arXiv subjects

Jun Du

Publications and source records attributed to Jun Du.

174 records · Page 10Linked to original sources

arXiv2018

Multi-Scale Attention with Dense Encoder for Handwritten Mathematical Expression Recognition

Handwritten mathematical expression recognition is a challenging problem due to the complicated two-dimensional structures, ambiguous handwriting input and variant scales of handwritten math symbols. To settle this problem, we utilize the attention based encoder-decoder model that recognizes mathematical expression images from two-dimensional layouts to one-dimensional LaTeX strings. We improve the encoder by employing densely connected convolutional networks as they can strengthen feature extraction and facilitate gradient propagation especially on a small training set. We also present a novel multi-scale attention model which is employed to deal with the recognition of math symbols in different scales and save the fine-grained details that will be dropped by pooling operations. Validated on the CROHME competition task, the proposed method significantly outperforms the state-of-the-art methods with an expression recognition accuracy of 52.8% on CROHME 2014 and 50.1% on CROHME 2016, by only using the official training dataset.

cs.CV↗

arXiv2018

Sliding Line Point Regression for Shape Robust Scene Text Detection

Traditional text detection methods mostly focus on quadrangle text. In this study we propose a novel method named sliding line point regression (SLPR) in order to detect arbitrary-shape text in natural scene. SLPR regresses multiple points on the edge of text line and then utilizes these points to sketch the outlines of the text. The proposed SLPR can be adapted to many object detection architectures such as Faster R-CNN and R-FCN. Specifically, we first generate the smallest rectangular box including the text with region proposal network (RPN), then isometrically regress the points on the edge of text by using the vertically and horizontally sliding lines. To make full use of information and reduce redundancy, we calculate x-coordinate or y-coordinate of target point by the rectangular box position, and just regress the remaining y-coordinate or x-coordinate. Accordingly we can not only reduce the parameters of system, but also restrain the points which will generate more regular polygon. Our approach achieved competitive results on traditional ICDAR2015 Incidental Scene Text benchmark and curve text detection dataset CTW1500.

cs.CV↗

arXiv2018

Trajectory-based Radical Analysis Network for Online Handwritten Chinese Character Recognition

Recently, great progress has been made for online handwritten Chinese character recognition due to the emergence of deep learning techniques. However, previous research mostly treated each Chinese character as one class without explicitly considering its inherent structure, namely the radical components with complicated geometry. In this study, we propose a novel trajectory-based radical analysis network (TRAN) to firstly identify radicals and analyze two-dimensional structures among radicals simultaneously, then recognize Chinese characters by generating captions of them based on the analysis of their internal radicals. The proposed TRAN employs recurrent neural networks (RNNs) as both an encoder and a decoder. The RNN encoder makes full use of online information by directly transforming handwriting trajectory into high-level features. The RNN decoder aims at generating the caption by detecting radicals and spatial structures through an attention model. The manner of treating a Chinese character as a two-dimensional composition of radicals can reduce the size of vocabulary and enable TRAN to possess the capability of recognizing unseen Chinese character classes, only if the corresponding radicals have been seen. Evaluated on CASIA-OLHWDB database, the proposed approach significantly outperforms the state-of-the-art whole-character modeling approach with a relative character error rate (CER) reduction of 10%. Meanwhile, for the case of recognition of 500 unseen Chinese characters, TRAN can achieve a character accuracy of about 60% while the traditional whole-character method has no capability to handle them.

cs.CV↗

arXiv2017

Observation of spin-orbit magnetoresistance in metallic thin films on magnetic insulators

A magnetoresistance effect induced by the Rashba spin-orbit interaction was predicted, but not yet observed, in bilayers consisting of normal metal and ferromagnetic insulator. Here, we present an experimental observation of this new type of spin-orbit magnetoresistance (SOMR) effect in a bilayer structure Cu[Pt]/Y3Fe5O12 (YIG), where the Cu/YIG interface is decorated with nanosize Pt islands. This new MR is apparently not caused by the bulk spin-orbit interaction because of the negligible spin-orbit interaction in Cu and the discontinuity of the Pt islands. This SOMR disappears when the Pt islands are absent or located away from the Cu/YIG interface, therefore we can unambiguously ascribe it to the Rashba spin-orbit interaction at the interface enhanced by the Pt decoration. The numerical Boltzmann simulations are consistent with the experimental SOMR results in the angular dependence of magnetic field and the Cu thickness dependence. Our finding demonstrates the realization of the spin manipulation by interface engineering.

