arXiv Science⌕ Search

arXiv subjects

Wei Cheng

Publications and source records attributed to Wei Cheng.

At least 235 records · Page 13Linked to original sources
arXiv2020

T$^2$-Net: A Semi-supervised Deep Model for Turbulence Forecasting

Accurate air turbulence forecasting can help airlines avoid hazardous turbulence, guide the routes that keep passengers safe, maximize efficiency, and reduce costs. Traditional turbulence forecasting approaches heavily rely on painstakingly customized turbulence indexes, which are less effective in dynamic and complex weather conditions. The recent availability of high-resolution weather data and turbulence records allows more accurate forecasting of the turbulence in a data-driven way. However, it is a non-trivial task for developing a machine learning based turbulence forecasting system due to two challenges: (1) Complex spatio-temporal correlations, turbulence is caused by air movement with complex spatio-temporal patterns, (2) Label scarcity, very limited turbulence labels can be obtained. To this end, in this paper, we develop a unified semi-supervised framework, T$^2$-Net, to address the above challenges. Specifically, we first build an encoder-decoder paradigm based on the convolutional LSTM to model the spatio-temporal correlations. Then, to tackle the label scarcity problem, we propose a novel Dual Label Guessing method to take advantage of massive unlabeled turbulence data. It integrates complementary signals from the main Turbulence Forecasting task and the auxiliary Turbulence Detection task to generate pseudo-labels, which are dynamically utilized as additional training data. Finally, extensive experimental results on a real-world turbulence dataset validate the superiority of our method on turbulence forecasting.

cs.LG↗
arXiv2020

Branching fractions and polarizations of $D\to V(ω,ρ, K^*) \ell ν_\ell$ within QCD LCSR

In this paper, we make a detailed study about the $D\to V$ helicity form factors (HFFs) within the framework of QCD light-cone sum rule (LCSR) up to twist-4 accuracy. After extrapolating the LCSR predictions of HFFs to the whole physical $q^2$-region, we get the longitudinal, transverse and total $|V_{cq}|$-independent decay widths of semileptonic decay $D\to V\ell^+ν_\ell$. Meanwhile, the branching fractions of these decays are also obtained by using the $D^0(D^+)$-meson lifetime, which agree well with the BES-III results within errors. As a further step, we also investigate the differential and mean predictions for charged lepton (vector meson) polarization in the final state $P_{\rm L,T}^\ell$ ($F_{\rm L,T}^\ell$), the forward-backward asymmetry ${\cal A}_{\rm FB}^\ell$, and the lepton-side convexity parameters ${\cal C}_{\rm F}^\ell$. Our predictions are consistent with Covariant Confining Quark Model results within the errors. Thus, we think the LCSR approach for HFFs is applicable for dealing with the $D$-meson semileptonic decays.

hep-ph↗
arXiv2020

Local singular characteristics on $\mathbb{R}^2$

The singular set of a viscosity solution to a Hamilton-Jacobi equation is known to propagate, from any noncritical singular point, along singular characteristics which are curves satisfying certain differential inclusions. In the literature, different notions of singular characteristics were introduced. However, a general uniqueness criterion for singular characteristics, not restricted to mechanical systems or problems in one space dimension, is missing at the moment. In this paper, we prove that, for a Tonelli Hamiltonian on $\mathbb{R}^2$, two different notions of singular characteristics coincide up to a bi-Lipschitz reparameterization. As a significant consequence, we obtain a uniqueness result for the class of singular characteristics that was introduced by Khanin and Sobolevski in the paper [On dynamics of Lagrangian trajectories for Hamilton-Jacobi equations. Arch. Ration. Mech. Anal., 219(2):861-885, 2016].

math.OC↗
arXiv2020

Generalizing Variational Autoencoders with Hierarchical Empirical Bayes

Variational Autoencoders (VAEs) have experienced recent success as data-generating models by using simple architectures that do not require significant fine-tuning of hyperparameters. However, VAEs are known to suffer from over-regularization which can lead to failure to escape local maxima. This phenomenon, known as posterior collapse, prevents learning a meaningful latent encoding of the data. Recent methods have mitigated this issue by deterministically moment-matching an aggregated posterior distribution to an aggregate prior. However, abandoning a probabilistic framework (and thus relying on point estimates) can both lead to a discontinuous latent space and generate unrealistic samples. Here we present Hierarchical Empirical Bayes Autoencoder (HEBAE), a computationally stable framework for probabilistic generative models. Our key contributions are two-fold. First, we make gains by placing a hierarchical prior over the encoding distribution, enabling us to adaptively balance the trade-off between minimizing the reconstruction loss function and avoiding over-regularization. Second, we show that assuming a general dependency structure between variables in the latent space produces better convergence onto the mean-field assumption for improved posterior inference. Overall, HEBAE is more robust to a wide-range of hyperparameter initializations than an analogous VAE. Using data from MNIST and CelebA, we illustrate the ability of HEBAE to generate higher quality samples based on FID score than existing autoencoder-based approaches.

