arXiv subjects

Wei Xue

Publications and source records attributed to Wei Xue.

At least 181 records · Page 10Linked to original sources

arXiv2021

Neural Kalman Filtering for Speech Enhancement

Statistical signal processing based speech enhancement methods adopt expert knowledge to design the statistical models and linear filters, which is complementary to the deep neural network (DNN) based methods which are data-driven. In this paper, by using expert knowledge from statistical signal processing for network design and optimization, we extend the conventional Kalman filtering (KF) to the supervised learning scheme, and propose the neural Kalman filtering (NKF) for speech enhancement. Two intermediate clean speech estimates are first produced from recurrent neural networks (RNN) and linear Wiener filtering (WF) separately and are then linearly combined by a learned NKF gain to yield the NKF output. Supervised joint training is applied to NKF to learn to automatically trade-off between the instantaneous linear estimation made by the WF and the long-term non-linear estimation made by the RNN. The NKF method can be seen as using expert knowledge from WF to regularize the RNN estimations to improve its generalization ability to the noise conditions unseen in the training. Experiments in different noisy conditions show that the proposed method outperforms the baseline methods both in terms of objective evaluation metrics and automatic speech recognition (ASR) word error rates (WERs).

eess.AS↗

arXiv2020

Re-examining the Solar Axion Explanation for the XENON1T Excess

The XENON1T collaboration has observed an excess in electronic recoil events below $5~\mathrm{keV}$ over the known background, which could originate from beyond-the-Standard-Model physics. The solar axion is a well-motivated model that has been proposed to explain the excess, though it has tension with astrophysical observations. The axions traveled from the Sun can be absorbed by the electrons in the xenon atoms via the axion-electron coupling. Meanwhile, they can also scatter with the atoms through the inverse Primakoff process via the axion-photon coupling, which emits a photon and mimics the electronic recoil signals. We found that the latter process cannot be neglected. After including the $\rm{keV}$ photon produced via inverse Primakoff in the detection, the tension with the astrophysical constraints can be significantly reduced. We also explore scenarios involving additional new physics to further alleviate the tension with the astrophysical bounds.

hep-ph↗

arXiv2020

Few-shot Object Detection with Self-adaptive Attention Network for Remote Sensing Images

In remote sensing field, there are many applications of object detection in recent years, which demands a great number of labeled data. However, we may be faced with some cases where only limited data are available. In this paper, we proposed a few-shot object detector which is designed for detecting novel objects provided with only a few examples. Particularly, in order to fit the object detection settings, our proposed few-shot detector concentrates on the relations that lie in the level of objects instead of the full image with the assistance of Self-Adaptive Attention Network (SAAN). The SAAN can fully leverage the object-level relations through a relation GRU unit and simultaneously attach attention on object features in a self-adaptive way according to the object-level relations to avoid some situations where the additional attention is useless or even detrimental. Eventually, the detection results are produced from the features that are added with attention and thus are able to be detected simply. The experiments demonstrate the effectiveness of the proposed method in few-shot scenes.

cs.CV↗

arXiv2020

An Adaptive Remote Stochastic Gradient Method for Training Neural Networks

We present the remote stochastic gradient (RSG) method, which computes the gradients at configurable remote observation points, in order to improve the convergence rate and suppress gradient noise at the same time for different curvatures. RSG is further combined with adaptive methods to construct ARSG for acceleration. The method is efficient in computation and memory, and is straightforward to implement. We analyze the convergence properties by modeling the training process as a dynamic system, which provides a guideline to select the configurable observation factor without grid search. ARSG yields $O(1/\sqrt{T})$ convergence rate in non-convex settings, that can be further improved to $O(\log(T)/T)$ in strongly convex settings. Numerical experiments demonstrate that ARSG achieves both faster convergence and better generalization, compared with popular adaptive methods, such as ADAM, NADAM, AMSGRAD, and RANGER for the tested problems. In particular, for training ResNet-50 on ImageNet, ARSG outperforms ADAM in convergence speed and meanwhile it surpasses SGD in generalization.

cs.LG↗

arXiv2020

Few-shot Object Detection with Feature Attention Highlight Module in Remote Sensing Images

