arXiv subjects

Lu Huang

Publications and source records attributed to Lu Huang.

36 records · Page 2Linked to original sources

arXiv2023

Revisiting progenitor-age dependence of type Ia supernova luminosity standardization process

Much of the research in supernova cosmology is based on an assumption that the peak luminosity of type Ia supernovae (SNe Ia), after a standardization process, is independent of the galactic environment. A series of recent studies suggested that there is a significant correlation between the standardized luminosity and the progenitor age of SNe Ia. The correlation found in the most recent work by Lee et al. is strong enough to explain the extra dimming of distant SNe Ia and therefore casts doubts on the direct evidence of cosmic acceleration. The present work incorporates the uncertainties of progenitor ages, which were ignored in Lee et al., into a fully Bayesian inference framework. We find a weaker dependence of supernova standardized luminosity on the progenitor age, but the detection of correlation remains significant (3.5$σ$). Assuming that such correlation can be extended to high redshift and applying it to the Pantheon SN Ia data set, we confirm that when the Hubble residual does not include intrinsic scatter, the age-bias could be the primary cause of the observed extra dimming of distant SNe Ia. Furthermore, we use the PAge formalism, which is a good approximation to many dark energy and modified gravity models, to do a model comparison. We find that if intrinsic scatter is included in the Hubble residual, the Lambda cold dark matter model remains a good fit. However, in a scenario without intrinsic scatter, the Lambda cold dark matter model faces a challenge.

astro-ph.CO↗

arXiv2023

Text-only Domain Adaptation using Unified Speech-Text Representation in Transducer

Domain adaptation using text-only corpus is challenging in end-to-end(E2E) speech recognition. Adaptation by synthesizing audio from text through TTS is resource-consuming. We present a method to learn Unified Speech-Text Representation in Conformer Transducer(USTR-CT) to enable fast domain adaptation using the text-only corpus. Different from the previous textogram method, an extra text encoder is introduced in our work to learn text representation and is removed during inference, so there is no modification for online deployment. To improve the efficiency of adaptation, single-step and multi-step adaptations are also explored. The experiments on adapting LibriSpeech to SPGISpeech show the proposed method reduces the word error rate(WER) by relatively 44% on the target domain, which is better than those of TTS method and textogram method. Also, it is shown the proposed method can be combined with internal language model estimation(ILME) to further improve the performance.

cs.CL↗

arXiv2022

Removal of point source leakage from time-order data filtering

Time-ordered data (TOD) from ground-based CMB experiments are generally filtered before map-making to remove or reduce the contamination from the ground and the atmospheric emissions. However, when the observation region contains strong point sources, the filtering process will result in considerable leakage around the point sources in a measured CMB map, and leave spurious polarization signals. Therefore, such signals need to be assessed and removed before CMB science exploitation. In this work, we present a new method that we call "template fitting" and can effectively remove these leakage signals in pixel domain, not only satisfying the requirement for measuring primordial gravitational waves from CMB-$B$ modes, but also avoiding time-consuming operations on TOD.

astro-ph.CO↗

arXiv2022

Redshift Evolution and Non-Universal Dispersion of Quasar Luminosity Correlation

The standard $Λ$CDM model is recently reported to deviate from the high-redshift Hubble diagram of type Ia supernovae (SNe) and quasars (QSOs) at $\sim4σ$ confidence level. In this work, we combine the PAge approximation (a nearly model-independent parameterization) and a high-quality QSO sample to search for the origins of the deviation. By visualizing the $Λ$CDM model and the marginalized $3σ$ constraints of SNe+QSOs into PAge space, we confirm that the SNe+QSOs constraints in both flat and non-flat PAge cases are in remarkable tension with the standard $Λ$CDM cosmology. Next, we investigate the tension from the perspective of redshift-evolution effects. We find that the QSO correlation coefficient $γ$ calibrated by SNe+low-z QSOs and SNe+high-z QSOs shows $\sim2.7σ$ and $\sim4σ$ tensions in flat and non-flat universes, respectively. The tensions for intrinsic dispersion $δ$ between different data sets are found to be $>4σ$ in both flat and non-flat cases. These results indicate that the QSO luminosity correlation suffers from significant redshift evolution and non-universal intrinsic dispersion. Using a redshift-dependence correlation to build QSO Hubble diagram could lead to biases. Thus, the $\sim4σ$ deviation from the standard $Λ$CDM probably originates from the redshift-evolution effects and non-universal dispersion of the QSO luminosity correlation rather than new physics.

