Searcharxiv⌕ Search

arXiv subjects

Dan Liu

Publications and source records attributed to Dan Liu.

At least 73 records · Page 4Linked to original sources
arXiv2020

Towards Interpretable and Robust Hand Detection via Pixel-wise Prediction

The lack of interpretability of existing CNN-based hand detection methods makes it difficult to understand the rationale behind their predictions. In this paper, we propose a novel neural network model, which introduces interpretability into hand detection for the first time. The main improvements include: (1) Detect hands at pixel level to explain what pixels are the basis for its decision and improve transparency of the model. (2) The explainable Highlight Feature Fusion block highlights distinctive features among multiple layers and learns discriminative ones to gain robust performance. (3) We introduce a transparent representation, the rotation map, to learn rotation features instead of complex and non-transparent rotation and derotation layers. (4) Auxiliary supervision accelerates the training process, which saves more than 10 hours in our experiments. Experimental results on the VIVA and Oxford hand detection and tracking datasets show competitive accuracy of our method compared with state-of-the-art methods with higher speed.

cs.CV↗
arXiv2020

LiftTiles: Constructive Building Blocks for Prototyping Room-scale Shape-changing Interfaces

Large-scale shape-changing interfaces have great potential, but creating such systems requires substantial time, cost, space, and efforts, which hinders the research community to explore interactions beyond the scale of human hands. We introduce modular inflatable actuators as building blocks for prototyping room-scale shape-changing interfaces. Each actuator can change its height from 15cm to 150cm, actuated and controlled by air pressure. Each unit is low-cost (8 USD), lightweight (10 kg), compact (15 cm), and robust, making it well-suited for prototyping room-scale shape transformations. Moreover, our modular and reconfigurable design allows researchers and designers to quickly construct different geometries and to explore various applications. This paper contributes to the design and implementation of highly extendable inflatable actuators, and demonstrates a range of scenarios that can leverage this modular building block.

cs.HC↗
arXiv2019

Enjoy the Untrusted Cloud: A Secure, Scalable and Efficient SQL-like Query Framework for Outsourcing Data

While the security of the cloud remains a concern, a common practice is to encrypt data before outsourcing them for utilization. One key challenging issue is how to efficiently perform queries over the ciphertext. Conventional crypto-based solutions, e.g. partially/fully homomorphic encryption and searchable encryption, suffer from low performance, poor expressiveness and weak compatibility. An alternative method that utilizes hardware-assisted trusted execution environment, i.e., Intel SGX, has emerged recently. On one hand, such work lacks of supporting scalable access control over multiple data users. On the other hand, existing solutions are subjected to the key revocation problem and knowledge extractor vulnerability. In this work, we leverage the newly hardware-assisted methodology and propose a secure, scalable and efficient SQL-like query framework named QShield. Building upon Intel SGX, QShield can guarantee the confidentiality and integrity of sensitive data when being processed on an untrusted cloud platform. Moreover, we present a novel lightweight secret sharing method to enable multi-user access control in QShield, while tackling the key revocation problem. Furthermore, with an additional trust proof mechanism, QShield guarantees the correctness of queries and significantly alleviates the possibility to build a knowledge extractor. We implemented a prototype for QShield and show that QShield incurs minimum performance cost.

cs.CR↗
arXiv2019

Catalytic Formation of Narrow Nb Nanowires inside Carbon Nanotubes

We propose a previously unexplored way to form Nb nanowires from NbCl$_3$ molecules inside carbon nanotubes (CNTs). We have studied this reaction by ab initio density functional calculations and found it to be catalytically promoted in presence of graphitic carbon. Our results suggest that chemisorption of NbCl$_3$ on the CNT is accompanied by a charge transfer of ~0.5 electrons to the nanotube wall, which significantly weakens the Nb-Cl bond. We found that the bcc structure of Nb is not affected by the small diameter of the nanowire inside a CNT. We have also identified strong covalent bonds between the nanowires and the surrounding nanotube that are accompanied by a similar charge transfer of <0.5 e from surface Nb atoms to the nanotube. The large electronic density of states of bulk Nb at E$_F$ is not changed much in the confined geometry, suggesting that ultra-narrow nanowires of Nb may keep their superconducting behavior and form Josephson junctions in the quasi-1D geometry while being protected from the ambient by the enclosing CNT structure.

