Xin Qian, Ziyi Zhong, Jieli Zhou
Multimodal machine translation is one of the applications that integrates computer vision and language processing. It is a unique task given that in the field of machine translation, many state-of-the-arts algorithms still only employ textual information. In this work, we explore the effectiveness of reinforcement learning in multimodal machine translation. We present a novel algorithm based on the Advantage Actor-Critic (A2C) algorithm that specifically cater to the multimodal machine translation task of the EMNLP 2018 Third Conference on Machine Translation (WMT18). We experiment our proposed algorithm on the Multi30K multilingual English-German image description dataset and the Flickr30K image entity dataset. Our model takes two channels of inputs, image and text, uses translation evaluation metrics as training rewards, and achieves better results than supervised learning MLE baseline models. Furthermore, we discuss the prospects and limitations of using reinforcement learning for machine translation. Our experiment results suggest a promising reinforcement learning solution to the general task of multimodal sequence to sequence learning.
Xin Qian, Diego Klabjan
Neural network pruning techniques reduce the number of parameters without compromising predicting ability of a network. Many algorithms have been developed for pruning both over-parameterized fully-connected networks (FCNs) and convolutional neural networks (CNNs), but analytical studies of capabilities and compression ratios of such pruned sub-networks are lacking. We theoretically study the performance of two pruning techniques (random and magnitude-based) on FCNs and CNNs. Given a target network {whose weights are independently sampled from appropriate distributions}, we provide a universal approach to bound the gap between a pruned and the target network in a probabilistic sense. The results establish that there exist pruned networks with expressive power within any specified bound from the target network.
Qian Xin, Steffen Duhm, Fabio Bussolotti, Kouki Akaike, Yoshihiro Kubozono, Hideo Aoki, Taichi Kosugi, Satoshi Kera, Nobuo Ueno
We have experimentally revealed the band structure and the surface Brillouin zone of insulating picene single crystals (SCs), the mother organic system for a recently discovered aromatic superconductor, with ultraviolet photoelectron spectroscopy (UPS) and low-energy electron diffraction with laser for photoconduction. A hole effective mass of 2.24 m_0 and the hole mobility mu_h >= 9.0 cm^2/Vs (298 K) were deduced in Gamma-Y direction. We have further shown that some picene SCs did not show charging during UPS even without the laser, which indicates that pristine UPS works for high-quality organic SCs.
Xin Qian
The Daya Bay experiment was designed to be the largest and the deepest underground among the many current-generation reactor antineutrino experiments. With functionally identical detectors deployed at multiple baselines, the experiment aims to achieve the most precise measurement of $\sin^2 2θ_{13}$. The antineutrino rates measured in the two near experimental halls are used to predict the rate at the far experimental hall (average distance of 1648 m from the reactors), assuming there is no neutrino oscillation. The ratio of the measured over the predicted far-hall antineutrino rate is then used to constrain the $\sin^2 2θ_{13}$. The relative systematic uncertainty on this ratio is expected to be 0.2$\sim$0.4%. In this talk, we present an improved measurement of the electron antineutrino disappearance at Daya Bay. With data of 139 days, the deficit of the antineutrino rate in the far experimental hall was measured to be 0.056 $\pm$ 0.007 (stat.) $\pm$ 0.003 (sys.). In the standard three-neutrino framework, the $\sin^2 2 θ_{13}$ was determined to be 0.089 $\pm$ 0.011 at the 68% confidence level in a rate-only analysis.
Xin Qian, Ryan A. Rossi, Fan Du, Sungchul Kim, Eunyee Koh, Sana Malik, Tak Yeon Lee, Joel Chan
Visualization recommendation seeks to generate, score, and recommend to users useful visualizations automatically, and are fundamentally important for exploring and gaining insights into a new or existing dataset quickly. In this work, we propose the first end-to-end ML-based visualization recommendation system that takes as input a large corpus of datasets and visualizations, learns a model based on this data. Then, given a new unseen dataset from an arbitrary user, the model automatically generates visualizations for that new dataset, derive scores for the visualizations, and output a list of recommended visualizations to the user ordered by effectiveness. We also describe an evaluation framework to quantitatively evaluate visualization recommendation models learned from a large corpus of visualizations and datasets. Through quantitative experiments, a user study, and qualitative analysis, we show that our end-to-end ML-based system recommends more effective and useful visualizations compared to existing state-of-the-art rule-based systems. Finally, we observed a strong preference by the human experts in our user study towards the visualizations recommended by our ML-based system as opposed to the rule-based system (5.92 from a 7-point Likert scale compared to only 3.45).