In recent years, there are many applications of object detection in remote sensing field, which demands a great number of labeled data. However, in many cases, data is extremely rare. In this paper, we proposed a few-shot object detector which is designed for detecting novel objects based on only a few examples. Through fully leveraging labeled base classes, our model that is composed of a feature-extractor, a feature attention highlight module as well as a two-stage detection backend can quickly adapt to novel classes. The pre-trained feature extractor whose parameters are shared produces general features. While the feature attention highlight module is designed to be light-weighted and simple in order to fit the few-shot cases. Although it is simple, the information provided by it in a serial way is helpful to make the general features to be specific for few-shot objects. Then the object-specific features are delivered to the two-stage detection backend for the detection results. The experiments demonstrate the effectiveness of the proposed method for few-shot cases.

cs.CV↗

arXiv2020

SkipConvNet: Skip Convolutional Neural Network for Speech Dereverberation using Optimally Smoothed Spectral Mapping

The reliability of using fully convolutional networks (FCNs) has been successfully demonstrated by recent studies in many speech applications. One of the most popular variants of these FCNs is the `U-Net', which is an encoder-decoder network with skip connections. In this study, we propose `SkipConvNet' where we replace each skip connection with multiple convolutional modules to provide decoder with intuitive feature maps rather than encoder's output to improve the learning capacity of the network. We also propose the use of optimal smoothing of power spectral density (PSD) as a pre-processing step, which helps to further enhance the efficiency of the network. To evaluate our proposed system, we use the REVERB challenge corpus to assess the performance of various enhancement approaches under the same conditions. We focus solely on monitoring improvements in speech quality and their contribution to improving the efficiency of back-end speech systems, such as speech recognition and speaker verification, trained on only clean speech. Experimental findings show that the proposed system consistently outperforms other approaches.

eess.AS↗

arXiv2019

Looking for MACHOs in the Spectra of Fast Radio Bursts

We explore a novel search strategy for dark matter in the form of massive compact halo objects (MACHOs) such as primordial black holes or dense mini-halos in the mass range from $10^{-4}$ to 0.1 solar masses. These objects can gravitationally lens the signal of fast radio bursts (FRBs), producing a characteristic interference pattern in the frequency spectrum, similar to the previously studied femtolensing signal in gamma ray burst spectra. Unlike traditional searches using microlensing, FRB lensing will probe the abundance of MACHOs at cosmological distance scales (~Gpc) rather than just their distribution in the neighborhood of the Milky Way. The method is thus particularly relevant for dark mini-halos, which may be inaccessible to microlensing due to their finite spatial extent or tidal disruption in galaxies. We find that the main complication in FRB lensing will be interstellar scintillation in the FRB's host galaxy and in the Milky Way. Scintillation is difficult to quantify because it heavily depends on turbulence in the interstellar medium, which is poorly understood. We show that, nevertheless, for realistic scintillation parameters, FRB lensing can set competitive limits on compact dark matter object, and we back our findings with explicit simulations.

astro-ph.CO↗

arXiv2019

Co-Interacting Dark Matter

We propose a novel mechanism with two component dark matter models. The subdominant dark matter can thermalize the dominant one in galaxies, and leads to core density profiles. Using ultralight dark photons and $\mathrm{GeV}$-$\mathrm{TeV}$ Dirac fermions as an example, we couple the two dark matter candidates with a $U(1)$ interaction. This mechanism differs from self-interacting dark matter, due to three effects: (1) higher occupation numbers, (2) forward-backward scattering cancellation, and (3) the multiple scatterings required for the heavy dark matter. Furthermore, the Bullet Cluster bound is evaded due to the reduced Bose enhancement factor. Unlike the fuzzy dark matter solution to the small structure problems which have tension with Lyman-$α$, the ultralight dark photons with mass $\gtrsim 10^{-21}$ eV can have a core profile through interactions with $ψ$ and are not constrained by other astrophysical observations.

hep-ph↗

arXiv2019

Orbital Angular Momentum Multiplexing in Highly Reverberant Environments

Previous studies on orbital angular momentum (OAM) communication mainly considered line-of-sight environments. In this letter, however, it is found that OAM communication with high-order modulation can be achieved in highly reverberant environments by combining the OAM multiplexing with a spatial equalizer. The OAM multiplexing exhibits comparable performance of conventional multiple-input multiple-output (MIMO) system.