astro-ph.CO↗

arXiv2022

Search for lensing signatures from the latest fast radio burst observations and constraints on the abundance of primordial black holes

The possibility that primordial black holes (PBHs) form a part of dark matter has been considered for a long time but poorly constrained over a wide mass range. Fast radio bursts (FRBs) are bright radio transients with millisecond duration. Lensing effect of them has been proposed as one of the cleanest probes for constraining the presence of PBHs in the stellar mass window. In this paper, we first apply the normalised cross-correlation algorithm to search and identify candidates of lensed FRBs in the latest public FRB observations, i.e. $593$ FRBs which mainly consist of the first Canadian Hydrogen Intensity Mapping Experiment FRB catalog, and then derive constraints on the abundance of PBHs from the null search result of lensing signature. For a monochromatic mass distribution, the fraction of dark matter made up of PBHs could be constrained to $\leq87\%$ for $\geq500~M_{\odot}$ at 95\% confidence level by assuming flux ratio thresholds dependent signal-to-noise ratio for each FRB and that apparently one-off events are intrinsic single bursts. This result would be improved by a three times factor when a conventional constant flux ratio threshold is considered. Moreover, we derive constraints on PBHs with a log-normal mass function naturally predicted by some popular inflation models and often investigated with gravitational wave detections. We find that, in this mass distribution scenario, the constraint from currently public FRB observations is relatively weaker than the one from gravitational wave detections. It is foreseen that upcoming complementary multi-messenger observations will yield considerable constraints on the possibilities of PBHs in this intriguing mass window.

astro-ph.CO↗

arXiv2022

Constraints on the abundance of primordial black holes with different mass distributions from lensing of fast radio bursts

Primordial black holes (PBHs) has been considered to form a part of dark matter for a long time but the possibility has been poorly constrained over a wide mass range, including the stellar mass range ($1-100~M_{\odot}$). However, due to the discovery of merger events of black hole binaries by LIGO-Virgo gravitational wave observatories, the interest for PBHs in the stellar mass window has been aroused again. Fast radio bursts (FRBs) are bright radio transients with millisecond duration and very high all-sky occurrence rate. Lensing effect of these bursts has been proposed as one of the optimal probes for constraining the abundance of PBHs in the stellar mass range. In this paper, we first investigate constraints on the abundance of PBHs from the latest $593$ FRB observations for both the monochromatic mass distribution and three other popular extended mass distributions related to different formation mechanisms of PBHs. It is found that constraints from currently public FRB observations are relatively weaker than those from existing gravitational wave detections. Furthermore, we forecast constraining power of future FRB observations on the abundance of PBHs with different mass distributions of PBHs and different redshift distributions of FRBs taken into account. Finally, We find that constraints of parameter space on extended mass distributions from $\sim10^5$ FRBs with $\overline{Δt}\leq1 ~\rm ms$ would be comparable with what can be constrained from gravitational wave events. It is foreseen that upcoming complementary multi-messenger observations will yield considerable constraints on the possibilities of PBHs in this intriguing mass window.

astro-ph.CO↗

arXiv2022

The $S_8$ Tension in Light of Updated Redshift-Space Distortion Data and PAge Approximation

