Searcharxiv⌕ Search

arXiv subjects

Li Wan

Publications and source records attributed to Li Wan.

51 records · Page 3Linked to original sources

Attention-Based Models for Text-Dependent Speaker Verification

Attention-based models have recently shown great performance on a range of tasks, such as speech recognition, machine translation, and image captioning due to their ability to summarize relevant information that expands through the entire length of an input sequence. In this paper, we analyze the usage of attention mechanisms to the problem of sequence summarization in our end-to-end text-dependent speaker recognition system. We explore different topologies and their variants of the attention layer, and compare different pooling methods on the attention weights. Ultimately, we show that attention-based models can improves the Equal Error Rate (EER) of our speaker verification system by relatively 14% compared to our non-attention LSTM baseline model.

eess.AS↗

Links: A High-Dimensional Online Clustering Method

We present a novel algorithm, called Links, designed to perform online clustering on unit vectors in a high-dimensional Euclidean space. The algorithm is appropriate when it is necessary to cluster data efficiently as it streams in, and is to be contrasted with traditional batch clustering algorithms that have access to all data at once. For example, Links has been successfully applied to embedding vectors generated from face images or voice recordings for the purpose of recognizing people, thereby providing real-time identification during video or audio capture.

stat.ML↗

Strong convergence rate in averaging principle for stochastic hyperbolic-parabolic equations with two time-scales

In this article, we investigate averaging principle for stochastic hyperbolic-parabolic equations with two time-scales, in which both the slow and fast components are perturbed by multiplicative noises. Particularly, we prove that the rate of strong convergence for the slow component to the averaged dynamics is of order $1/2$, which significantly improves the order $1/4$ established in our previous work.

math.PR↗

Effective slip length of confined fluid with wavy fluid-solid interfaces

We have studied the effective slip length of confined fluid with nano-structured fluid-solid interfaces. A formula bridging the effective slip length and nano structures of the interfaces has been obtained analytically. For the application of the formula, a confined fluid with wavy fluid-solid interfaces has been studied as an example. In the study, the criteria of the formula have been detailed, and verified by the finite element method.

physics.flu-dyn↗

Large Deviations for Chiral Transition through Path Integral

An noise-induced mechanism has been revealed by some authors recently for the homochirality in a chiral system. Motivated by such stochastic process, we study the noise-induced transition in the system. The chiral transition, say the transition between chiral states, is impossible in the deterministic view if the homochirality has been converged, but is a rare event when noise is involved. We study the rare event by using the large deviation theory (LDT) and figure out the LDT rate functional for the transition probability through the Doi-Peliti second quantization path integral method. In order to check the correctness of our work, we have derived the Langevin equations through the path integral method by using the Hubbard-Stratonovich transformation, and recover the equations reported by other authors through a different method.

cond-mat.stat-mech↗

Weak order in averaging principle for stochastic wave equations with a fast oscillation

This article deals with the weak errors for averaging principle for a stochastic wave equation in a bounded interval $[0,L]$, perturbed by a oscillating term arising as the solution of a stochastic reaction-diffusion equation evolving with respect to the fast time. Under suitable conditions, it is proved that the rate of weak convergence to the averaged effective dynamics is of order $1$ via an asymptotic expansion approach.

math.PR↗

Slip length of confined liquid with small roughness of solid-liquid interfaces

We have studied the slip length of confined liquid with small roughness of solid-liquid interfaces. Dyadic Green function and perturbation expansion have been applied to get the slip length quantitatively. In the slip length, both effects of the roughness of the interfaces and the chemical interaction between the liquid and the solid surface are involved. For the numerical calculation, Monte Carlo method has been used to simulate the rough interfaces and the physical quantities are obtained statistically over the interfaces. Results show that the total slip length of the system is linearly proportional to the slip length contributed from the chemical interaction. And the roughness of the interfaces plays its role as the proportionality factor. For the roughness, the variance of the roughness decreases the total slip length while the correlation length of the roughness can enhance the slip length dramatically to a saturation value.

physics.flu-dyn↗

Iso-electric point of fluid

Iso-electric point(IEP) is the PH, at which the $ζ$ potential is measured to be zero. The occurrence of IEP has been understood due to the neutralization of surface charge density (SCD) at the solid-liquid interface. In this work, we use the potential trap model to study the sources of the surface charge density at verious PC and PH, by taking the water-silica system as an example. It is revealed that in the case of $PH<8$, the SCD is mainly originated from the dissociation of water molecules. And the bulk ions trapped at the interface can dominate the SCD when $PH>9$. Due to the mass action law, the dissociation of water molecules is suppressed at the PH close to IEP, leading to a zero surface charge density. In this way, zero $ζ$ potential is obtained at the IEP. It has also been obtained that the increase of the salt concentration in the water can decrease the $ζ$ potential, but increase the surface charge density.