cond-mat.mtrl-sci↗

arXiv2017

A GRU-based Encoder-Decoder Approach with Attention for Online Handwritten Mathematical Expression Recognition

In this study, we present a novel end-to-end approach based on the encoder-decoder framework with the attention mechanism for online handwritten mathematical expression recognition (OHMER). First, the input two-dimensional ink trajectory information of handwritten expression is encoded via the gated recurrent unit based recurrent neural network (GRU-RNN). Then the decoder is also implemented by the GRU-RNN with a coverage-based attention model. The proposed approach can simultaneously accomplish the symbol recognition and structural analysis to output a character sequence in LaTeX format. Validated on the CROHME 2014 competition task, our approach significantly outperforms the state-of-the-art with an expression recognition accuracy of 52.43% by only using the official training dataset. Furthermore, the alignments between the input trajectories of handwritten expressions and the output LaTeX sequences are visualized by the attention mechanism to show the effectiveness of the proposed method.

cs.CV↗

arXiv2017

Temperature effects of the magnetic tunnel junctions with periodic grating barrier

We have developed a tunneling theory to describe the temperature dependence of tunneling magnetoresistance (TMR) of the magnetic tunnel junctions (MTJs) with periodic grating barrier. Through the Patterson function approach, the theory can handle easily the influence of the lattice distortion of the barrier on the tunneling process of the electrons. The lattice distortion of the barrier is sensible to the temperature and can be quite easily weakened by the thermal relaxation of the strain, and thus the tunneling process of the electrons gets changed highly with the variation of the temperature of the system. That is just the physical mechanism for the temperature dependence of the TMR. From it, we find that the decrease of TMR with rising temperature is mostly carried by a change in the antiparallel resistance ($R_{AP}$), and the parallel resistance ($R_{P}$) changes so little that it seems roughly constant, if compared to the $R_{AP}$, and that, for the annealed MTJ, the $R_{AP}$ is significantly more sensitive to the strain than the $R_{P}$, and for non-annealed MTJ, both the $R_{P}$ and $R_{AP}$ are not sensitive to the strain. They are both in agreement with the experiments of the MgO-based MTJs. Other relevant properties are also discussed.

cond-mat.mes-hall↗

arXiv2017

Multi-Objective Learning and Mask-Based Post-Processing for Deep Neural Network Based Speech Enhancement

We propose a multi-objective framework to learn both secondary targets not directly related to the intended task of speech enhancement (SE) and the primary target of the clean log-power spectra (LPS) features to be used directly for constructing the enhanced speech signals. In deep neural network (DNN) based SE we introduce an auxiliary structure to learn secondary continuous features, such as mel-frequency cepstral coefficients (MFCCs), and categorical information, such as the ideal binary mask (IBM), and integrate it into the original DNN architecture for joint optimization of all the parameters. This joint estimation scheme imposes additional constraints not available in the direct prediction of LPS, and potentially improves the learning of the primary target. Furthermore, the learned secondary information as a byproduct can be used for other purposes, e.g., the IBM-based post-processing in this work. A series of experiments show that joint LPS and MFCC learning improves the SE performance, and IBM-based post-processing further enhances listening quality of the reconstructed speech.

cs.SD↗

arXiv2016

Synthesis, Optical, and Magnetic Properties of Ba$_2$Ni$_3$F$_{10}$ Nanowires

A low temperature hydrothermal route has been developed, and pure phase Ba$_2$Ni$_3$F$_{10}$ nanowires have been successfully prepared under the optimized conditions. Under the 325 nm excitation, the Ba$_2$Ni$_3$F$_{10}$ nanowires exhibit three emission bands with peak positions locating at 360 nm, 530 nm, and 700 nm, respectively. Combined with the first-principles calculations, the photoluminescence property can be explained by the electron transitions between the t2g and eg orbitals. Clear hysteresis loops observed below the temperature of 60 K demonstrates the weak ferromagnetism in Ba$_2$Ni$_3$F$_{10}$ nanowires, which has been attributed to the surface strain of nanowires. Exchange bias with blocking temperature of 55 K has been observed, which originates from the magnetization pinning under the cooling field due to antiferromagnetic core/weak ferromagnetic shell structure of Ba2Ni3F10 nanowires.

