arXiv subjects

Siyang Wang

Publications and source records attributed to Siyang Wang.

At least 55 records · Page 3Linked to original sources

arXiv2021

Integrated Speech and Gesture Synthesis

Text-to-speech and co-speech gesture synthesis have until now been treated as separate areas by two different research communities, and applications merely stack the two technologies using a simple system-level pipeline. This can lead to modeling inefficiencies and may introduce inconsistencies that limit the achievable naturalness. We propose to instead synthesize the two modalities in a single model, a new problem we call integrated speech and gesture synthesis (ISG). We also propose a set of models modified from state-of-the-art neural speech-synthesis engines to achieve this goal. We evaluate the models in three carefully-designed user studies, two of which evaluate the synthesized speech and gesture in isolation, plus a combined study that evaluates the models like they will be used in real-world applications -- speech and gesture presented together. The results show that participants rate one of the proposed integrated synthesis models as being as good as the state-of-the-art pipeline system we compare against, in all three tests. The model is able to achieve this with faster synthesis time and greatly reduced parameter count compared to the pipeline system, illustrating some of the potential benefits of treating speech and gesture synthesis together as a single, unified problem. Videos and code are available on our project page at https://swatsw.github.io/isg_icmi21/

cs.HC↗

arXiv2021

A conservative and energy stable discontinuous spectral element method for the shifted wave equation in second order form

In this paper, we develop a provably energy stable and conservative discontinuous spectral element method for the shifted wave equation in second order form. The proposed method combines the advantages and central ideas of very successful numerical techniques, the summation-by-parts finite difference method, the spectral method and the discontinuous Galerkin method. We prove energy-stability, discrete conservation principle, and derive error estimates in the energy norm for the (1+1)-dimensions shifted wave equation in second order form. The energy-stability results, discrete conservation principle, and the error estimates generalise to multiple dimensions using tensor products of quadrilateral and hexahedral elements. Numerical experiments, in (1+1)-dimensions and (2+1)-dimensions, verify the theoretical results and demonstrate optimal convergence of $L^2$ numerical errors at subsonic, sonic and supersonic regimes.

math.NA↗

arXiv2021

Elastic wave propagation in curvilinear coordinates with mesh refinement interfaces by a fourth order finite difference method

We develop a fourth order accurate finite difference method for the three dimensional elastic wave equation in isotropic media with the piecewise smooth material property. In our model, the material property can be discontinuous at curved interfaces. The governing equations are discretized in second order form on curvilinear meshes by using a fourth order finite difference operator satisfying a summation-by-parts property. The method is energy stable and high order accurate. The highlight is that mesh sizes can be chosen according to the velocity structure of the material so that computational efficiency is improved. At the mesh refinement interfaces with hanging nodes, physical interface conditions are imposed by using ghost points and interpolation. With a fourth order predictor-corrector time integrator, the fully discrete scheme is energy conserving. Numerical experiments are presented to verify the fourth order convergence rate and the energy conserving property.

math.NA↗

arXiv2020

Summation-by-parts approximations of the second derivative: Pseudoinverses of singular operators and revisiting the sixth order accurate narrow-stencil operator

We consider finite difference approximations of the second derivative, exemplified in Poisson's equation, the heat equation and the wave equation. The finite difference operators satisfy a summation-by-parts property, which mimics the integration-by-parts. Since the operators approximate the second derivative, they are singular by construction. To impose boundary conditions, these operators are modified using Simultaneous Approximation Terms. This makes the modified matrices non-singular, for most choices of boundary conditions. Recently, inverses of such matrices were derived. However, when considering Neumann boundary conditions on both boundaries, the modified matrix is still singular. For such matrices, we have derived an explicit expression for the Moore-Penrose pseudoinverse, which can be used for solving elliptic problems and some time-dependent problems. The condition for this new pseudoinverse to be valid, is that the modified matrix does not have more than one zero eigenvalue. We have reconstructed the sixth order accurate narrow-stencil operator with a free parameter and show that more than one zero eigenvalue can occur. We have performed a detailed analysis on the free parameter to improve the properties of the second derivative operator. We complement the derivations by numerical experiments to demonstrate the improvements of the new second derivative operator.

