arXiv subjects

Ali Davody

Publications and source records attributed to Ali Davody.

At least 19 recordsLinked to original sources

arXiv2022

On the correlation functions in stable first-order relativistic hydrodynamics

First-order relativistic conformal hydrodynamics in a general (hydrodynamic) frame is characterized by a shear viscosity coefficient and two UV-regulator parameters. Within a certain range of these parameters, the equilibrium is stable and propagation is causal. In this work we study the correlation functions of fluctuations in this theory. We first compute hydrodynamic correlation functions in the linear response regime. Then we use the linear response results to explore the analytical structure of response functions beyond the linear response. A method is developed to numerically calculate the branch cut structure from the well-known Landau equations. We apply our method to the shear channel and find the branch cuts of a certain response function, without computing the response function itself. We then solve the Landau equations analytically and find the threshold singularities of the same response function. Using these results, we achieve the leading singularity in momentum space, by which, we find the long-time tail of the correlation function. The results turn out to be in complete agreement with the loop calculations in effective field theory.

hep-th↗

arXiv2022

Privacy Guarantees for De-identifying Text Transformations

Machine Learning approaches to Natural Language Processing tasks benefit from a comprehensive collection of real-life user data. At the same time, there is a clear need for protecting the privacy of the users whose data is collected and processed. For text collections, such as, e.g., transcripts of voice interactions or patient records, replacing sensitive parts with benign alternatives can provide de-identification. However, how much privacy is actually guaranteed by such text transformations, and are the resulting texts still useful for machine learning? In this paper, we derive formal privacy guarantees for general text transformation-based de-identification methods on the basis of Differential Privacy. We also measure the effect that different ways of masking private information in dialog transcripts have on a subsequent machine learning task. To this end, we formulate different masking strategies and compare their privacy-utility trade-offs. In particular, we compare a simple redact approach with more sophisticated word-by-word replacement using deep learning models on multiple natural language understanding tasks like named entity recognition, intent detection, and dialog act classification. We find that only word-by-word replacement is robust against performance drops in various tasks.

cs.CL↗

arXiv2022

TOKEN is a MASK: Few-shot Named Entity Recognition with Pre-trained Language Models

Transferring knowledge from one domain to another is of practical importance for many tasks in natural language processing, especially when the amount of available data in the target domain is limited. In this work, we propose a novel few-shot approach to domain adaptation in the context of Named Entity Recognition (NER). We propose a two-step approach consisting of a variable base module and a template module that leverages the knowledge captured in pre-trained language models with the help of simple descriptive patterns. Our approach is simple yet versatile and can be applied in few-shot and zero-shot settings. Evaluating our lightweight approach across a number of different datasets shows that it can boost the performance of state-of-the-art baselines by 2-5% F1-score.

cs.CL↗

arXiv2021

On the effect of normalization layers on Differentially Private training of deep Neural networks

Differentially private stochastic gradient descent (DPSGD) is a variation of stochastic gradient descent based on the Differential Privacy (DP) paradigm, which can mitigate privacy threats that arise from the presence of sensitive information in training data. However, one major drawback of training deep neural networks with DPSGD is a reduction in the models accuracy. In this paper, we study the effect of normalization layers on the performance of DPSGD. We demonstrate that normalization layers significantly impact the utility of deep neural networks with noisy parameters and should be considered essential ingredients of training with DPSGD. In particular, we propose a novel method for integrating batch normalization with DPSGD without incurring an additional privacy loss. With our approach, we are able to train deeper networks and achieve a better utility-privacy trade-off.

cs.LG↗

arXiv2021

Preventing Author Profiling through Zero-Shot Multilingual Back-Translation

Documents as short as a single sentence may inadvertently reveal sensitive information about their authors, including e.g. their gender or ethnicity. Style transfer is an effective way of transforming texts in order to remove any information that enables author profiling. However, for a number of current state-of-the-art approaches the improved privacy is accompanied by an undesirable drop in the down-stream utility of the transformed data. In this paper, we propose a simple, zero-shot way to effectively lower the risk of author profiling through multilingual back-translation using off-the-shelf translation models. We compare our models with five representative text style transfer models on three datasets across different domains. Results from both an automatic and a human evaluation show that our approach achieves the best overall performance while requiring no training data. We are able to lower the adversarial prediction of gender and race by up to $22\%$ while retaining $95\%$ of the original utility on downstream tasks.

cs.CL↗

arXiv2020

SuperCoder: Program Learning Under Noisy Conditions From Superposition of States

We propose a new method of program learning in a Domain Specific Language (DSL) which is based on gradient descent with no direct search. The first component of our method is a probabilistic representation of the DSL variables. At each timestep in the program sequence, different DSL functions are applied on the DSL variables with a certain probability, leading to different possible outcomes. Rather than handling all these outputs separately, whose number grows exponentially with each timestep, we collect them into a superposition of variables which captures the information in a single, but fuzzy, state. This state is to be contrasted at the final timestep with the ground-truth output, through a loss function. The second component of our method is an attention-based recurrent neural network, which provides an appropriate initialization point for the gradient descent that optimizes the probabilistic representation. The method we have developed surpasses the state-of-the-art for synthesising long programs and is able to learn programs under noise.