cond-mat.mes-hall↗
arXiv2019

In-Plane Breathing and Shear Modes in Low-Dimensional Nanostructures

We use continuum elasticity theory to revise scaling laws for radial breathing-like and shear-like vibration modes in quasi-2D nanostructures including finite-width nanoribbons and finite-size thin circular discs. Such modes can be observed spectroscopically in corresponding nanostructures of graphene and phosphorene and can be determined numerically by atomistic ab initio density functional theory and classical force-field calculation. The revised scaling laws differ from previously used expressions, some of which display an unphysical asymptotic behavior. Apart from model assumptions describing the effect of edge termination, the continuum scaling laws have no adjustable parameters and display correct asymptotic behavior. These scaling laws yield excellent agreement with experimental and numerical results for vibration frequencies in both isotropic and anisotropic structures as well as useful expressions for the frequency dependence on structure size and edge termination.

cond-mat.mes-hall↗
arXiv2019

Dynamical Jahn-Teller effect in the first excited C$_{60}^-$

The Jahn-Teller effect of C$_{60}$ anions in the first electronically excited states was theoretically investigated. The orbital vibronic coupling parameters for the $t_{1g}$ next lowest unoccupied molecular orbitals were derived from the Kohn-Sham orbital levels with hybrid B3LYP functional by using the frozen phonon approach. With the use of these coupling parameters, the vibronic states of the first excited C$_{60}^-$ were derived by exactly diagonalizing the dynamical Jahn-Teller Hamiltonian. The dynamical Jahn-Teller stabilization energy of the first excited C$_{60}^-$ is stronger than that of the ground electronic states.

physics.chem-ph↗
arXiv2019

Scale Invariant Fully Convolutional Network: Detecting Hands Efficiently

Existing hand detection methods usually follow the pipeline of multiple stages with high computation cost, i.e., feature extraction, region proposal, bounding box regression, and additional layers for rotated region detection. In this paper, we propose a new Scale Invariant Fully Convolutional Network (SIFCN) trained in an end-to-end fashion to detect hands efficiently. Specifically, we merge the feature maps from high to low layers in an iterative way, which handles different scales of hands better with less time overhead comparing to concatenating them simply. Moreover, we develop the Complementary Weighted Fusion (CWF) block to make full use of the distinctive features among multiple layers to achieve scale invariance. To deal with rotated hand detection, we present the rotation map to get rid of complex rotation and derotation layers. Besides, we design the multi-scale loss scheme to accelerate the training process significantly by adding supervision to the intermediate layers of the network. Compared with the state-of-the-art methods, our algorithm shows comparable accuracy and runs a 4.23 times faster speed on the VIVA dataset and achieves better average precision on Oxford hand detection dataset at a speed of 62.5 fps.

cs.CV↗
arXiv2019

Learning Efficient Lexically-Constrained Neural Machine Translation with External Memory

Recent years has witnessed dramatic progress of neural machine translation (NMT), however, the method of manually guiding the translation procedure remains to be better explored. Previous works proposed to handle such problem through lexcially-constrained beam search in the decoding phase. Unfortunately, these lexically-constrained beam search methods suffer two fatal disadvantages: high computational complexity and hard beam search which generates unexpected translations. In this paper, we propose to learn the ability of lexically-constrained translation with external memory, which can overcome the above mentioned disadvantages. For the training process, automatically extracted phrase pairs are extracted from alignment and sentence parsing, then further be encoded into an external memory. This memory is then used to provide lexically-constrained information for training through a memory-attention machanism. Various experiments are conducted on WMT Chinese to English and English to German tasks. All the results can demonstrate the effectiveness of our method.