One of the most prominent challenges to the standard Lambda cold dark matter ($Λ$CDM) cosmology is the tension between the structure growth parameter $S_8$ constrained by the cosmic microwave background (CMB) data and the smaller one suggested by the cosmic shear data. Recent studies show that, for $Λ$CDM cosmology, redshift-space distortion (RSD) data also prefers a smaller $S_8$ that is $\sim 2$-$3σ$ lower than the CMB value, but the result is sensitive to the cosmological model. In the present work we update the RSD constraint on $S_8$ with the most up-to-date RSD data set where the correlation between data points are properly taken into account. To reduce the model dependence, we add in our Monte Carlo Markov Chain calculation the most up-to-date data sets of Type Ia supernovae (SN) and baryon acoustic oscillations (BAO), whose correlation with RSD is also taken into account, to constrain the background geometry. For $Λ$CDM cosmology we find $S_8= 0.812 \pm 0.026$, which is $\sim 2σ$ larger than previous studies, and hence is consistent with the CMB constraint. By replacing $Λ$CDM with the Parameterization based on cosmic Age (PAge), an almost model-independent description of the late universe, we find that the RSD + SN + BAO constraint on $S_8$ is insensitive to the cosmological model.

astro-ph.CO↗

arXiv2021

FGF-GAN: A Lightweight Generative Adversarial Network for Pansharpening via Fast Guided Filter

Pansharpening is a widely used image enhancement technique for remote sensing. Its principle is to fuse the input high-resolution single-channel panchromatic (PAN) image and low-resolution multi-spectral image and to obtain a high-resolution multi-spectral (HRMS) image. The existing deep learning pansharpening method has two shortcomings. First, features of two input images need to be concatenated along the channel dimension to reconstruct the HRMS image, which makes the importance of PAN images not prominent, and also leads to high computational cost. Second, the implicit information of features is difficult to extract through the manually designed loss function. To this end, we propose a generative adversarial network via the fast guided filter (FGF) for pansharpening. In generator, traditional channel concatenation is replaced by FGF to better retain the spatial information while reducing the number of parameters. Meanwhile, the fusion objects can be highlighted by the spatial attention module. In addition, the latent information of features can be preserved effectively through adversarial training. Numerous experiments illustrate that our network generates high-quality HRMS images that can surpass existing methods, and with fewer parameters.

cs.CV↗

arXiv2021

A More Accurate Parameterization based on cosmic Age (MAPAge)

Recently, several statistically significant tensions between different cosmological datasets have raised doubts about the standard Lambda cold dark matter ($Λ$CDM) model. A recent letter~\citet{Huang:2020mub} suggests to use "Parameterization based on cosmic Age" (PAge) to approximate a broad class of beyond-$Λ$CDM models, with a typical accuracy $\sim 1\%$ in angular diameter distances at $z\lesssim 10$. In this work, we extend PAge to a More Accurate Parameterization based on cosmic Age (MAPAge) by adding a new degree of freedom $η_2$. The parameter $η_2$ describes the difference between physically motivated models and their phenomenological PAge approximations. The accuracy of MAPAge, typically of order $10^{-3}$ in angular diameter distances at $z\lesssim 10$, is significantly better than PAge. We compare PAge and MAPAge with current observational data and forecast data. The conjecture in~\citet{Huang:2020mub}, that PAge approximation is sufficiently good for current observations, is quantitatively confirmed in this work. We also show that the extension from PAge to MAPAge is important for future observations, which typically requires sub-percent accuracy in theoretical predictions.

astro-ph.CO↗

arXiv2021

HMM-Free Encoder Pre-Training for Streaming RNN Transducer

This work describes an encoder pre-training procedure using frame-wise label to improve the training of streaming recurrent neural network transducer (RNN-T) model. Streaming RNN-T trained from scratch usually performs worse than non-streaming RNN-T. Although it is common to address this issue through pre-training components of RNN-T with other criteria or frame-wise alignment guidance, the alignment is not easily available in end-to-end manner. In this work, frame-wise alignment, used to pre-train streaming RNN-T's encoder, is generated without using a HMM-based system. Therefore an all-neural framework equipping HMM-free encoder pre-training is constructed. This is achieved by expanding the spikes of CTC model to their left/right blank frames, and two expanding strategies are proposed. To our best knowledge, this is the first work to simulate HMM-based frame-wise label using CTC model for pre-training. Experiments conducted on LibriSpeech and MLS English tasks show the proposed pre-training procedure, compared with random initialization, reduces the WER by relatively 5%~11% and the emission latency by 60 ms. Besides, the method is lexicon-free, so it is friendly to new languages without manually designed lexicon.