Xin Qian
We say that a metric space $X$ is $(ε,G)$-homogeneous if $G<Iso(X)$ is a discrete group of isometries with $diam(X/G)<ε$.\ A sequence of $(ε_i,G_i)$-homogeneous spaces $X_i$ with $ε_i\to0$ is called a sequence of almost homogeneous spaces. In this paper we show that the Gromov-Hausdorff limit of a sequence of almost homogeneous RCD$(K,N)$ spaces must be a nilpotent Lie group with $Ric\geqslant K$. We also obtain a topological rigidity theorem for $(ε,G)$-homogeneous RCD$(K,N)$ spaces, which generalizes a recent result by Wang. Indeed, if $X$ is an $(ε,G)$-homogeneous RCD$(K,N)$ space and $G$ is an almost-crystallographic group, then $X/G$ is bi-Hölder to an infranil orbifold. Moreover, we study $(ε,G)$-homogeneous spaces in the smooth setting and prove rigidity and $ε$-regularity theorems for Riemannian orbifolds with Einstein metrics and bounded Ricci curvatures respectively.
Te-Huan Liu, Tianyu Wang, Jun Zhou, Xin Qian, Ronggui Yang
Nonequilibrium multi-carrier thermal transport is essential for both scientific research and technological applications in electronic, spintronic, and energy conversion devices. This article reviews the fundamentals of phonon, electron, spin, and ion transport driven by temperature gradients in solid-state and soft condensed matters, and the microscopic interactions between energy/charge carriers that can be leveraged for manipulating electrical and thermal transport in energy conversion devices, such as electron-phonon coupling, spin-phonon interaction, and ion-solvent interactions, etc. In coupled electron-phonon transport, we discuss the basics of electron-phonon interactions and their effects on phonon dynamics, thermalization, and nonequilibrium thermal transport. For the phonon-spin interaction, nonequilibrium transport formulation is introduced first, followed by the physics of spin thermoelectric effect and strategies to manipulate them. Contributions to thermal conductivity from magnons as heat carriers are also reviewed. For coupled transport of heat and ions/molecules, we highlight the importance of local molecular configurations that determine the magnitude of the electrochemical gradient, which is the key to improving the efficiency of low-grade heat energy conversion.
Xin Qian, Chuang Zhang, Te-Huan Liu, Ronggui Yang
Hydrodynamic second sound can be generated by heat pulses when the phonon-phonon interaction is dominantly momentum conserving, and the propagation of the temperature field becomes wavelike rather than diffusive. While the Boltzmann transport equation (BTE) has been widely applied to study phonon dynamics and thermal transport at the nanoscale, modeling the hydrodynamic transport regime remains challenging. The widely used relaxation time approximation (RTA) treats all phonon interactions as resistive without considering momentum conservation, resulting in the absence of phonon hydrodynamics. Rigorously solving BTE by inverting the full scattering matrix, however, is extremely computationally demanding and has been only applied to model one-dimensional temperature variations. Here, we present an analytical Green's function formalism for solving multidimensional Boltzmann transport equation (BTE) using phonon properties from first-principles calculations. This formalism involves Callaway's scattering approximation with separate relaxation times for momentum-conserving and momentum-destroying scattering events. The Green's function captures phonon dynamics in a wide range of temperature, spatial, and temporal scales, and successfully reproduces the transition from ballistic, hydrodynamic, to diffusive transport regimes. Our method avoids the computationally demanding inversion of large scattering matrices and shows good accuracies in predicting the temperature oscillation in ultrafast pump-probe characterizations with different geometries of thermal excitation.
Xin Qian
In this paper, we study some structure properties on the (revised) fundamental group of RCD(0,N) spaces. Our main result generalizes earlier work of Sormani on Riemannian manifolds with nonnegative Ricci curvature and small linear diameter growth. We prove that the revised fundamental group is finitely generated if assuming small linear diameter growth on RCD(0,N) spaces.
