arXiv subjects

Shi Pu

Publications and source records attributed to Shi Pu.

At least 91 records · Page 5Linked to original sources

arXiv2022

Alignment-Uniformity aware Representation Learning for Zero-shot Video Classification

Most methods tackle zero-shot video classification by aligning visual-semantic representations within seen classes, which limits generalization to unseen classes. To enhance model generalizability, this paper presents an end-to-end framework that preserves alignment and uniformity properties for representations on both seen and unseen classes. Specifically, we formulate a supervised contrastive loss to simultaneously align visual-semantic features (i.e., alignment) and encourage the learned features to distribute uniformly (i.e., uniformity). Unlike existing methods that only consider the alignment, we propose uniformity to preserve maximal-info of existing features, which improves the probability that unobserved features fall around observed data. Further, we synthesize features of unseen classes by proposing a class generator that interpolates and extrapolates the features of seen classes. Besides, we introduce two metrics, closeness and dispersion, to quantify the two properties and serve as new measurements of model generalizability. Experiments show that our method significantly outperforms SoTA by relative improvements of 28.1% on UCF101 and 27.0% on HMDB51. Code is available.

cs.CV↗

arXiv2022

Lorentz transformation in Maxwell equations for slowly moving media

We use the method of field decomposition, a technique widely used in relativistic magnetohydrodynamics, to study the small velocity approximation (SVA) of the Lorentz transformation in Maxwell equations for slowly moving media. The "deformed" Maxwell equations derived under the SVA in the lab frame can be put into the conventional form of Maxwell equations in the medium's comoving frame. Our results show that the Lorentz transformation in the SVA up to $O(v/c)$ ($v$ is the speed of the medium and $c$ is the speed of light in vacuum) is essential to derive these equations: the time and charge density must also change when transforming to a different frame even in the SVA, not just the position and current density as in the Galilean transformation. This marks the essential difference of the Lorentz transformation from the Galilean one. We show that the integral forms of Faraday and Ampere equations for slowly moving surfaces are consistent with Maxwell equations. We also present Faraday equation the covariant integral form in which the electromotive force can be defined as a Lorentz scalar independent of the observer's frame. No evidences exist to support an extension or modification of Maxwell equations.

physics.class-ph↗

arXiv2021

Reexamination of local spin polarization beyond global equilibrium in relativistic heavy ion collisions

We study local spin polarization in the relativistic hydrodynamic model. Generalizing the Wigner functions previously obtained from chiral kinetic theory by Y. Hidaka et al. [Phys. Rev. D 97, 016004 (2018)] to the massive case, we present the possible contributions up to the order of $\hbar$ from thermal vorticity, shear viscous tensor, other terms associated with the temperature and chemical-potential gradients, and electromagnetic fields to the local spin polarization. We then implement the (3+1)-dimensional viscous hydrodynamic model to study the spin polarizations from these sources with a small chemical potential and ignorance of electromagnetic fields by adopting an equation of state different from those in other recent studies. Although the shear correction alone upon the local polarization results in a sign and azimuthal-angle dependence more consistent with experimental observations, as also discovered in other recent studies, it is mostly suppressed by the contributions from thermal vorticity and other terms that yield an opposite trend. It is found that the total local spin polarization can be very sensitive to the equation of states, the ratio of shear viscosity to entropy density, and the freeze-out temperature.

hep-ph↗

arXiv2021

Analytic solutions of relativistic dissipative spin hydrodynamics with Bjorken expansion

We have studied analytically the longitudinally boost-invariant motion of a relativistic dissipative fluid with spin. We have derived the analytic solutions of spin density and spin chemical potential as a function of proper time $τ$ in the presence of viscous tensor and the second order relaxation time corrections for spin. Interestingly, analogous to the ordinary particle number density and chemical potential, we find that the spin density and spin chemical potential decay as $\simτ^{-1}$ and $\simτ^{-1/3}$, respectively. It implies that the initial spin density may not survive at the freezeout hyper-surface. These solutions can serve both to gain insight on the dynamics of spin polarization in relativistic heavy-ion collisions and as testbeds for further numerical codes.

nucl-th↗

arXiv2021

Compressed Gradient Tracking Methods for Decentralized Optimization with Linear Convergence