stat.ML↗
arXiv2020

$D \to P(π,K)$ helicity form factors within light-cone sum rule approach

In this paper, the $D\to P(π, K)$ helicity form factors (HFFs) are studied by applying the QCD light-cone sum rule (LCSR) approach. The calculation accuracy is up to next-to-leading order (NLO) gluon radiation correction of twist-(2,3) distribution amplitude. The resultant HFFs at large recoil point are ${\cal P}_{t,0}^π(0) = 0.688^{+0.020}_{-0.024}$, ${\cal P}_{t,0}^K(0)=0.780^{+0.024}_{-0.029}$. In which, the contributions from three particles of the leading order (LO) are so small that can be safely neglected, and the maximal contribution of the NLO gluon radiation correction for ${\cal P}_{t,0}^{π,K}(0)$ is less than $3\%$. After extrapolating the LCSR predictions for these HFFs to whole $q^2$-region, we obtain the decay widths for semileptonic decay processes $D\to P\ellν_\ell$, which are consistent with BES-III collaboration predictions within errors. After considering the $D^{+}/D^{0}$-meson lifetime, we give the branching fractions of $D\to P\ellν_\ell$ with $\ell = e, μ$, our predictions also agree with BES-III collaboration within errors, especially for $D\to π\ellν_\ell$ decay process. Finally, we present the forward-backward asymmetry ${\cal A}_{\rm FB}^\ell(q^2)$ and lepton convexity parameter ${\cal C}_F^\ell(q^2)$, and further calculate the mean value of these two observations $\langle{\cal A}_{\rm FB}^\ell\rangle$ and $\langle{\cal C}_F^\ell\rangle$, which may provide a way to test those HFFs in future experiments.

hep-ph↗
arXiv2020

Weak KAM approach to first-order Mean Field Games with state constraints

We study the asymptotic behavior of solutions to the constrained MFG system as the time horizon $T$ goes to infinity. For this purpose, we analyze first Hamilton-Jacobi equations with state constraints from the viewpoint of weak KAM theory, constructing a Mather measure for the associated variational problem. Using these results, we show that a solution to the constrained ergodic mean field games system exists and the ergodic constant is unique. Finally, we prove that any solution of the first-order constrained MFG problem on $[0,T]$ converges to the solution of the ergodic system as $T \to +\infty$.

math.AP↗
arXiv2020

The $D\to ρ$ semileptonic and radiative decays within the light-cone sum rules

The measured branching ratio of the $D$ meson semileptonic decay $D \to ρe^+ ν_e$, which is based on the $0.82~{\rm fb^{-1}}$ CLEO data taken at the peak of $ψ(3770)$ resonance, disagrees with the traditional SVZ sum rules analysis by about three times. In the paper, we show that this discrepancy can be eliminated by applying the QCD light-cone sum rules (LCSR) approach to calculate the $D\to ρ$ transition form factors $A_{1,2}(q^2)$ and $V(q^2)$. After extrapolating the LCSR predictions of these TFFs to whole $q^2$-region, we obtain $1/|V_{\rm cd}|^2 \times Γ(D \to ρe ν_e) =(55.45^{+13.34}_{-9.41})\times 10^{-15}~{\rm GeV}$. Using the CKM matrix element and the $D^0(D^+)$ lifetime from the Particle Data Group, we obtain ${\cal B} (D^0\to ρ^- e^+ ν_e) = (1.749^{+0.421}_{-0.297}\pm 0.006)\times 10^{-3}$ and ${\cal B} (D^+ \to ρ^0 e^+ ν_e) = (2.217^{+0.534}_{-0.376}\pm 0.015)\times 10^{-3}$, which agree with the CLEO measurements within errors. We also calculate the branching ratios of the two $D$ meson radiative processes and obtain ${\cal B}(D^0\to ρ^0 γ)= (1.744^{+0.598}_{-0.704})\times 10^{-5}$ and ${\cal B}(D^+ \to ρ^+ γ) = (5.034^{+0.939}_{-0.958})\times 10^{-5}$, which also agree with the Belle measurements within errors. Thus we think the LCSR approach is applicable for dealing with the $D$ meson decays.