eess.SP↗

arXiv2019

Direct anthropic bound on the weak scale from supernovae explosions

Core-collapse supernovae presumably explode because trapped neutrinos push the material out of the stellar envelope. This process is directly controlled by the weak scale $v$: we argue that supernova explosions happen only if fundamental constants are tuned within a factor of few as $v \sim Λ_{\rm QCD}^{3/4} M_{\rm Pl}^{1/4}$, such that neutrinos are trapped in supernovae for a time comparable to the gravitational time-scale. We provide analytic arguments and simulations in spherical approximation, that need to be validated by more comprehensive simulations. The above result can be important for fundamental physics, because core-collapse supernova explosions seem anthropically needed, as they spread intermediate-mass nuclei presumably necessary for `life'. We also study stellar burning, finding that it does not provide anthropic boundaries on $v$.

astro-ph.HE↗

arXiv2019

Searching in CMS Open Data for Dimuon Resonances with Substantial Transverse Momentum

We study dimuon events in 2.11/fb of 7 TeV pp collisions, using CMS Open Data, and search for a narrow dimuon resonance with moderate mass (14-66 GeV) and substantial transverse momentum (pT). Applying dimuon pT cuts of 25 GeV and 60 GeV, we explore two overlapping samples: one with isolated muons, and one with prompt muons without an isolation requirement. Using the latter sample requires information about detector effects and QCD backgrounds, which we obtain directly from the CMS Open Data. We present model-independent limits on the product of cross section, branching fraction, acceptance, and efficiencies. These limits are stronger, relative to a corresponding inclusive search without a pT cut, by factors of as much as nine. Our "pT-enhanced" dimuon search strategy provides improved sensitivity to models in which a new particle is produced mainly in the decay of something heavier, as could occur, for example, in decays of the Higgs boson or of a TeV-scale top partner. An implementation of this method with the current 13 TeV data should improve the sensitivity to such signals further by roughly an order of magnitude.

hep-ph↗

arXiv2019

Impeded Dark Matter

We consider dark matter models in which the mass splitting between the dark matter particles and their annihilation products is tiny. Compared to the previously proposed Forbidden Dark Matter scenario, the mass splittings we consider are much smaller, and are allowed to be either positive or negative. To emphasize this modification, we dub our scenario "Impeded Dark Matter". We demonstrate that Impeded Dark Matter can be easily realized without requiring tuning of model parameters. For negative mass splitting, we demonstrate that the annihilation cross-section for Impeded Dark Matter depends linearly on the dark matter velocity or may even be kinematically forbidden, making this scenario almost insensitive to constraints from the cosmic microwave background and from observations of dwarf galaxies. Accordingly, it may be possible for Impeded Dark Matter to yield observable signals in clusters or the Galactic center, with no corresponding signal in dwarfs. For positive mass splitting, we show that the annihilation cross-section is suppressed by the small mass splitting, which helps light dark matter to survive increasingly stringent constraints from indirect searches. As specific realizations for Impeded Dark Matter, we introduce a model of vector dark matter from a hidden $SU(2)$ sector, and a composite dark matter scenario based on a QCD-like dark sector.

hep-ph↗

arXiv2019

Can the ANITA anomalous events be due to new physics?

The ANITA collaboration has observed two ultra-high-energy upgoing air shower events that cannot originate from Standard Model neutrinos that have traversed the Earth. Several beyond-the-standard-model physics scenarios have been proposed as explanations for these events. In this paper we present some general arguments making it challenging for new physics to explain the events. One exceptional class of models that could work is pointed out, in which metastable dark matter decays to a highly boosted lighter dark matter particle, that can interact in the Earth to produce the observed events.