Communication compression techniques are of growing interests for solving the decentralized optimization problem under limited communication, where the global objective is to minimize the average of local cost functions over a multi-agent network using only local computation and peer-to-peer communication. In this paper, we first propose a novel compressed gradient tracking algorithm (C-GT) that combines gradient tracking technique with communication compression. In particular, C-GT is compatible with a general class of compression operators that unifies both unbiased and biased compressors. We show that C-GT inherits the advantages of gradient tracking-based algorithms and achieves linear convergence rate for strongly convex and smooth objective functions. In the second part of this paper, we propose an error feedback based compressed gradient tracking algorithm (EF-C-GT) to further improve the algorithm efficiency for biased compression operators. Numerical examples complement the theoretical findings and demonstrate the efficiency and flexibility of the proposed algorithms.

math.OC↗

arXiv2021

Lepton pair production in ultra-peripheral collisions

The lepton pair production in ultra-peripheral collisions (UPC) is studied in the classical field approximation. We derive a general form of the cross section in terms of photon distributions which depend on the transverse momentum and coordinate based on the wave packet form of nuclear wave functions. Such a general form of the cross section in the classical field approximation contains the results of the generalized equivalent photon approximation (EPA) as well as the corrections beyond EPA in the Born approximation. By rewriting the general form of the cross section in light-cone coordinates, we find a good connection with the transverse momentum dependent distribution (TMD) factorization formalism in the Born approximation. Our numerical results are consistent with current experimental data.

hep-ph↗

arXiv2021

Improving the Transient Times for Distributed Stochastic Gradient Methods

We consider the distributed optimization problem where $n$ agents each possessing a local cost function, collaboratively minimize the average of the $n$ cost functions over a connected network. Assuming stochastic gradient information is available, we study a distributed stochastic gradient algorithm, called exact diffusion with adaptive stepsizes (EDAS) adapted from the Exact Diffusion method and NIDS and perform a non-asymptotic convergence analysis. We not only show that EDAS asymptotically achieves the same network independent convergence rate as centralized stochastic gradient descent (SGD) for minimizing strongly convex and smooth objective functions, but also characterize the transient time needed for the algorithm to approach the asymptotic convergence rate, which behaves as $K_T=\mathcal{O}\left(\frac{n}{1-λ_2}\right)$, where $1-λ_2$ stands for the spectral gap of the mixing matrix. To the best of our knowledge, EDAS achieves the shortest transient time when the average of the $n$ cost functions is strongly convex and each cost function is smooth. Numerical simulations further corroborate and strengthen the obtained theoretical results.

math.OC↗

arXiv2021

A Sharp Estimate on the Transient Time of Distributed Stochastic Gradient Descent

This paper is concerned with minimizing the average of $n$ cost functions over a network in which agents may communicate and exchange information with each other. We consider the setting where only noisy gradient information is available. To solve the problem, we study the distributed stochastic gradient descent (DSGD) method and perform a non-asymptotic convergence analysis. For strongly convex and smooth objective functions, DSGD asymptotically achieves the optimal network independent convergence rate compared to centralized stochastic gradient descent (SGD). Our main contribution is to characterize the transient time needed for DSGD to approach the asymptotic convergence rate, which we show behaves as $K_T=\mathcal{O}\left(\frac{n}{(1-ρ_w)^2}\right)$, where $1-ρ_w$ denotes the spectral gap of the mixing matrix. Moreover, we construct a "hard" optimization problem for which we show the transient time needed for DSGD to approach the asymptotic convergence rate is lower bounded by $Ω\left(\frac{n}{(1-ρ_w)^2} \right)$, implying the sharpness of the obtained result. Numerical experiments demonstrate the tightness of the theoretical results.

math.OC↗

arXiv2020

Causality and stability in relativistic viscous non-resistive magneto-fluid dynamics

