arXiv subjects

Qin Gao

Publications and source records attributed to Qin Gao.

13 recordsLinked to original sources

arXiv2024

Generating Gender Alternatives in Machine Translation

Machine translation (MT) systems often translate terms with ambiguous gender (e.g., English term "the nurse") into the gendered form that is most prevalent in the systems' training data (e.g., "enfermera", the Spanish term for a female nurse). This often reflects and perpetuates harmful stereotypes present in society. With MT user interfaces in mind that allow for resolving gender ambiguity in a frictionless manner, we study the problem of generating all grammatically correct gendered translation alternatives. We open source train and test datasets for five language pairs and establish benchmarks for this task. Our key technical contribution is a novel semi-supervised solution for generating alternatives that integrates seamlessly with standard MT models and maintains high performance without requiring additional components or increasing inference overhead.

cs.CL↗

arXiv2020

Maximal nilpotent complex structures

Let the pair $(\mathfrak{g},J)$ be a nilpotent Lie algebra $\mathfrak{g}$ (NLA for short) endowed with a nilpotent complex structure $J$. In this paper, motivated by a question in the work of Cordero, Fernández, Gray and Ugarte, we prove that $2\leq ν(J) \leq 3$ for $(\mathfrak{g},J)$ when $ν(\mathfrak{g})=2$, where $ν(\mathfrak{g})$ is the step of $\mathfrak{g}$ and $ν(J)$ is the unique smallest integer such that $\mathfrak{a}(J)_{ν(J)}=\mathfrak{g}$ as in Definition 1 and 8 of the paper by Cordero, Fernández, Gray and Ugarte. When $ν(\mathfrak{g})=3$, for arbitrary $n \geq 3$, there exists a pair $(\mathfrak{g},J)$ such that $ν(J)=\dim_{\mathbb{C}}\mathfrak{g}=n$, for which we call the $J$ in the pair $(\mathfrak{g},J)$, satisfying $ν(J)=\dim_{\mathbb{C}}\mathfrak{g}=n$, a maximal nilpotent (MaxN for short) complex structure. The algebraic dimension of a nilmanifold endowed with a left invariant MaxN complex structure is discussed. Furthermore, a structure theorem is proved for the pair $(\mathfrak{g},J)$, where $ν(\mathfrak{g})=3$ and $J$ is a MaxN complex structure.

math.DG↗

arXiv2017

Phase Diagram of Carbon Nickel Tungsten: Superatom Model

Carbon solubility in face-centered cubic Ni-W alloys and the phase diagram of C-Ni-W are investigated by means of first principle calculations and semi-grand canonical Monte Carlo simulations. With density functional theory (DFT) total energies as fitting data, we build a superatom model for efficient simulation. Multi-histogram analysis is utilized to predict free energies for different compositions and temperatures. By comparing free energies of competing phases, we are able to predict carbon solubility and phase diagrams of C-Ni-W at different temperatures. A simple ideal mixing approximation gives qualitatively similar predictions.

cond-mat.mtrl-sci↗

arXiv2016

Phase Diagram of Boron Carbide With Variable Carbon Composition

Boron carbide exhibits intrinsic substitutional disorder over a broad composition range. The structure consists of 12-atom icosahedra placed at the vertices of a rhombohedral lattice, together with a 3-atom chain along the 3-fold axis. In the high carbon limit, one or two carbons can replace borons on the icosahedra while the chains are primarily of type C-B-C. We fit an interatomic pair interaction model to density functional theory total energies to investigate the substitutional carbon disorder. Monte Carlo simulations with sampling improved by replica exchange and augmented by 2d multiple histogram analysis, predicts three phases. The low temperature, high carbon composition monoclinic Cm "tilted polar" structure disorders through a pair of phase transitions, first via an Ising-like transition to a "bipolar" state with space group C2/m, then via a first order 3-state Potts-like transition to the experimentally observed "nonpolar" \bar{R}3m symmetry.

cond-mat.mtrl-sci↗

arXiv2016

Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation

Neural Machine Translation (NMT) is an end-to-end learning approach for automated translation, with the potential to overcome many of the weaknesses of conventional phrase-based translation systems. Unfortunately, NMT systems are known to be computationally expensive both in training and in translation inference. Also, most NMT systems have difficulty with rare words. These issues have hindered NMT's use in practical deployments and services, where both accuracy and speed are essential. In this work, we present GNMT, Google's Neural Machine Translation system, which attempts to address many of these issues. Our model consists of a deep LSTM network with 8 encoder and 8 decoder layers using attention and residual connections. To improve parallelism and therefore decrease training time, our attention mechanism connects the bottom layer of the decoder to the top layer of the encoder. To accelerate the final translation speed, we employ low-precision arithmetic during inference computations. To improve handling of rare words, we divide words into a limited set of common sub-word units ("wordpieces") for both input and output. This method provides a good balance between the flexibility of "character"-delimited models and the efficiency of "word"-delimited models, naturally handles translation of rare words, and ultimately improves the overall accuracy of the system. Our beam search technique employs a length-normalization procedure and uses a coverage penalty, which encourages generation of an output sentence that is most likely to cover all the words in the source sentence. On the WMT'14 English-to-French and English-to-German benchmarks, GNMT achieves competitive results to state-of-the-art. Using a human side-by-side evaluation on a set of isolated simple sentences, it reduces translation errors by an average of 60% compared to Google's phrase-based production system.

cs.CL↗

arXiv2016

Thickness characterization of atomically-thin WSe$_2$ on epitaxial graphene by low-energy electron reflectivity oscillations