Yu Pang, Jinjin Liu, Zeyu Xiang, Xuanhui Fan, Jie Zhu, Zhiwei Wang, Yugui Yao, Xin Qian, Ronggui Yang
This work reports the thermal conductivity of RbV3Sb5 and CsV3Sb5 with three-dimensional charge density wave phase transitions from 80 K to 400 K measured by pump-probe thermoreflectance techniques. The in-plane (basal plane) thermal conductivities are found moderate, i.e., 12 W/mK of RbV3Sb5 and 8.8 W/mK of CsV3Sb5 at 300 K. Low cross-plane (stacking direction) thermal conductivities are observed, with 0.72 W/mK of RbV3Sb5 and 0.49 W/mK of CsV3Sb5 at 300 K. A unique glass-like temperature dependence in the cross-plane thermal conductivity is observed, which decreases monotonically even lower than the Cahill-Pohl limit as the temperature decreases below the phase transition point TCDW. This temperature dependence is found to obey the hopping transport picture. In addition, a peak in cross-plane thermal conductivity is observed at TCDW as a fingerprint of the modulated structural distortion along the stacking direction.
Xin Qian, Te-Huan Liu, Ronggui Yang
Ionic Seebeck effect of electrolytes has shown promising applications in harvesting energy from low-grade waste-heat sources with small temperature difference from the environment, which can power sensors and Internet-of-Things devices. Recent experiments have demonstrated giant thermopower (~ 10 mV/K) of electrolytes under confinement due to the overlapping of electric double layer (EDL). Nonetheless, there has been no consensus on the theory of the ionic Seebeck effect, especially whether the thermopower depends on ionic diffusivities, imposing confusion on the theoretical interpretation of experimental discovery on giant thermopower of confined electrolytes. This article presents a linear perturbative solution of Poisson-Nernst-Planck (PNP) equations to describe the ionic Seebeck effect of confined liquid electrolytes. We provide both analytical and numerical solutions to the PNP equations for closed systems and open systems connected to reservoirs of electrolytes. The analytical solution captured the confinement effect both along and perpendicular to the temperature gradient, and showed excellent agreement with numerically solved PNP equations for a wide range of EDL potentials, channel widths, and lengths. Finally, we show that for polyelectrolytes with largely mismatched diffusivities, thermopower can only be enhanced for closed system through confinement perpendicular to the temperature gradient.
Xiangpan Ji, Wenqiang Gu, Xin Qian, Hanyu Wei, Chao Zhang
We describe an approximation to the widely-used Poisson-likelihood chi-square using a linear combination of Neyman's and Pearson's chi-squares, namely "combined Neyman-Pearson chi-square" ($χ^2_{\mathrm{CNP}}$). Through analytical derivations and toy model simulations, we show that $χ^2_\mathrm{CNP}$ leads to a significantly smaller bias on the best-fit model parameters compared to those using either Neyman's or Pearson's chi-square. When the computational cost of using the Poisson-likelihood chi-square is high, $χ^2_\mathrm{CNP}$ provides a good alternative given its natural connection to the covariance matrix formalism.
Xin Qian, Diego Klabjan
The mini-batch stochastic gradient descent (SGD) algorithm is widely used in training machine learning models, in particular deep learning models. We study SGD dynamics under linear regression and two-layer linear networks, with an easy extension to deeper linear networks, by focusing on the variance of the gradients, which is the first study of this nature. In the linear regression case, we show that in each iteration the norm of the gradient is a decreasing function of the mini-batch size $b$ and thus the variance of the stochastic gradient estimator is a decreasing function of $b$. For deep neural networks with $L_2$ loss we show that the variance of the gradient is a polynomial in $1/b$. The results back the important intuition that smaller batch sizes yield lower loss function values which is a common believe among the researchers. The proof techniques exhibit a relationship between stochastic gradient estimators and initial weights, which is useful for further research on the dynamics of SGD. We empirically provide further insights to our results on various datasets and commonly used deep network structures.