hep-ph↗
arXiv2019

Asymmetrical Hierarchical Networks with Attentive Interactions for Interpretable Review-Based Recommendation

Recently, recommender systems have been able to emit substantially improved recommendations by leveraging user-provided reviews. Existing methods typically merge all reviews of a given user or item into a long document, and then process user and item documents in the same manner. In practice, however, these two sets of reviews are notably different: users' reviews reflect a variety of items that they have bought and are hence very heterogeneous in their topics, while an item's reviews pertain only to that single item and are thus topically homogeneous. In this work, we develop a novel neural network model that properly accounts for this important difference by means of asymmetric attentive modules. The user module learns to attend to only those signals that are relevant with respect to the target item, whereas the item module learns to extract the most salient contents with regard to properties of the item. Our multi-hierarchical paradigm accounts for the fact that neither are all reviews equally useful, nor are all sentences within each review equally pertinent. Extensive experimental results on a variety of real datasets demonstrate the effectiveness of our method.

cs.IR↗
arXiv2019

Singularities of solutions of time dependent Hamilton-Jacobi equations. Applications to Riemannian geometry

If $U:[0,+\infty[\times M$ is a uniformly continuous viscosity solution of the evolution Hamilton-Jacobi equation $$\partial_tU+ H(x,\partial_xU)=0,$$ where $M$ is a not necessarily compact manifold, and $H$ is a Tonelli Hamiltonian, we prove the set $Σ(U)$, of points where $U$ is not differentiable, is locally contractible. Moreover, we study the homotopy type of $Σ(U)$. We also give an application to the singularities of a distance function to a closed subset of a complete Riemannian manifold.

math.AP↗
arXiv2019

The $B\to π\ellν_\ell$ semileptonic decay within the LCSR approach under heavy quark effective field theory

The heavy quark effective field theory (HQEFT) provides an effective way to deal with the heavy meson decays. In the paper, we adopt two different correlators to derive the light-cone sum rules of the $B \to π$ transition form factors (TFFs) within the framework of HQEFT. We label those two LCSR results as LCSR-${\cal U}$ and LCSR-${\cal R}$, which are for conventional correlator and right-handed correlator, respectively. We observe that the correlation parameter $|ρ_{\rm RU}|$ for the branching ratio ${\cal B}(B \to πl ν_{l})$ is $\sim 0.85$, implying the consistency of the LCSRs under different correlators. Moreover, we obtain $|V_{\rm ub}|_{{\rm LCSR}-{\cal U}}=(3.45^{+0.28}_{-0.20}\pm{0.13}_{\rm{exp}})\times10^{-3}$ and $|V_{\rm ub}|_{{\rm LCSR}-{\cal R}} =(3.38^{+0.22}_{-0.16} \pm{0.12}_{\rm{exp}})\times10^{-3}$. We then obtain $\mathcal{R}_π|_{{\rm LCSR}-{\cal U}}=0.68^{+0.10}_{-0.09}$ and $\mathcal{R}_π|_{{\rm LCSR}-{\cal R}}=0.65^{+0.13}_{-0.11}$, both of them agree with the Lattice QCD predictions. Thus the HQEFT provides a useful framework for studying the $B$ meson decays. Moreover, by using right-handed correlator, the twist-2 terms shall dominant the TFF $f^+(q^2)$, which approaches over $\sim97\%$ contribution in the whole $q^2$-region; and the large twist-3 uncertainty for the conventional correlator is greatly suppressed. One can thus adopt the LCSR-${\cal R}$ prediction to test the properties of the various models for the pion twist-2 distribution amplitudes.

hep-ph↗
arXiv2019

Semantic Graph Convolutional Network for Implicit Discourse Relation Classification

Implicit discourse relation classification is of great importance for discourse parsing, but remains a challenging problem due to the absence of explicit discourse connectives communicating these relations. Modeling the semantic interactions between the two arguments of a relation has proven useful for detecting implicit discourse relations. However, most previous approaches model such semantic interactions from a shallow interactive level, which is inadequate on capturing enough semantic information. In this paper, we propose a novel and effective Semantic Graph Convolutional Network (SGCN) to enhance the modeling of inter-argument semantics on a deeper interaction level for implicit discourse relation classification. We first build an interaction graph over representations of the two arguments, and then automatically extract in-depth semantic interactive information through graph convolution. Experimental results on the English corpus PDTB and the Chinese corpus CDTB both demonstrate the superiority of our model to previous state-of-the-art systems.