We investigate the causality and the stability of the relativistic viscous magneto-hydrodynamics in the framework of the Israel-Stewart (IS) second-order theory, and also within a modified IS theory which incorporates the effect of magnetic fields in the relaxation equations of the viscous stress. We compute the dispersion relation by perturbing the fluid variables around their equilibrium values. In the ideal magnetohydrodynamics limit, the linear dispersion relation yields the well-known propagating modes: the Alfvén and the magneto-sonic modes.In the presence of bulk viscous pressure, the causality bound is found to be independent of the magnitude of the magnetic field. The same bound also remains true, when we take the full non-linear form of the equation using the method of characteristics. In the presence of shear viscous pressure, the causality bound is independent of the magnitude of the magnetic field for the two magneto-sonic modes. The causality bound for the shear-Alfvén modes, however, depends both on the magnitude and the direction of the propagation. For modified IS theory in the presence of shear viscosity, new non-hydrodynamic modes emerge but the asymptotic causality condition is the same as that of IS. In summary, although the magnetic field does influence the wave propagation in the fluid, the study of the stability and asymptotic causality conditions in the fluid rest frame shows that the fluid remains stable and causal given that they obey certain asymptotic causality condition.

nucl-th↗

arXiv2020

Multimodal Topic Learning for Video Recommendation

Facilitated by deep neural networks, video recommendation systems have made significant advances. Existing video recommendation systems directly exploit features from different modalities (e.g., user personal data, user behavior data, video titles, video tags, and visual contents) to input deep neural networks, while expecting the networks to online mine user-preferred topics implicitly from these features. However, the features lacking semantic topic information limits accurate recommendation generation. In addition, feature crosses using visual content features generate high dimensionality features that heavily downgrade the online computational efficiency of networks. In this paper, we explicitly separate topic generation from recommendation generation, propose a multimodal topic learning algorithm to exploit three modalities (i.e., tags, titles, and cover images) for generating video topics offline. The topics generated by the proposed algorithm serve as semantic topic features to facilitate preference scope determination and recommendation generation. Furthermore, we use the semantic topic features instead of visual content features to effectively reduce online computational cost. Our proposed algorithm has been deployed in the Kuaibao information streaming platform. Online and offline evaluation results show that our proposed algorithm performs favorably.

cs.IR↗

arXiv2020

Spin Hydrodynamics and Symmetric Energy-Momentum Tensors -- A current induced by the spin vorticity --

We discuss a puzzle in relativistic spin hydrodynamics; in the previous formulation the spin source from the antisymmetric part of the canonical energy-momentum tensor (EMT) is crucial. The Belinfante improved EMT is pseudo-gauge transformed from the canonical EMT and is usually a physically sensible choice especially when gauge fields are coupled as in magnetohydrodynamics, but the Belinfante EMT has no antisymmetric part. We find that pseudo-transformed entropy currents are physically inequivalent in nonequilibrium situations. We also identify a current induced by the spin vorticity read from the Belinfante symmetric EMT.

hep-th↗

arXiv2020

A general framework for decentralized optimization with first-order methods

Decentralized optimization to minimize a finite sum of functions over a network of nodes has been a significant focus within control and signal processing research due to its natural relevance to optimal control and signal estimation problems. More recently, the emergence of sophisticated computing and large-scale data science needs have led to a resurgence of activity in this area. In this article, we discuss decentralized first-order gradient methods, which have found tremendous success in control, signal processing, and machine learning problems, where such methods, due to their simplicity, serve as the first method of choice for many complex inference and training tasks. In particular, we provide a general framework of decentralized first-order methods that is applicable to undirected and directed communication networks alike, and show that much of the existing work on optimization and consensus can be related explicitly to this framework. We further extend the discussion to decentralized stochastic first-order methods that rely on stochastic gradients at each node and describe how local variance reduction schemes, previously shown to have promise in the centralized settings, are able to improve the performance of decentralized methods when combined with what is known as gradient tracking. We motivate and demonstrate the effectiveness of the corresponding methods in the context of machine learning and signal processing problems that arise in decentralized environments.

cs.LG↗

arXiv2020

A Robust Gradient Tracking Method for Distributed Optimization over Directed Networks