Xin Qian, Jungwoo Shin, Yaodong Tu, James Han Zhang, Gang Chen
Harvesting waste heat with temperatures lower than 100 oC can improve system efficiency and reduce greenhouse gas emissions, yet it has been a longstanding and challenging task. Electrochemical methods for harvesting low-grade heat have attracted research interest in recent years due to the relatively high effective temperature coefficient of the electrolytes (> 1 mV/K) compared with the thermopower of traditional thermoelectric devices. Comparing with other electrochemical devices such as temperature-variation based thermally regenerative electrochemical cycle and temperature-difference based thermogalvanic cells, the thermally regenerative flow battery (TRFB) has the advantages of providing a continuous power output, decoupling the heat source and heat sink and recuperating heat, and compatible with stacking for scaling up. However, TRFB suffers from the issue of stable operation due to the challenge of pH matching between catholyte and anolyte solutions with desirable temperature coefficients. In this work, we demonstrate a PH-neutral TRFB based on KI/KI3 and K3Fe(CN)6/K4Fe(CN)6 as the catholyte and anolyte, respectively, with a cell temperature coefficient of 1.9 mV/K and a power density of 9 uW/cm2. This work also presents a comprehensive model with a coupled analysis of mass transfer and reaction kinetics in a porous electrode that can accurately capture the flow rate dependence of power density and energy conversion efficiency. We estimate that the efficiency of the pH-neutral TRFB can reach 11% of the Carnot efficiency at the maximum power output with a temperature difference of 37 K. Via analysis, we identify that the mass transfer overpotential inside the porous electrode and the resistance of the ion exchange membrane are the two major factors limiting the efficiency and power density, pointing to directions for future improvements.
Xin Qian, Ryan A. Rossi, Fan Du, Sungchul Kim, Eunyee Koh, Sana Malik, Tak Yeon Lee, Nesreen K. Ahmed
Visualization recommendation work has focused solely on scoring visualizations based on the underlying dataset and not the actual user and their past visualization feedback. These systems recommend the same visualizations for every user, despite that the underlying user interests, intent, and visualization preferences are likely to be fundamentally different, yet vitally important. In this work, we formally introduce the problem of personalized visualization recommendation and present a generic learning framework for solving it. In particular, we focus on recommending visualizations personalized for each individual user based on their past visualization interactions (e.g., viewed, clicked, manually created) along with the data from those visualizations. More importantly, the framework can learn from visualizations relevant to other users, even if the visualizations are generated from completely different datasets. Experiments demonstrate the effectiveness of the approach as it leads to higher quality visualization recommendations tailored to the specific user intent and preferences. To support research on this new problem, we release our user-centric visualization corpus consisting of 17.4k users exploring 94k datasets with 2.3 million attributes and 32k user-generated visualizations.
Xin Qian, Guanda Quan, Te-Huan Liu, Ronggui Yang
API Phonons is a Python software package to predict the transport dynamics of heat-carrying phonons. Using the powerful syntax of Python, this package provides modules and functions interfacing between different packages for atomistic simulations, lattice dynamics, and phonon-phonon interaction calculations including LAMMPS, Quippy, Phonopy, and ShengBTE. API Phonons enabled complex phonon calculations, including (1) extracting harmonic and anharmonic force constants from arbitrary interatomic potentials, which can be used as inputs for solving Boltzmann transport equations; (2) predicting thermal conductivity using Kubo's linear response theory, which captures both quasiparticle transport and inter-band coherent transport; and (3) modeling of ultrafast pump-probe thermal responses using a Green's function approach based on mode-resolved phonon properties for studying ballistic, hydrodynamic, and diffusive transport dynamics. The package provides a flexible, easy-to-use, and extensive platform for modeling phonon transport physics through Python programming.
Xin Qian, Ronggui Yang
Quantitative descriptions of the structure-thermal property correlation have been a bottleneck in designing materials with superb thermal properties. In the past decade, the first-principles phonon calculations using density functional theory and the Boltzmann transport equation have become a common practice for predicting the thermal conductivity of new materials. However, first-principles calculations are too costly for high-throughput material screening and multi-scale structural design. First-principles calculations also face several fundamental challenges in modeling thermal transport properties, e.g., of crystalline materials with defects, of amorphous materials, and for materials at high temperatures. In the past five years, machine learning started to play a role in solving these challenges. This review provides a comprehensive summary and discussion on the state-of-the-art, future opportunities, and the remaining challenges in implementing machine learning for studying thermal conductivity. After an introduction to the working principles of machine learning and descriptors of material structures, recent research using machine learning to study thermal transport is discussed. Three major applications of machine learning for predicting thermal properties are discussed. First, machine learning is applied to solve the challenges in modeling phonon transport of crystals with defects, in amorphous materials, and at high temperatures. Machine learning is used to build high-fidelity interatomic potentials to bridge the gap between first-principles calculations and molecular dynamics simulations. Second, machine learning can be used to study the correlation between thermal conductivity and other properties for high-throughput materials screening. Finally, machine learning is a powerful tool for structural design to achieve target thermal conductance or thermal conductivity.