In this work, low-energy electron microscopy is employed to probe structural as well as electronic information in few-layer WSe$_2$ on epitaxial graphene on SiC. The emergence of unoccupied states in the WSe$_2$--graphene heterostructures are studied using spectroscopic low-energy electron reflectivity. Reflectivity minima corresponding to specific WSe$_2$ states that are localized between the monolayers of each vertical heterostructure are shown to reveal the number of layers for each point on the surface. A theory for the origin of these states is developed and utilized to explain the experimentally observed featured in the WSe$_2$ electron reflectivity.

cond-mat.mes-hall↗

arXiv2015

Machine Learning methods for interatomic potentials: application to boron carbide

Total energies of crystal structures can be calculated to high precision using quantum-based density functional theory (DFT) methods, but the calculations can be time consuming and scale badly with system size. Cluster expansions of total energy as a linear superposition of pair, triplet and higher interactions can efficiently approximate the total energies but are best suited to simple lattice structures. To model the total energy of boron carbide, with a complex crystal structure, we explore the utility of machine learning methods ($L_1$-penalized regression, neural network, Gaussian process and support vector regression) that capture certain non-linear effects associated with many-body interactions despite requiring only pair frequencies as input. Our interaction models are combined with Monte Carlo simulations to evaluate the thermodynamics of chemical ordering.

cond-mat.mtrl-sci↗

arXiv2014

Inelastic Effects in Low-Energy Electron Reflectivity of Two-dimensional Materials

A simple method is proposed for inclusion of inelastic effects (electron absorption) in computations of low-energy electron reflectivity (LEER) spectra. The theoretical spectra are formulated by matching of electron wavefunctions obtained from first-principles computations in a repeated vacuum-slab-vacuum geometry. Inelastic effects are included by allowing these states to decay in time in accordance with an imaginary term in the potential of the slab, and by mixing of the slab states in accordance with the same type of distribution as occurs in a free-electron model. LEER spectra are computed for various two-dimensional materials, including free-standing multilayer graphene, graphene on copper substrates, and hexagonal boron nitride (h-BN) on cobalt substrates.

cond-mat.mes-hall↗

arXiv2014

Theory of Graphene-Insulator-Graphene Tunnel Junctions

Graphene-insulator-graphene vertical tunneling structures are discussed from a theoretical perspective. Momentum conservation in such devices leads to highly nonlinear current-voltage characteristics, which with gates on the tunnel junction form potentially useful transistor structures. Two prior theoretical treatments of such devices are discussed; the treatments are shown to be formally equivalent, although some differences in their implementations are identified. The limit of zero momentum conservation in the theory is explicitly considered, with a formula involving the density-of-states of the graphene electrodes recovered in this limit. Various predictions of the theory are compared to experiment.

cond-mat.mes-hall↗

arXiv2013

First Principles Study of Bismuth Films at Transition Metal Grain Boundaries

Recent experiments suggest that Bi impurities segregate to form bilayer films on Ni and Cu grain boundaries but do not segregate in Fe. To explain these phenomena, we study the total energies of Bi films on transition metal (TM) $Σ$3(111) and $Σ$5(012) GBs using density functional theory. Our results agree with the observed stabilities. We propose a model to predict Bi bilayer stability at Ni GBs which suggests that Bi bilayer is not stable on (111) twist CSL GBs but is stable in most (100) twist CSL GBs. We investigate the interaction and bonding character between Bi and TMs to explain the differences among TMs based on localization of orbitals and magnetism.

cond-mat.mes-hall↗

arXiv2013

Low-Energy Electron Reflectivity of Graphene on Copper and other Substrates

The reflectivity of low energy electrons from graphene on copper substrates is studied both experimentally and theoretically. Well-known oscillations in the reflectivity of electrons with energies 0 - 8 eV above the vacuum level are observed in the experiment. These oscillations are reproduced in theory, based on a first-principles density functional description of interlayer states forming for various thicknesses of multilayer graphene. It is demonstrated that n layers of graphene produce a regular series of n-1 minima in the reflectance spectra, together with a possible additional minimum associated with an interlayer state forming between the graphene and the substrate. Both (111) and (001) orientations of the copper substrates are studied. Similarities in their reflectivity spectra arise from the interlayer states, whereas differences are found because of the different Cu band structures along those orientations. Results for graphene on other substrates, including Pt(111) and Ir(111), are also discussed.

cond-mat.mes-hall↗

arXiv2013

First Principles Study of Bismuth Films on the Nickel(111) Surface

A recent experiment(Bollmann11) suggested that bismuth forms hexagonal close packed (HCP) films on the Ni(111) surface, of heights 3, 5 and 7 layers. A quantum size effect based on free electrons was proposed in explanation. To test this idea, we calculate the total energies of Bi on the Ni(111) surface using density functional theory. We find that HCP film stabilities disagree with the observed odd layer preferences, and the structures are mechanically destabilized by adding capping atoms which pucker the HCP layers. Furthermore, we find that rhombohedral films based on the bulk Bi structure are energetically more favorable than the proposed HCP films. These structures also favor odd numbers of layers, but owing to covalent chemical bonding rather than confinement of free electrons. Specifically, a strongly bound adsorbed surface monolayer forms, followed by bulk-like rhombohedral bilayers.

cond-mat.mes-hall↗

arXiv2013

Low-energy Electron Reflectivity from Graphene

Low-energy reflectivity of electrons from single- and multi-layer graphene is examined both theoretically and experimentally. A series of minima in the reflectivity over the energy range of 0 - 8 eV are found, with the number of minima depending on the number of graphene layers. Using first-principles computations, it is demonstrated that a free standing n-layer graphene slab produces n-1 reflectivity minima. This same result is also found experimentally for graphene supported on SiO2. For graphene bonded onto other substrates it is argued that a similar series of reflectivity minima is expected, although in certain cases an additional minimum occurs, at an energy that depends on the graphene-substrate separation and the effective potential in that space.

cond-mat.mes-hall↗