cs.CL↗
arXiv2019

Learning Robust Representations with Graph Denoising Policy Network

Graph representation learning, aiming to learn low-dimensional representations which capture the geometric dependencies between nodes in the original graph, has gained increasing popularity in a variety of graph analysis tasks, including node classification and link prediction. Existing representation learning methods based on graph neural networks and their variants rely on the aggregation of neighborhood information, which makes it sensitive to noises in the graph. In this paper, we propose Graph Denoising Policy Network (short for GDPNet) to learn robust representations from noisy graph data through reinforcement learning. GDPNet first selects signal neighborhoods for each node, and then aggregates the information from the selected neighborhoods to learn node representations for the down-stream tasks. Specifically, in the signal neighborhood selection phase, GDPNet optimizes the neighborhood for each target node by formulating the process of removing noisy neighborhoods as a Markov decision process and learning a policy with task-specific rewards received from the representation learning phase. In the representation learning phase, GDPNet aggregates features from signal neighbors to generate node representations for down-stream tasks, and provides task-specific rewards to the signal neighbor selection phase. These two phases are jointly trained to select optimal sets of neighbors for target nodes with maximum cumulative task-specific rewards, and to learn robust representations for nodes. Experimental results on node classification task demonstrate the effectiveness of GDNet, outperforming the state-of-the-art graph representation learning methods on several well-studied datasets. Additionally, GDPNet is mathematically equivalent to solving the submodular maximizing problem, which theoretically guarantees the best approximation to the optimal solution with GDPNet.

cs.LG↗
arXiv2019

Gravitational waves triggered by $B-L$ charged hidden scalar and leptogenesis

We study the electroweak symmetry breaking in the framework of a classically conformal $U(1)_{B-L}$ theory, where three right-handed neutrinos (RHNs) and a hidden scalar are introduced, with the latter playing the role of dark matter (DM). It is found that the DM and RHN sectors are crucial for the spontaneous symmetry breaking of the $U(1)_{B-L}$ symmetry, strong first order phase transition in the conformal theory and the resultant gravitational wave (GW) prospects at future space-based interferometer LISA and other GW experiments. The baryon asymmetry of the Universe is addressed by the resonant leptogenesis mechanism, which is potentially disturbed by the hidden scalar. To make the GW spectra detectable by LISA and resonant leptogenesis work in the conformal $U(1)_{B-L}$ theory, the hidden scalar can not fully saturate the observed DM relic density.

hep-ph↗
arXiv2019

Herglotz' variational principle and Lax-Oleinik evolution

We develop an elementary method to give a Lipschitz estimate for the minimizers in the problem of Herglotz' variational principle proposed in \cite{CCWY2018} in the time-dependent case. We deduce Erdmann's condition and the Euler-Lagrange equation separately under different sets of assumptions, by using a generalized du Bois-Reymond lemma. As an application, we obtain a representation formula for the viscosity solution of the Cauchy problem for the Hamilton-Jacobi equation \begin{align*} D_tu(t,x)+H(t,x,D_xu(t,x),u(t,x))=0 \end{align*} and study the related Lax-Oleinik evolution.

math.AP↗
arXiv2019

Higgs inflation and cosmological electroweak phase transition with N scalars in the post-Higgs era

We study inflation and cosmological electroweak phase transitions utilizing the Standard model augmented by $N$ scalars respecting a global $O(N)$ symmetry. We observe that the representation of the global symmetry is restricted by the inflationary observables and the condition of a strongly first order electroweak phase transition. Theoretical constraints including the stability, perturbativity and unitarity are used to bound the model parameter space. The Electroweak precision observables and Higgs precisions limit the representation of the symmetry. We evaluate the possibility to simultaneously address the inflation and the dark matter after considering the experimental constraints from the future leptonic colliders. When the $O (N)$ symmetry respected by the N-scalar is spontaneously broken to the $O (N-1)$ symmetry, both the one-step and two-step SFOEWPT can occur within the inflation viable parameter regions, which will be tested by the future CEPC, ILC and FCC-ee. The relation between the number of Goldstones and the SFOEWPT condition depends on phase transition patterns. The situation of Goldstone faking neutrinos and contributing to the dark radiation are investigated.

hep-ph↗