cs.CL↗
arXiv2019

Effect of Net Charge on the Relative Stability of 2D Boron Allotropes

We study the effect of electron doping on the bonding character and stability of two-dimensional (2D) structures of elemental boron, called borophene, which is known to form many stable allotropes. Our {\em ab initio} calculations for the neutral system reveal previously unknown stable 2D $ε$-B and $ω$-B structures. We find that the chemical bonding characteristic in this and other boron structures is strongly affected by extra charge. Beyond a critical degree of electron doping, the most stable allotrope changes from $ε$-B to a buckled honeycomb structure. Additional electron doping, mimicking a transformation of boron to carbon, causes a gradual decrease in the degree of buckling of the honeycomb lattice that can be interpreted as piezoelectric response. Net electron doping can be achieved by placing borophene in direct contact with layered electrides such as Ca$_{2}$N. We find that electron doping can be doubled by changing from the B/Ca$_{2}$N bilayer to the Ca$_{2}$N/B/Ca$_{2}$N sandwich geometry.

cond-mat.mtrl-sci↗
arXiv2018

Two-dimensional Mechanical Metamaterials with Unusual Poisson Ratio Behavior

We design two-dimensional (2D) mechanical metamaterials that may be deformed substantially at little or no energy cost. Examples of such deformable structures are assemblies of rigid isosceles triangles hinged in their corners on the macro-scale, or polymerized phenanthrene molecules forming porous graphene on the nano-scale. In these and in a large class of related structures, the Poisson ratio $ν$ diverges for particular strain values. $ν$ also changes its magnitude and sign, and displays a shape memory effect.

cond-mat.mes-hall↗
arXiv2018

The USTC-NEL Speech Translation system at IWSLT 2018

This paper describes the USTC-NEL system to the speech translation task of the IWSLT Evaluation 2018. The system is a conventional pipeline system which contains 3 modules: speech recognition, post-processing and machine translation. We train a group of hybrid-HMM models for our speech recognition, and for machine translation we train transformer based neural machine translation models with speech recognition output style text as input. Experiments conducted on the IWSLT 2018 task indicate that, compared to baseline system from KIT, our system achieved 14.9 BLEU improvement.

cs.CL↗
arXiv2018

Is Twisted Bilayer Graphene Stable under Shear?

In twisted bilayer graphene (TBLG), extremely small deviations from the magic twist angle $θ_m{\approx}1.08^\circ$ change its electronic structure near the Fermi level drastically, causing a meV-wide flat band to appear or disappear. In view of such sensitivity to minute structural deformations, we investigate the combined effect of shear and atomic relaxation on the electronic structure. Using precise experimental data for monolayer and bilayer graphene as input in a simplified formalism for the electronic structure and elastic energy, we find TBLG near $θ_m$ to be unstable with respect to global shear by the angle $α{\approx}0.08^\circ$. In TBLG, the effect of shear on the electronic structure is as important as that of atomic relaxation. Under optimum global shear, calculated $θ_m$ is reduced by $0.04^\circ$ and agrees with the observed value.

cond-mat.mes-hall↗
arXiv2018

Microscopic Mechanism of the Helix-to-Layer Transformation in Elemental Group VI Solids

We study the conversion of bulk Se and Te, consisting of intertwined a helices, to structurally very dissimilar, atomically thin two-dimensional (2D) layers of these elements. Our ab initio calculations reveal that previously unknown and unusually stable δ- and η-2D allotropes may form in an intriguing multi-step process that involves a concerted motion of many atoms at dislocation defects. We identify such a complex reaction path involving zipper-like motion of such dislocations that initiate structural changes. With low activation barriers <0.3 eV along the optimum path, the conversion process may occur at moderate temperatures. We find all one-dimensional (1D) and 2D chalcogen structures to be semiconducting.