math.NA↗

arXiv2020

Theory of the Chromatic Dispersion, Revisited

We derive general analytic expressions for the chromatic dispersion orders valid to infinity, due to the k vector or phase ϕ dependence on the wavelength. Additionally, we identify polynomials and recursion relations associated with the chromatic dispersion orders and draw analogy to the generalized Lah and Laguerre transformations. Further, we give explicitly the dispersion terms to the 10th order and visualize the chromatic dispersion for material, grating and prism-pair compressors and hollow-core photonic anti-resonant fiber. These simple formulas are applicable for material dispersion, compressors, stretchers, waveguides, and any other type of known frequency-dependent phase.

physics.optics↗

arXiv2020

Hydride growth mechanism in Zircaloy-4: investigation of the partitioning of alloying elements

The long-term safety of water-based nuclear reactors relies in part on the reliability of zirconium-based nuclear fuel. Yet the progressive ingress of hydrogen during service makes zirconium alloys subject to delayed hydride cracking. Here, we use a combination of electron back-scattered diffraction and atom probe tomography to investigate specific microstructural features from the as-received sample and in the blocky-alpha microstructure, before and after electrochemical charging with hydrogen or deuterium followed by a low temperature heat treatment at 400C for 5 hours followed by furnace cooling at a rate of 0. 5C per min. Specimens for atom probe were prepared at cryogenic temperature to avoid the formation of spurious hydrides. We report on the compositional evolution of grains and grain boundaries over the course of the sample's thermal history, as well as the ways the growth of the hydrides modifies locally the composition and the structure of the alloy. We observe a significant amount of deuterium left in the matrix, even after the slow cooling and growth of the hydrides. Stacking faults form ahead of the growth front and Sn segregates at the hydride-matrix interface and on these faults. We propose that this segregation may facilitate further growth of the hydride. Our systematic investigation enables us discuss how the solute distribution affects the evolution of the alloy's properties during its service lifetime.

cond-mat.mtrl-sci↗

arXiv2020

Slip-hydride interactions in Zircaloy-4: Multiscale mechanical testing and characterisation

The interactions between δ-hydrides and plastic slip in a commercial zirconium alloy, Zircaloy-4, under stress were studied using in situ secondary electron microscope (SEM) micropillar compression tests of single crystal samples and ex situ digital image correlation (DIC) macroscale tensile tests of polycrystalline samples. The hydrides decorate near basal planes in orientation, and for micropillars orientated for basal slip localised shear at the hydride-matrix interface is favoured over slip in α-Zr matrix due to a lower shear stress required. In contrast, for pillars oriented for prismatic slip the shear stress needed to trigger plastic slip within the hydride is slightly higher than the critical resolved shear stress (CRSS) for the prismatic slip system. In this case, slip in the hydride is likely achieved through <110>-type shear which is parallel to the activated -type shear in the parent matrix. At a longer lengthscale, these results are used to inform polycrystalline samples analysed using high spatial resolution DIC. Here localised interface shear remains to be a significant deformation path which can both cause and be caused by matrix slip on planes closely-oriented to the phase boundaries. Matrix slip on planes nearly perpendicular to the adjacent hydride-matrix interfaces can either result in plastic slip within the hydrides or get arrested at the interfaces, generating local stress concentration. Through these mechanisms, the presence of δ-hydrides leads to enhanced strain localisation in Zircaloy-4 early in the plastic regime.

cond-mat.mtrl-sci↗

arXiv2020

Numerical upscaling for heterogeneous materials in fractured domains

We consider numerical solution of elliptic problems with heterogeneous diffusion coefficients containing thin highly conductive structures. Such problems arise e.g. in fractured porous media, reinforced materials, and electric circuits. The main computational challenge is the high resolution needed to resolve the data variation. We propose a multiscale method that models the thin structures as interfaces and incorporate heterogeneities in corrected shape functions. The construction results in an accurate upscaled representation of the system that can be used to solve for several forcing functions or to simulate evolution problems in an efficient way. By introducing a novel interpolation operator, defining the fine scale of the problem, we prove exponential decay of the shape functions which allows for a sparse approximation of the upscaled representation. An a priori error bound is also derived for the proposed method together with numerical examples that verify the theoretical findings. Finally we present a numerical example to show how the technique can be applied to evolution problems.