In this paper, we consider the problem of distributed consensus optimization over multi-agent networks with directed network topology. Assuming each agent has a local cost function that is smooth and strongly convex, the global objective is to minimize the average of all the local cost functions. To solve the problem, we introduce a robust gradient tracking method (R-Push-Pull) adapted from the recently proposed Push-Pull/AB algorithm. R-Push-Pull inherits the advantages of Push-Pull and enjoys linear convergence to the optimal solution with exact communication. Under noisy information exchange, R-Push-Pull is more robust than the existing gradient tracking based algorithms; the solutions obtained by each agent reach a neighborhood of the optimum in expectation exponentially fast under a constant stepsize policy. We provide a numerical example that demonstrate the effectiveness of R-Push-Pull.

math.OC↗

arXiv2020

Relativistic Kelvin circulation theorem for ideal Magnetohydrodynamics

We have studied the relativistic Kelvin circulation theorem for ideal Magnetohydrodynamics. The relativistic Kelvin circulation theorem is a conservation equation for the called $T$-vorticity. We have briefly reviewed the ideal magnetohydrodynamics in relativistic heavy ion collisions. The highlight of this work is that we have obtained the general expression of relativistic Kelvin circulation theorem for ideal Magnetohydrodynamics. We have also applied the analytic solutions of ideal magnetohydrodynamics in Bjorken flow to check our results. Our main results can also be implemented to relativistic magnetohydrodynamics in relativistic heavy ion collisions.

nucl-th↗

arXiv2020

Chirality Production with Mass Effects-Schwinger Pair Production and the Axial Ward Identity

The anomalous generation of chirality with mass effects via the axial Ward identity and its dependence on the Schwinger mechanism is reviewed, utilizing parity violating homogeneous electromagnetic background fields. The role vacuum asymptotic states play on the interpretation of expectation values is examined. It is discussed that observables calculated with an in-out scattering matrix element predict a scenario under Euclidean equilibrium. A notable ramification of which is a vanishing of the chiral anomaly. In contrast, it is discussed observables calculated under an in-in, or real-time, formalism predict a scenario out-of equilibrium, and capture effects of mean produced particle anti-particle pairs due to the Schwinger mechanism. The out-of equilibrium chiral anomaly is supplemented with exponential quadratic mass suppression as anticipated for the Schwinger mechanism. Similar behavior in and out-of equilibrium is reviewed for applications including the chiral magnetic effect and chiral condensate.

hep-ph↗

arXiv2020

Recent developments in chiral and spin polarization effects in heavy-ion collisions

We give a brief overview of recent theoretical and experimental results on the chiral magnetic effect and spin polarization effect in heavy-ion collisions. We present updated experimental results for the chiral magnetic effect and related phenomena. The time evolution of the magnetic fields in different models is discussed. The newly developed quantum kinetic theory for massive fermions is reviewed. We present theoretical and experimental results for the polarization of $Λ$ hyperons and the $ρ_{00}$ value of vector mesons.

hep-ph↗

arXiv2020

Relativistic decomposition of the orbital and the spin angular momentum in chiral physics and Feynman's angular momentum paradox

Over recent years we have witnessed tremendous progresses in our understanding on the angular momentum decomposition. In the context of the proton spin problem in high energy processes the angular momentum decomposition by Jaffe and Manohar, which is based on the canonical definition, and the alternative by Ji, which is based on the Belinfante improved one, have been revisited under light shed by Chen et al. leading to seminal works by Hatta, Wakamatsu, Leader, etc. In chiral physics as exemplified by the chiral vortical effect and applications to the relativistic nucleus-nucleus collisions, sometimes referred to as a relativistic extension of the Barnett and the Einstein--de Haas effects, such arguments of the angular momentum decomposition would be of crucial importance. We pay our special attention to the fermionic part in the canonical and the Belinfante conventions and discuss a difference between them, which is reminiscent of a classical example of Feynman's angular momentum paradox. We point out its possible relevance to early-time dynamics in the nucleus-nucleus collisions, resulting in excess by the electromagnetic angular momentum.

hep-ph↗

arXiv2020

Anomalous magnetohydrodynamics with constant anisotropic electric conductivities

We study anomalous magnetohydrodynamics in a longitudinal boost invariant Bjorken flow with constant anisotropic electric conductivities as outlined in Ref. [1]. For simplicity, we consider a neutral fluid and a force-free magnetic field in the transverse direction. We derived analytic solutions of the electromagnetic fields in the laboratory frame, the chiral density, and the energy density as functions of proper time.

hep-ph↗