cond-mat.mes-hall↗
arXiv2018

Quadratic Jahn-Teller effect of fullerene anions

The quadratic Jahn-Teller effect of C$_{60}^{n-}$ ($n=$ 1-5) is investigated from the first principles. Employing the density functional theory calculations with hybrid functional, the quadratic vibronic coupling constants of C$_{60}^-$ were derived. The warping of the adiabatic potential energy surface of C$_{60}^-$ by the quadratic vibronic coupling is estimated about 2 meV, which is much smaller than the Jahn-Teller stabilization energy ($\approx$ 50 meV). Because of the selection rule and the vibronic reduction, the quadratic coupling slightly modifies the vibronic states of C$_{60}$ anions. Particularly, in the case of C$_{60}^{3-}$, parity and symmetry selection rule significantly reduces the effect of quadratic coupling on vibronic states. The present results confirm that the low-energy vibronic dynamics of C$_{60}^{n-}$ is of pseudorotational type.

cond-mat.str-el↗
arXiv2018

Towards room-temperature superconductivity in low-dimensional C60 nanoarrays: An ab initio study

We propose to raise the critical temperature $T_c$ for superconductivity in doped C$_{60}$ molecular crystals by increasing the electronic density of states at the Fermi level $N(E_F)$ and thus the electron-phonon coupling constant in low-dimensional C$_{60}$ nanoarrays. We consider both electron and hole doping and present numerical results for $N(E_F)$, which increases with decreasing bandwidth of the partly filled $h_u$ and $t_{1u}$ derived frontier bands with decreasing coordination number of C$_{60}$. Whereas a significant increase of $N(E_F)$ occurs in 2D arrays of doped C$_{60}$ intercalated in-between graphene layers, we propose that the highest $T_c$ values approaching room temperature may occur in bundles of nanotubes filled by 1D arrays of externally doped C$_{60}$ or La@C$_{60}$, or in diluted 3D crystals, where quasi-1D arrangements of C$_{60}$ form percolation paths.

cond-mat.supr-con↗
arXiv2018

Dynamical Jahn-Teller effect of fullerene anions

The dynamical Jahn-Teller effect of C$_{60}^{n-}$ anions ($n = $ 1-5) is studied using the numerical diagonalization of the linear $p^n \otimes 8d$ Jahn-Teller Hamiltonian with the currently established coupling parameters. It is found that in all anions the Jahn-Teller effect stabilizes the low-spin states, resulting in the violation of Hund's rule. The energy gain due to the Jahn-Teller dynamics is found to be comparable to the static Jahn-Teller stabilization. The Jahn-Teller dynamics influences the thermodynamic properties via strong variation of the density of vibronic states with energy. Thus, the large vibronic entropy in the low-spin states enhances the effective spin gap of C$_{60}^{3-}$ quenching the spin crossover. From the calculations of the effective spin gap in function of the Hund's rule coupling, we found that the latter should amount 40 $\pm$ 5 meV in order to cope with the violation of Hund's rule and to reproduce the large spin gap. With the obtained numerical solutions the matrix elements of electronic operators for the low-lying vibronic levels and the vibronic reduction factors are calculated for all anions.

cond-mat.mtrl-sci↗
arXiv2017

A Novel Way of Identifying Cyber Predators

Recurrent Neural Networks with Long Short-Term Memory cell (LSTM-RNN) have impressive ability in sequence data processing, particularly for language model building and text classification. This research proposes the combination of sentiment analysis, new approach of sentence vectors and LSTM-RNN as a novel way for Sexual Predator Identification (SPI). LSTM-RNN language model is applied to generate sentence vectors which are the last hidden states in the language model. Sentence vectors are fed into another LSTM-RNN classifier, so as to capture suspicious conversations. Hidden state enables to generate vectors for sentences never seen before. Fasttext is used to filter the contents of conversations and generate a sentiment score so as to identify potential predators. The experiment achieves a record-breaking accuracy and precision of 100% with recall of 81.10%, exceeding the top-ranked result in the SPI competition.

cs.CL↗