Zeyu Xiang, Yu Pang, Xin Qian, Ronggui Yang
Characterizing materials with spatially varying thermal conductivities is significant to unveil the structure-property relation for a wide range of functional materials, such as chemical-vapor-deposited diamonds, ion-irradiated materials, nuclear materials under radiation, and battery electrode materials. Although the development of thermal conductivity microscopy based on time/frequency-domain thermoreflectance (TDTR/FDTR) enabled in-plane scanning of thermal conductivity profile, measuring depth-dependent thermal conductivity remains challenging. This work proposed a machine-learning-based reconstruction method for extracting depth-dependent thermal conductivity K(z) directly from frequency-domain phase signals. We demonstrated that the simple supervised-learning algorithm kernel ridge regression (KRR) can reconstruct K(z) without requiring pre-knowledge about the functional form of the profile. The reconstruction method can not only accurately reproduce typical K(z) distributions such as the pre-assumed exponential profile of chemical-vapor-deposited (CVD) diamonds and Gaussian profile of ion-irradiated materials, but also complex profiles artificially constructed by superimposing Gaussian, exponential, polynomial, and logarithmic functions. In addition to FDTR, the method also shows excellent performances of reconstructing K(z) of ion-irradiated semiconductors from TDTR signals. This work demonstrates that combining machine learning with pump-probe thermoreflectance is an effective way for depth-dependent thermal property mapping.
Yuchi Chen, Qiangqiang Huang, Te-Huan Liu, Xin Qian, Ronggui Yang
Recent advancements in thermogalvanic batteries offer a promising route to efficient harvesting of low-grade heat with temperatures below 100 °C. The thermogalvanic temperature coefficient α, usually referred to as effective thermopower, is the key parameter determining the power density and efficiency of thermogalvanic batteries. However, the current understanding of improving α of redox pairs remains at the phenomenological level without microscopic insights, and the development of electrolytes with high α largely relies on experimental trial and error. This work applies the free energy perturbation method based on molecular dynamics simulations to predict the α of the {Fe^{3+}/Fe^{2+}} redox pair in aqueous and acetone solutions. We showed that α of the {Fe^{3+}/Fe^{2+}} redox pair can be increased from 1.5{\pm}0.3 mV/K to 4.1{\pm}0.4 mV/K with the increased acetone to water fraction. The predicted α of {Fe^{3+}/Fe^{2+}} both in pure water and acetone show excellent agreement with experimental values. By monitoring the fluctuation of dipole orientations in the first solvation shell, we discovered a significant change in the variance of solvent dipole orientation between Fe^{3+} and Fe^{2+}, which can be a microscopic indicator for large magnitudes of α. The effect of acetone weight fraction in the mixed acetone-water solvent on the α of {Fe^{3+}/Fe^{2+}} is also studied. Acetone molecules are found to intercalate into the first solvation shell of the {Fe^{2+}} ion at high acetone fractions, while this phenomenon is not observed in the solvation shell of the Fe^{3+} ion. Such solvation shell structure change of {Fe^{2+}} ions contributes to the enhanced α at high acetone fractions. Our discovery provides atomistic insights into how solvation shell order can be leveraged to develop electrolytes with high thermopower.
Xin Qian, Jen-Chieh Peng
Neutrinos produced by nuclear reactors have played a major role in advancing our knowledge of the properties of neutrinos. The first direct detection of the neutrino, confirming its existence, was performed using reactor neutrinos. More recent experiments utilizing reactor neutrinos have also found clear evidence for neutrino oscillation, providing unique input for the determination of neutrino mass and mixing. Ongoing and future reactor neutrino experiments will explore other important issues, including the neutrino mass hierarchy and the search for sterile neutrinos and other new physics beyond the standard model. In this article, we review the recent progress in physics using reactor neutrinos and the opportunities they offer for future discoveries.