hep-ph↗

arXiv2018

Femtolensing by Dark Matter Revisited

Femtolensing of gamma ray bursts (GRBs) has been put forward as an exciting possibility to probe exotic astrophysical objects with masses below $10^{-13}$ solar masses such as small primordial black holes or ultra-compact dark matter minihalos, made up for instance of QCD axions. In this paper we critically review this idea, properly taking into account the extended nature of the source as well as wave optics effects. We demonstrate that most GRBs are inappropriate for femtolensing searches due to their large sizes. This removes the previous femtolensing bounds on primordial black holes, implying that vast regions of parameter space for primordial black hole dark matter are not robustly constrained. Still, we entertain the possibility that a small fraction of GRBs, characterized by fast variability can have smaller sizes and be useful. However, a large number of such bursts would need to be observed to achieve meaningful constraints. We study the sensitivity of future observations as a function of the number of detected GRBs and of the size of the emission region.

astro-ph.CO↗

arXiv2018

Dark Matter in the Standard Model?

We critically reexamine two possible Dark Matter candidate within the Standard Model. First, we consider the $uuddss$ exa-quark. Its QCD binding energy could be large enough to make it (quasi) stable. We show that the cosmological Dark Matter abundance is reproduced thermally if its mass is 1.2 GeV. However, we also find that such mass is excluded by the stability of Oxygen nuclei. Second, we consider the possibility that the instability in the Higgs potential leads to the formation of primordial black holes while avoiding vacuum decay during inflation. We show that the non-minimal Higgs coupling to gravity must be as small as allowed by quantum corrections, $|ξ_H| < 0.01$. Even so, one must assume that the Universe survived in $e^{120}$ independent regions to fluctuations that lead to vacuum decay with probability 1/2 each.

hep-ph↗

arXiv2018

Serendipity in dark photon searches

Searches for dark photons provide serendipitous discovery potential for other types of vector particles. We develop a framework for recasting dark photon searches to obtain constraints on more general theories, which includes a data-driven method for determining hadronic decay rates. We demonstrate our approach by deriving constraints on a vector that couples to the $B\!-\!L$ current, a leptophobic $B$ boson that couples directly to baryon number and to leptons via $B$-$γ$ kinetic mixing, and on a vector that mediates a protophobic force. Our approach can easily be generalized to any massive gauge boson with vector couplings to the Standard Model fermions, and software to perform any such recasting is provided at https://gitlab.com/philten/darkcast .

hep-ph↗

arXiv2018

Aspect Based Sentiment Analysis with Gated Convolutional Networks

Aspect based sentiment analysis (ABSA) can provide more detailed information than general sentiment analysis, because it aims to predict the sentiment polarities of the given aspects or entities in text. We summarize previous approaches into two subtasks: aspect-category sentiment analysis (ACSA) and aspect-term sentiment analysis (ATSA). Most previous approaches employ long short-term memory and attention mechanisms to predict the sentiment polarity of the concerned targets, which are often complicated and need more training time. We propose a model based on convolutional neural networks and gating mechanisms, which is more accurate and efficient. First, the novel Gated Tanh-ReLU Units can selectively output the sentiment features according to the given aspect or entity. The architecture is much simpler than attention layer used in the existing models. Second, the computations of our model could be easily parallelized during training, because convolutional layers do not have time dependency as in LSTM layers, and gating units also work independently. The experiments on SemEval datasets demonstrate the efficiency and effectiveness of our models.

cs.CL↗

arXiv2018

Exposing Dark Sector with Future Z-Factories

We investigate the prospects of searching dark sector models via exotic Z-boson decay at future $e^+ e^-$ colliders with Giga Z and Tera Z options. Four general categories of dark sector models: Higgs portal dark matter, vector portal dark matter, inelastic dark matter and axion-like particles, are considered. Focusing on channels motivated by the dark sector models, we carry out a model independent study of the sensitivities of Z-factories in probing exotic decays. The limits on branching ratios of the exotic Z decay are typically $\mathcal{O} (10^{-6} - 10^{-8.5}) $ for the Giga Z and $\mathcal{O} (10^{-7.5} - 10^{-11})$ for the Tera Z, and they are compared with the projection for the high luminosity LHC. We demonstrate that future Z-factories can provide its unique and leading sensitivity, and highlight the complementarity with other experiments, including the indirect and direct dark matter search limits, and the existing collider limits. Future Z factories will play a leading role to uncover the hidden sector of the universe in the future.

hep-ph↗