math.NA↗

arXiv2020

Bridging physics-based and equivalent circuit models for lithium-ion batteries

In this article, a novel implementation of a widely used pseudo-two-dimensional (P2D) model for lithium-ion battery simulation is presented with a transmission line circuit structure. This implementation represents an interplay between physical and equivalent circuit models. The discharge processes of an NMC-graphite lithium-ion battery under different currents are simulated, and it is seen the results from the circuit model agree well with the results obtained from a physical simulation carried out in COMSOL Multiphysics, including both terminal voltage and concentration distributions. Finally we demonstrated how the circuit model can contribute to the understanding of the cell electrochemistry, exemplified by an analysis of the overpotential contributions by various processes.

physics.chem-ph↗

arXiv2019

Unaligned Image-to-Sequence Transformation with Loop Consistency

We tackle the problem of modeling sequential visual phenomena. Given examples of a phenomena that can be divided into discrete time steps, we aim to take an input from any such time and realize this input at all other time steps in the sequence. Furthermore, we aim to do this without ground-truth aligned sequences -- avoiding the difficulties needed for gathering aligned data. This generalizes the unpaired image-to-image problem from generating pairs to generating sequences. We extend cycle consistency to loop consistency and alleviate difficulties associated with learning in the resulting long chains of computation. We show competitive results compared to existing image-to-image techniques when modeling several different data sets including the Earth's seasons and aging of human faces.

cs.CV↗

arXiv2019

Fourth order finite difference methods for the wave equation with mesh refinement interfaces

We analyze two types of summation-by-parts finite difference operators for approximating the second derivative with variable coefficient. The first type uses ghost points, while the second type does not use any ghost points. A previously unexplored relation between the two types of summation-by-parts operators is investigated. By combining them we develop a new fourth order accurate finite difference discretization with hanging nodes on the mesh refinement interface. We take the model problem as the two-dimensional acoustic wave equation in second order form in terms of acoustic pressure, and prove energy stability for the proposed method. Compared to previous approaches using ghost points, the proposed method leads to a smaller system of linear equations that needs to be solved for the ghost point values. Another attractive feature of the proposed method is that the explicit time step does not need to be reduced relative to the corresponding periodic problem. Numerical experiments, both for smoothly varying and discontinuous material properties, demonstrate that the proposed method converges to fourth order accuracy. A detailed comparison of the accuracy and the time-step restriction with the simultaneous-approximation-term penalty method is also presented.

math.NA↗

arXiv2019

Attention is all you need for Videos: Self-attention based Video Summarization using Universal Transformers

Video Captioning and Summarization have become very popular in the recent years due to advancements in Sequence Modelling, with the resurgence of Long-Short Term Memory networks (LSTMs) and introduction of Gated Recurrent Units (GRUs). Existing architectures extract spatio-temporal features using CNNs and utilize either GRUs or LSTMs to model dependencies with soft attention layers. These attention layers do help in attending to the most prominent features and improve upon the recurrent units, however, these models suffer from the inherent drawbacks of the recurrent units themselves. The introduction of the Transformer model has driven the Sequence Modelling field into a new direction. In this project, we implement a Transformer-based model for Video captioning, utilizing 3D CNN architectures like C3D and Two-stream I3D for video extraction. We also apply certain dimensionality reduction techniques so as to keep the overall size of the model within limits. We finally present our results on the MSVD and ActivityNet datasets for Single and Dense video captioning tasks respectively.

cs.CV↗

arXiv2019

The effect of δ-hydride on the micromechanical deformation of Zircaloy-4 studied by in situ high angular resolution electron backscatter diffraction

Zircaloy-4 is used extensively as nuclear fuel cladding materials and hydride embrittlement is a major failure mechanism. To explore the effect of δ-hydride on plastic deformation and performance of Zircaloy-4, in situ high angular resolution electron backscatter diffraction (HR-EBSD) was used to quantify stress and geometrically necessary dislocation (GND) density during bending tests of hydride-free and hydride-containing single crystal Zircaloy-4 microcantilevers. Results suggest that while the stress applied was accommodated by plastic slip in the hydride-free cantilever, the hydride-containing cantilever showed precipitation induced GND pile-up at hydride-matrix interface pre-deformation, and considerable locally increasing GND density under tensile stress upon plastic deformation.