cs.LG↗

arXiv2019

Standardized Cumulants of Flow Harmonic Fluctuations

The distribution of flow harmonics in heavy ion experiment can be characterized by standardized cumulants. We first model the ellipticity and power parameters of the elliptic-power distribution by employing MC-Glauber model. Then we use the elliptic-power distribution together with the hydrodynamic linear response approximation to study the two dimensional standardized cumulants of elliptic and triangular flow ($v_2$ and $v_3$) distribution. For the second harmonic, it turns out that finding two dimensional cumulants in terms of $2q$-particle correlation functions $c_2\{2q\}$ is limited to the skewness. We also show that $c_3\{2\}$, $c_3\{4\}$, and $c_3\{6\}$, are related to the second, fourth, and sixth standardized cumulants of the $v_3$ distribution, respectively. The cumulant $c_{n}\{2q\}$ can be also written in terms of $v_n\{2q\}$. Specifically, $-(v_3\{4\}/v_3\{2\})^4$ turns out to be the kurtosis of the $v_3$ event-by-event fluctuation distribution. We introduce a new parametrization for the distribution $p(v_3)$ with $v_3\{2\}$, kurtosis and sixth-order standardized cumulant being its free parameters. Compared to the Gaussian distribution, it indicates a more accurate fit with experimental results. Finally, we compare the kurtosis obtained from simulation with that of extracted from experimental data for the $v_3$ distribution.

nucl-th↗

arXiv2017

Collective Excitations in Hot QCD Plasma

We study the long wavelength excitations in rotating QCD fluid in presence of an external magnetic field at finite vector and axial charge densities. We consider the fluctuations of vector and axial charge currents coupled to energy and momentum fluctuations and compute the $SO(3)$ covariant dispersion relations of the six corresponding hydrodynamic modes. Among them, there are always two scalar Chiral Magnetic-Vortical-Heat (CMVH) waves; In the absence of magnetic field (vorticity) these waves reduce to CVH (CMH) waves. While CMVH waves are the mixed of CMH and CVH waves, they have generally different velocities compared to the sum of velocities of the latter waves. The other four modes, which are made out of scalar-vector fluctuations, are mixed Sound-Alfvén waves. We show that in the direction parallel to both magnetic field and vorticity, these four modes are the two ordinary sound modes together with two Chiral Alfvén waves (CAW).

nucl-th↗

arXiv2016

On Butterfly effect in Higher Derivative Gravities

We study butterfly effect in $D$-dimensional gravitational theories containing terms quadratic in Ricci scalar and Ricci tensor. One observes that due to higher order derivatives in the corresponding equations of motion there are two butterfly velocities. The velocities are determined by the dimension of operators whose sources are provided by the metric. The three dimensional TMG model is also studied where we get two butterfly velocities at generic point of the moduli space of parameters. At critical point two velocities coincide.

hep-th↗

arXiv2016

Hydrodynamic Waves in an Anomalous Charged Fluid

We study the collective excitations in a relativistic fluid with an anomalous $U(1)$ current. In $3+1$ dimensions at zero chemical potential, in addition to ordinary sound modes we find two propagating modes in presence of an external magnetic field. The first one which is a transverse degenerate mode, propagates with a velocity proportional to the coefficient of gravitational anomaly; this is in fact the Chiral Alfvén wave recently found in \cite{Yamamoto:2015ria}. Another one is a wave of density perturbation, namely a chiral magnetic wave (CMW). The velocity dependence of CMW on the chiral anomaly coefficient is well known. We compute the dependence of CMW's velocity on the coefficient of gravitational anomaly as well. We also show that the dissipation splits the degeneracy of CAW. At finite chiral charge density we show that in general there may exist five chiral hydrodynamic waves. Of these five waves, one is the CMW while the other four are mixed Modified Sound-Alfvén waves. It turns out that in propagation transverse to the magnetic field no anomaly effect appear while in parallel to the magnetic field we find sound waves become dispersive due to anomaly.

hep-th↗

arXiv2015

The Energy Loss of a Heavy Quark Moving in a Viscous Fluid

To study the rate of energy and momentum loss of a heavy quark in QGP, specifically in the hydrodynamic regime, we use fluid/gravity duality and construct a perturbative procedure to find the string solution in gravity side. We show that by this construction the drag force exerted on the quark can be computed perturbatively, order by order in a boundary derivative expansion. At ideal order, our result is just the drag force exerted on a moving quark in thermal plasma with thermodynamics variables promoted to become local functions of space and time. Furthermore, we apply this procedure to a transverse quark in Bjorken flow and compute the first-derivative corrections, namely the viscous corrections, to the drag force.