cond-mat.soft↗

Finite Size Effects of Thermal Conductivity for One-Dimensional Mesoscopic Systems

The finite size effects of the thermal conductivity $κ$ have been studied in the phonon space. It is found that only a few phonon modes are selected to take part in the thermal transport when the size $L$ of the system is decreased. The amount of the selected phonon modes is proportional to the $L$. In this way, $κ$ decreases with the decreasing of $L$. Such mechanism for the size effect of $κ$ found in this work is beyond the Phonon-Boundary scattering. The exponent $α$ of the power law $κ\sim L^α$ has been fitted, showing that the exponent is not universal.

cond-mat.mes-hall↗

Quantum noise theory for phonon transport through nanostructures

We have developed a quantum noise approach to study the phonon transport through nanostructures. The nanostructures acting as phonon channels are attached to two phonon reservoirs. And the temperature drop between the two reservoirs drives the phonon transport through the channels. We have derived a quantum Langevin equation(QLE) to describe the phonon transport with the quantum noise originated from the thermal fluctuation of the reservoirs. Within the Markov approximation, the QLE is used to get the thermal conductivity $κ$ of the nanostructures and the finite size effect of the $κ$ then is studied. In this study, the advantage of the quantum noise approach lays on the fact that no any local temperature needs to be defined for the nanostructures in its non-equilibrium state.

cond-mat.mes-hall↗

The Poisson Boltzmann equation and the charge separation phenomenon at the silica-water interface: A holistic approach

The Poisson Boltzmann equation is known for its success in describing the Debye layer that arises from the charge separation phenomenon at the silica/water interface. However, by treating only the mobile ionic charges in the liquid, the Poisson Boltzmann equation accounts for only half of the electrical double layer, with the other half, the surface charge layer, being beyond its computational domain. In this work, we take a holistic approach to the charge separation phenomenon at the silica/water interface by treating, within a single computational domain, the electrical double layer that comprises both the mobile ions in the liquid and the surface charge density. The Poisson Nernst Planck equations are used as the rigorous basis for our methodology. The holistic approach has the advantage of being able to predict surface charge variations that arise either from the addition of salt and acid to the liquid, or from the decrease of the liquid channel width to below twice the Debye length. As the electrical double layer must be overall neutral, we use this constraint to derive both the form of the static limit of the Poisson Nernst Planck equations, as well as a global chemical potential that replaces the classical zeta potential as the boundary value for the PB equation, which can be re-derived from our formalism. We present several predictions of our theory that are beyond the framework of the PB equation alone, e.g., the surface capacitance and the so-called pK and pL values, the isoelectronic point at which the surface charge layer is neutralized, and the appearance of a Donnan potential that arises from the formation of an electrical double layer at the inlet regions of a nano-channel connected to the bulk reservoir. All theory predictions are shown to be in good agreement with the experimental observations.

cond-mat.soft↗

End-to-End Integration of a Convolutional Network, Deformable Parts Model and Non-Maximum Suppression

Deformable Parts Models and Convolutional Networks each have achieved notable performance in object detection. Yet these two approaches find their strengths in complementary areas: DPMs are well-versed in object composition, modeling fine-grained spatial relationships between parts; likewise, ConvNets are adept at producing powerful image features, having been discriminatively trained directly on the pixels. In this paper, we propose a new model that combines these two approaches, obtaining the advantages of each. We train this model using a new structured loss function that considers all bounding boxes within an image, rather than isolated object instances. This enables the non-maximal suppression (NMS) operation, previously treated as a separate post-processing stage, to be integrated into the model. This allows for discriminative training of our combined Convnet + DPM + NMS model in end-to-end fashion. We evaluate our system on PASCAL VOC 2007 and 2011 datasets, achieving competitive results on both benchmarks.

cs.CV↗

A new method to find full complex roots of a complex dispersion equation for light propagation

A new numerical method is presented to find full complex roots of a complex dispersion equation. For the application of the solution, the complex dispersion equation of a cylindrical metallic nanowire is investigated. By using this method, locus of Brewster angle, complex dispersion curves of Surface Plasmon Polaritons (SPPs) and complex bulk modes can be obtained in once calculation. Approximate analytical solution to the complex dispersion equation has also been derived to verify our method.

physics.comp-ph↗

Analytical solutions to zeroth-order dispersion relations of a cylindrical metallic nanowire

Zeroth-order complex dispersion relations of a cylindrical metallic nanowire have been solved out analytically with approximate methods. The analytical solutions are valid for the sections of the dispersion relations whose frequencies are close to the Surface Plasmon frequency. The back bending of the Surface Plasmon-Polaritons(SPPs) can be well described by the analytical solutions, confirming that the back bending is originated from the metal Ohmic loss. The utility of the back bending point in the dispersion relation for the measurement of the metallic Ohimc loss has also been suggested.

cond-mat.mes-hall↗