eess.AS↗

arXiv2021

Reconciling low and high redshift GRB luminosity correlations

The correlation between the peak spectra energy ($E_p$) and the equivalent isotropic energy ($E_{\rm iso}$) of long gamma-ray bursts (GRBs), the so-called Amati relation, is often used to constrain the high-redshift Hubble diagram. Assuming Lambda cold dark matter ($Λ$CDM) cosmology, Wang et al. found a $\gtrsim 3σ$ tension in the data-calibrated Amati coefficients between low- and high-redshift GRB samples. To reduce the impact of fiducial cosmology, we use the Parameterization based on cosmic Age (PAge), an almost model-independent framework to trace the cosmological expansion history. We find that the low- and high-redshift tension in Amati coefficients stays almost the same for the broad class of models covered by PAge, indicating that the cosmological assumption is not the dominant driver of the redshift evolution of GRB luminosity correlation. Next, we analyze the selection effect due to flux limits in observations. We find Amati relation evolves much more significantly across energy scales of $E_{\rm iso}$. We debias the GRB data by selectively discarding samples to match low-$z$ and high-$z$ $E_{\rm iso}$ distributions. After debiasing, the Amati coefficients agree well between low-$z$ and high-$z$ data groups, whereas the evidence of $E_{\rm iso}$-dependence of Amati relation remains to be strong. Thus, the redshift evolution of GRB luminosity correlation can be fully interpreted as a selection bias, and does not imply cosmological evolution of GRBs.

astro-ph.CO↗

arXiv2021

Deep Convolutional Sparse Coding Network for Pansharpening with Guidance of Side Information

Pansharpening is a fundamental issue in remote sensing field. This paper proposes a side information partially guided convolutional sparse coding (SCSC) model for pansharpening. The key idea is to split the low resolution multispectral image into a panchromatic image related feature map and a panchromatic image irrelated feature map, where the former one is regularized by the side information from panchromatic images. With the principle of algorithm unrolling techniques, the proposed model is generalized as a deep neural network, called as SCSC pansharpening neural network (SCSC-PNN). Compared with 13 classic and state-of-the-art methods on three satellites, the numerical experiments show that SCSC-PNN is superior to others. The codes are available at https://github.com/xsxjtu/SCSC-PNN.

cs.CV↗

arXiv2020

Reaffirming the Cosmic Acceleration without Supernova and CMB

Recent discussions about supernova magnitude evolution have raised doubts about the robustness of the late-universe acceleration. In a previous letter, Huang did a null test of the cosmic acceleration by using a Parameterization based on the cosmic Age (PAge), which covers a broad class of cosmological models including the standard $Λ$ cold dark matter model and its many extensions. In this work, we continue to explore the cosmic expansion history with the PAge approximation. Using baryon acoustic oscillations ({\it without} a CMB prior on the acoustic scale), gravitational strong lens time delay, and passively evolving early galaxies as cosmic chronometers, we obtain $\gtrsim 4σ$ detections of cosmic acceleration for both flat and nonflat PAge universes. In the nonflat case, we find a $\gtrsim 3σ$ tension between the spatial curvatures derived from baryon acoustic oscillations and strong lens time delay. Implications and possible systematics are discussed.

astro-ph.CO↗

arXiv2019

Utterance-level Permutation Invariant Training with Latency-controlled BLSTM for Single-channel Multi-talker Speech Separation

Utterance-level permutation invariant training (uPIT) has achieved promising progress on single-channel multi-talker speech separation task. Long short-term memory (LSTM) and bidirectional LSTM (BLSTM) are widely used as the separation networks of uPIT, i.e. uPIT-LSTM and uPIT-BLSTM. uPIT-LSTM has lower latency but worse performance, while uPIT-BLSTM has better performance but higher latency. In this paper, we propose using latency-controlled BLSTM (LC-BLSTM) during inference to fulfill low-latency and good-performance speech separation. To find a better training strategy for BLSTM-based separation network, chunk-level PIT (cPIT) and uPIT are compared. The experimental results show that uPIT outperforms cPIT when LC-BLSTM is used during inference. It is also found that the inter-chunk speaker tracing (ST) can further improve the separation performance of uPIT-LC-BLSTM. Evaluated on the WSJ0 two-talker mixed-speech separation task, the absolute gap of signal-to-distortion ratio (SDR) between uPIT-BLSTM and uPIT-LC-BLSTM is reduced to within 0.7 dB.

