scheme by jointly considering the downlink (DL) and uplink (UL) UE pair selection and UL power control. In particular, the optimal scheme is derived via the exhaustive search for the UE pair and UL power. The second scheme reduces the complexity of the UE pair selection via choosing the DL UE and the UL UE separately, and the third scheme further reduces the UL power setting by reducing the candidate UL power levels. The simulation results verify that compared to the VFD scheme without power control, the largest spectral efficiency gain is obtained by the optimal scheme with the highest complexity, while the smallest spectral efficiency gain is achieved by the third scheme with the lowest complexity. © 2015 IEEE.


Shi Z.,Fujitsu Research and Development Center | Liu R.,Fujitsu Research and Development Center
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Year: 2015

Online and stochastic gradient methods have emerged as potent tools in large scale optimization with both smooth convex and nonsmooth convex problems from the classes C1,1(Rp) and C1,0(Rp) respectively. However, to our best knowledge, there is few paper using incremental gradient methods to optimization the intermediate classes of convex problems with Hölder continuous functions C1, v(Rp). In order to fill the difference and the gap between the methods for smooth and nonsmooth problems, in this work, we propose several online and stochastic universal gradient methods, which we do not need to know the actual degree of the smoothness of the objective function in advance. We expanded the scope of the problems involved in machine learning to Hölder continuous functions and to propose a general family of first-order methods. Regret and convergent analysis shows that our methods enjoy strong theoretical guarantees. For the first time, we establish algorithms that enjoys a linear convergence rate for convex functions that have Hölder continuous gradients. © Springer International Publishing Switzerland 2015.


Wu J.,Fujitsu Research and Development Center
IEEE International Conference on Communications | Year: 2010

In this paper, we newly design a non-uniform QAM signal constellation, which could be appropriately applied for hierarchical modulation. This novel constellation is constructed by a certain number of equilateral triangles and hierarchically divided into two layers; one for MSB and the other for LSB. The advantage of using this constellation compared to uniform rectangular QAM constellation is that the MED for LSB is constantly maintained whereas the MED for MSB is remarkably increased. To this end, we evaluate the performance in terms of SER by means of link level simulation. Consequently, the newly designed constellation does not compromise the modulation performance between MSB and LSB, and the SNR gain of MSB as opposed to uniform rectangular QAM constellation reaches 0.5∼0.8dB but without sacrificing the performance of LSB. ©2010 IEEE.


Shi Z.,Fujitsu Research and Development Center | Liu R.,Fujitsu Research and Development Center
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Year: 2015

In this work, we generalized and unified two recent completely different works of Jascha [10] and Lee [2] respectively into one by proposing the proximal stochastic Newton-type gradient (PROXTONE) method for optimizing the sums of two convex functions: one is the average of a huge number of smooth convex functions, and the other is a nonsmooth convex function. Our PROXTONE incorporates second order information to obtain stronger convergence results, that it achieves a linear convergence rate not only in the value of the objective function, but also for the solution. The proofs are simple and intuitive, and the results and technique can be served as a initiate for the research on the proximal stochastic methods that employ second order information. The methods and principles proposed in this paper can be used to do logistic regression, training of deep neural network and so on. Our numerical experiments shows that the PROXTONE achieves better computation performance than existing methods. © Springer International Publishing Switzerland 2015.


Wang W.,Fujitsu Research and Development Center | Wang X.,Fujitsu Research and Development Center
IEEE Vehicular Technology Conference | Year: 2015

In this paper, a virtual full duplex scheme is proposed by considering the joint selection of transmission point (TP) and downlink/uplink (DL/UL) configuration. Specifically, in a heterogenous network (HetNet), consisting of multiple macro TPs (MTPs) and small TPs (STPs) connected to the same virtual base station (BS) pool, the UEs in the coverage of the same small cell can be served by an MTP and an STP simultaneously with different DL and UL settings. Moreover, the interference from the DL transmission to the UL reception can be canceled via successive interference cancellation since the DL data and the corresponding scheduling information are well known at virtual BS pool. The analysis and system-level simulation results demonstrate that the highest DL and UL spectral efficiency is achieved by the scheme that MTP and STP serve UEs in DL and UL, respectively, at the same time. © 2015 IEEE.

Loading Fujitsu Research and Development Center collaborators
Loading Fujitsu Research and Development Center collaborators