hep-th↗

arXiv2015

The Energy Loss of a Heavy Quark Moving Through a General Fluid Dynamical Flow

We determine the most general form of the covariant drag force exerted on a quark moving through a fluid dynamical flow. Up to first order in derivative expansion, our general formula requires the specification of seven coefficient functions. We use the perturbative method introduced in arXiv:1202.2737 and find all these coefficients in the hydrodynamic regime of a $\mathcal{N}=4$ SYM plasma. Having this general formula, we can obtain the rate of the energy and momentum loss of a quark, namely the drag force, in a general flow. This result makes it possible to perturbatively study the motion of heavy quarks moving through the Bjorken flow up to first order in derivative expansion.

hep-th↗

arXiv2015

Dissipative Charged Fluid in a Magnetic Field

We study the collective excitations in a dissipative charged fluid at zero chemical potential when an external magnetic field is present. While in the absence of magnetic field, four collective excitations appear in the fluid, we find five hydrodynamic modes in presence of magnetic field. This implies that the magnetic field splits the degeneracy between the transverse shear modes. Using linear response theory, we then compute the retarded response functions. In particular, it turns out that the correlation between charge and the energy fluctuations will no longer vanish, even at zero chemical potential. By use of the response functions, we also derive the relevant Kubo formulas for the transport coefficients.

hep-th↗

arXiv2014

Monte Carlo Computation of Spectral Density Function in Real-Time Scalar Field Theory

Non-perturbative study of "real-time" field theories is difficult due to the sign problem. We use Bold Schwinger-Dyson (SD) equations to study the real-time $ϕ^4$ theory in $d=4$ beyond the perturbative regime. Combining SD equations in a particular way, we derive a non-linear integral equation for the two-point function. Then we introduce a new method by which one can analytically perform the momentum part of loop integrals in this equation. The price we must pay for such simplification is to numerically solve a non-linear integral equation for the spectral density function. Using Bold diagrammatic Monte Carlo method we find non-perturbative spectral function of theory and compare it with the one obtained from perturbation theory. At the end we utilize our Monte Carlo result to find the full vertex function as the basis for the computation of real-time scattering amplitudes.

hep-ph↗

arXiv2014

Nonconformal holographic model for D-meson suppression at energies available at the CERN Large Hadron Collider

The drag force of charm quarks propagating through a thermalized system of Quark Gluon Plasma (QGP) has been considered within the framework of both conformal/non-conformal Anti de Sitter (AdS) correspondence. Newly derived Einstein Fluctuation-Dissipation relation has been used to calculate the heavy flavor diffusion coefficients. Using the drag and diffusion coefficients as inputs Langevin equation has been solved to study the heavy flavor suppression factor. It has been shown that within conformal AdS correspondence the D-meson suppression at LHC energy can be reproduced where as the non-conformal AdS correspondence fail to reproduce the experimental results. It suggests collisional loss alone within non-conformal AdS correspondence can not reproduce the experimental results and inclusion of radiative loss becomes important.

hep-ph↗

arXiv2013

Bold Diagrammatic Monte Carlo Study of $ϕ^4$ Theory

By incorporating renormalization procedure into Bold Diagrammatic Monte Carlo (BDMC), we propose a method for studying quantum field theories in the strong coupling regime. BDMC essentially samples Feynman diagrams using local Metropolis-type updates and does not suffer from the sign problem. Applying the method to three dimensional $ϕ^4$ theory, we analyze the strong coupling limit of the theory and confirm the existence of a nontrivial IR fixed point in agreement with prior studies. Interestingly, we find that working with bold correlation functions as building blocks of the Monte Carlo procedure, renders the scheme convergent and no further resummation method is needed.

hep-lat↗

arXiv2011

Noncritical Holographic QCD in External Electric Field

We investigate behavior of a noncritical model in external electric field and explore its phase structure in the quenched approximation N_f>> N_c. We compute the conductivity of QCD plasma in this model and compare it with the predictions of Sakai-Sugimoto model, D3-D7 system and the lattice simulation. We find that, while the behavior of conductivity in noncritical model as a function of temperature and baryon density is similar to those of D3-D7 system, the phase diagram of noncritical model resembles the phase diagram of Sakai-Sugimoto model.

hep-th↗

arXiv2009

Weyl Anomaly in NonRelativistic CFTs

We study Weyl symmetry for non-relativistic conformal filed theories on curved spatial spaces, and calculate it's quantum anomaly. We show that there is no geometric anomaly, and the non-relativistic Weyl anomaly can appear only due to interaction. Also we study the anomaly by using the light-cone approach.

hep-th↗