cs.SD↗

arXiv2019

Can Non-standard Recombination Resolve the Hubble Tension?

The inconsistent Hubble constant values derived from cosmic microwave background (CMB) observations and from local distance-ladder measurements may suggest new physics beyond the standard $Λ$CDM paradigm. It has been found in earlier works that, at least phenomenologically, non-standard recombination histories can reduce the $\gtrsim 4σ$ Hubble tension to $\sim 2σ$. Following this path, we vary physical and phenomenological parameters in RECFAST, the standard code to compute ionization history of the universe, to explore possible physics beyond standard recombination. We find that the CMB constraint on the Hubble constant is sensitive to the Hydrogen ionization energy and $2s \rightarrow 1s$ two-photon decay rate, both of which are atomic constants, and is insensitive to other details of recombination. Thus, the Hubble tension is very robust against perturbations of recombination history, unless exotic physics modifies the atomic constants during the recombination epoch.

astro-ph.CO↗

arXiv2019

The Domain Shift Problem of Medical Image Segmentation and Vendor-Adaptation by Unet-GAN

Convolutional neural network (CNN), in particular the Unet, is a powerful method for medical image segmentation. To date Unet has demonstrated state-of-art performance in many complex medical image segmentation tasks, especially under the condition when the training and testing data share the same distribution (i.e. come from the same source domain). However, in clinical practice, medical images are acquired from different vendors and centers. The performance of a U-Net trained from a particular source domain, when transferred to a different target domain (e.g. different vendor, acquisition parameter), can drop unexpectedly. Collecting a large amount of annotation from each new domain to retrain the U-Net is expensive, tedious, and practically impossible. In this work, we proposed a generic framework to address this problem, consisting of (1) an unpaired generative adversarial network (GAN) for vendor-adaptation, and (2) a Unet for object segmentation. In the proposed Unet-GAN architecture, GAN learns from Unet at the feature level that is segmentation-specific. We used cardiac cine MRI as the example, with three major vendors (Philips, Siemens, and GE) as three domains, while the methodology can be extended to medical images segmentation in general. The proposed method showed significant improvement of the segmentation results across vendors. The proposed Unet-GAN provides an annotation-free solution to the cross-vendor medical image segmentation problem, potentially extending a trained deep learning model to multi-center and multi-vendor use in real clinical scenario.

eess.IV↗

arXiv2017

An Improved Residual LSTM Architecture for Acoustic Modeling

Long Short-Term Memory (LSTM) is the primary recurrent neural networks architecture for acoustic modeling in automatic speech recognition systems. Residual learning is an efficient method to help neural networks converge easier and faster. In this paper, we propose several types of residual LSTM methods for our acoustic modeling. Our experiments indicate that, compared with classic LSTM, our architecture shows more than 8% relative reduction in Phone Error Rate (PER) on TIMIT tasks. At the same time, our residual fast LSTM approach shows 4% relative reduction in PER on the same task. Besides, we find that all this architecture could have good results on THCHS-30, Librispeech and Switchboard corpora.

cs.CL↗

arXiv2010

A quantum model for the stock market

Beginning with several basic hypotheses of quantum mechanics, we give a new quantum model in econophysics. In this model, we define wave functions and operators of the stock market to establish the Schrödinger equation for the stock price. Based on this theoretical framework, an example of a driven infinite quantum well is considered, in which we use a cosine distribution to simulate the state of stock price in equilibrium. After adding an external field into the Hamiltonian to analytically calculate the wave function, the distribution and the average value of the rate of return are shown.

q-fin.ST↗