cond-mat.mtrl-sci↗

arXiv2016

Unsaturated magnetoconductance of epitaxial La0.7Sr0.3MnO3 thin films in pulsed magnetic fields up to 60 T

We report on the temperature and field dependence of resistance of La0.7Sr0.3MnO3 thin films over a wide temperature range and in pulsed magnetic fields up to 60 T. The epitaxial La0.7Sr0.3MnO3 thin films were deposited by laser molecular beam epitaxy. High magnetic field magnetoresistance curves were fitted by the Brillouin function, which indicated the existence of magnetically polarized regions and the underlying hopping mechanism. The unsaturated magnetoconductance was the most striking finding observed in pulsed magnetic fields up to 60 T. These observations can deepen the fundamental understanding of the colossal magnetoresistance in manganites with strong correlation of transport properties and magnetic ordering.

cond-mat.str-el↗

arXiv2016

Elementary specific spin and orbital moments of ultrathin CoFeB amorphous films on GaAs(100)

Nanoscale CoFeB amorphous films have been synthesized on GaAs(100) and studied with X-ray magnetic circular dichroism (XMCD) and transmission electron microscopy (TEM). We have found that the ratios of the orbital to spin magnetic moments of both the Co and Fe in the ultrathin amorphous film have been enhanced by more than 300% compared with those of the bulk crystalline Co and Fe, and in specifically, a large orbital moment of 0.56*10^-6 B from the Co atoms has been observed and at the same time the spin moment of the Co atoms remains comparable to that of the bulk hcp Co. The results indicate that the large uniaxial magnetic anisotropy (UMA) observed in the ultrathin CoFeB film on GaAs(100) is related to the enhanced spin-orbital coupling of the Co atoms in the CoFeB. This work offers experimental evidences of the correlation between the UMA and the elementary specific spin and orbital moments in the CoFeB amorphous film on the GaAs(100) substrate, which is significant for spintronics applications.

cond-mat.mtrl-sci↗

arXiv2016

Magnetic coherent tunnel junctions with periodic grating barrier

A new spintronic theory has been developed for the magnetic tunnel junction (MTJ) with single-crystal barrier. The barrier will be treated as a diffraction grating with intralayer periodicity, the diffracted waves of tunneling electrons thus contain strong coherence, both in charge and especially in spin. The theory can answer the two basic problems present in MgO-based MTJs: (1) Why does the tunneling magnetoresistance (TMR) oscillate with the barrier thickness? (2) Why is the TMR still far away from infinity when the two electrodes are both half-metallic? Other principal features of TMR can also be explained and reproduced by the present work. It also provides possible ways to modulate the oscillation of TMR, and to enhance TMR so that it can tend to infinity. Within the theory, the barrier, as a periodic diffraction grating, can get rid of the confinement in width, it can vary from nanoscale to microscale. Based on those results, a future-generation MTJ is proposed where the three pieces can be fabricated separately and then assembled together, it is especially appropriate for the layered materials, e.g., MoS2 and graphite, and most feasible for industries.

cond-mat.mes-hall↗

arXiv2015

Magnetoelectricity coupled exchange bias in BaMnF$_4$

Multiferroic BaMnF$_4$ powder were prepared by hydrothermal method. Hysteretic field dependent magnetization curve at 5 K confirms the weak ferromagnetism aroused from the canted antiferromagnetic spins by magnetoelectric coupling. The blocking temperature of 65 K for exchange bias coincides well with the peak at 65 K in the zero-field cooled temperature-dependent magnetization curve, which has been assigned to the onset temperature of two-dimensional antiferromagnetism. An upturn kink of exchange field and coercivity with decreasing temperature was observed from 40 K to 20 K, which is consistent with the two-dimensional to three-dimensional antiferromagnetic transition at Néel temperature (~26 K). In contrast to the conventional mechanism of magnetization pinned by interfacial exchange coupling in multiphases, the exchange bias in BaMnF$_4$ is argued to be a bulk effect in single phase, due to the magnetization pinned by the polarization through magnetoelectric coupling.

cond-mat.mtrl-sci↗