cond-mat.mtrl-sci↗

arXiv2019

An Energy Based Discontinuous Galerkin Method for Coupled Elasto-Acoustic Wave Equations in Second Order Form

We consider wave propagation in a coupled fluid-solid region, separated by a static but possibly curved interface. The wave propagation is modeled by the acoustic wave equation in terms of a velocity potential in the fluid, and the elastic wave equation for the displacement in the solid. At the fluid solid interface, we impose suitable interface conditions to couple the two equations. We use a recently developed, energy based discontinuous Galerkin method to discretize the governing equations in space. Both energy conserving and upwind numerical fluxes are derived to impose the interface conditions. The highlights of the developed scheme include provable energy stability and high order accuracy. We present numerical experiments to illustrate the accuracy property and robustness of the developed scheme.

math.NA↗

arXiv2019

Quantification challenges for atom probe tomography of hydrogen and deuterium in Zircaloy-4

Analysis and understanding of the role of hydrogen in metals is a significant challenge for the future of materials science, and this is a clear objective of recent work in the atom probe tomography (APT) community. Isotopic marking by deuteration has often been proposed as the preferred route to enable quantification of hydrogen by APT. Zircaloy-4 was charged electrochemically with hydrogen and deuterium under the same conditions to form large hydrides and deuterides. Our results from a Zr hydride and a Zr deuteride highlight challenges associated to accurate quantification of hydrogen and deuterium, in particular associated to the overlap of peaks at low mass-to-charge ratio and of hydrogen/deuterium containing molecular ions. We discuss possible ways to ensure that appropriate information is extracted from APT analysis of hydrogen in a zirconium alloy systems that is important for nuclear power.

cond-mat.mtrl-sci↗

arXiv2018

Microstructure and formation mechanisms of δ-hydrides in variable grain size Zircaloy-4 studied by electron backscatter diffraction

Microstructure and crystallography of δ phase hydrides in as-received fine grain and 'blocky' alpha large grain Zircaloy-4 (average grain size ~11 μm and >200 μm, respectively) were examined using electron backscatter diffraction. Results suggest that the the matrix-hydride orientation relationship is {0001}α||{111}δ;<11-20>α||<110>δ for all the cases studied. The habit plane of intragranular hydrides and some intergranular hydrides has been found to be {10-17} of the surrounding matrix. The morphology of intergranular hydrides can vary depending upon the angle between the grain boundary and the hydride habit plane. The misfit strain between α-Zr and δ-hydride is accommodated by high density of dislocations and twin structures in the hydrides, and a mechanism of twin formation in the hydrides has been proposed. The growth of hydrides across grain boundaries is achieved through an auto-catalytic manner similar to the growth pattern of intragranular hydrides. Easy collective shear along <1-100> makes it possible for hydride nucleation at any grain boundaries, while the process seems to favour grain boundaries with low (<40°) and high (>80°) c-axis misorientation angles. Moreover, the angle between the grain boundary and the adjacent basal planes does not influence the propensity for hydride nucleation.

cond-mat.mtrl-sci↗

arXiv2018

Controllable Top-down Feature Transformer

We study the intrinsic transformation of feature maps across convolutional network layers with explicit top-down control. To this end, we develop top-down feature transformer (TFT), under controllable parameters, that are able to account for the hidden layer transformation while maintaining the overall consistency across layers. The learned generators capture the underlying feature transformation processes that are independent of particular training images. Our proposed TFT framework brings insights to and helps the understanding of, an important problem of studying the CNN internal feature representation and transformation under the top-down processes. In the case of spatial transformations, we demonstrate the significant advantage of TFT over existing data-driven approaches in building data-independent transformations. We also show that it can be adopted in other applications such as data augmentation and image style transfer.

cs.CV↗

arXiv2018

The effect of cooling rate and grain size on hydride formation in Zircaloy-4

We explore the distribution, morphology and structure of zirconium hydrides formed using different cooling rates through the solid state Zr+[H] --> Zr + hydride transus, in fine and blocky alpha Zircaloy-4. We observe that cooling rate and grain size control the phase and distribution of hydrides. The blocky alpha (coarse grain, > 200 micrometer) Zircaloy-4, has a smaller grain boundary area to grain volume ratio and this significantly affects nucleation and growth of hydrides as compared to fine grain size (~11 micrometer) material.

cond-mat.mtrl-sci↗