arXiv subjects

D. Liu

Publications and source records attributed to D. Liu.

At least 55 records · Page 3Linked to original sources

arXiv2023

The [CII] 158 $μ$m emission line as a gas mass tracer in high redshift quiescent galaxies

Many efforts have been done in recent years to probe the gas fraction evolution of massive quiescent galaxies (QGs); however, a clear picture has not yet been established. Recent spectroscopic confirmations at z>3 offer the chance to measure the residual gas reservoirs of massive galaxies a few hundreds of Myr after their death and to study how fast quenching proceeds in a highly star-forming Universe. Even so, stringent constraints at z$>$2 remain hardly accessible with ALMA when adopting molecular gas tracers commonly used for the quenched population. In this letter, we propose overcoming this impasse by using the carbon [CII] 158 $μ$m emission line to systematically probe the gaseous budget of unlensed QGs at z>2.8, when these galaxies could still host non-negligible star formation on an absolute scale and when the line becomes best observable with ALMA (Bands 8 and 7). So far predominantly used for star-forming galaxies, this emission line is the best choice to probe the gas budget of spectroscopically confirmed QGs at $z>3$, reaching 2-4 and 13-30 times deeper than dust continuum (ALMA band 7) and CO(2-1)/(1-0) (VLA K-K$α$ bands), respectively, at fixed integration time. Exploiting archival ALMA observations, we place conservative 3$σ$ upper limits on the molecular gas fraction (f$_{\rm{mol}}=M_{\rm{H_2}}/M_{\star}$) of ADF22-QG1 (f$_{\rm{mol}}$<21%), ZF-COS-20115 (f$_{\rm{mol}}$<3.2%), two of the best-studied high-z QGs in the literature, and GS-9209 (f$_{\rm{mol}}$<72%), the most distant massive QG discovered to date. The deep upper limit found for ZF-COS-20115 is 3 times lower than previously anticipated for high-z QGs suggesting, at best, the existence of a large scatter in the f$_{\rm{mol}}$ distribution of the first QGs. Lastly, we discuss the current limitations of the method and propose ways to mitigate some of them by exploiting ALMA bands 9 and 10.

astro-ph.GA↗

arXiv2023

COSMOS2020: The Galaxy Stellar Mass Function: the assembly and star formation cessation of galaxies at $0.2\lt z \leq 7.5$

How galaxies form, assemble, and cease their star-formation is a central question within the modern landscape of galaxy evolution studies. These processes are indelibly imprinted on the galaxy stellar mass function (SMF). We present constraints on the shape and evolution of the SMF, the quiescent galaxy fraction, and the cosmic stellar mass density across 90% of the history of the Universe from $z=7.5\rightarrow0.2$ via the COSMOS survey. Now with deeper and more homogeneous near-infrared coverage exploited by the COSMOS2020 catalog, we leverage the large 1.27 deg$^{2}$ effective area to improve sample statistics and understand cosmic variance particularly for rare, massive galaxies and push to higher redshifts with greater confidence and mass completeness than previous studies. We divide the total stellar mass function into star-forming and quiescent sub-samples through $NUVrJ$ color-color selection. Measurements are then fitted with Schechter functions to infer the intrinsic SMF, the evolution of its key parameters, and the cosmic stellar mass density out to $z=7.5$. We find a smooth, monotonic evolution in the galaxy SMF since $z=7.5$, in agreement with previous studies. The number density of star-forming systems seems to have undergone remarkably consistent growth spanning four decades in stellar mass from $z=7.5\rightarrow2$ whereupon high-mass systems become predominantly quiescent (i.e. downsizing). An excess of massive systems at $z\sim2.5-5.5$ with strikingly red colors, some newly identified, increase the observed number densities to the point where the SMF cannot be reconciled with a Schechter function. Systematics including cosmic variance and/or AGN contamination are unlikely to fully explain this excess, and so we speculate that there may be contributions from dust-obscured objects similar to those found in FIR surveys. (abridged)

astro-ph.GA↗

arXiv2023

Review of medical data analysis based on spiking neural networks

Medical data mainly includes various types of biomedical signals and medical images, which can be used by professional doctors to make judgments on patients' health conditions. However, the interpretation of medical data requires a lot of human cost and there may be misjudgments, so many scholars use neural networks and deep learning to classify and study medical data, which can improve the efficiency and accuracy of doctors and detect diseases early for early diagnosis, etc. Therefore, it has a wide range of application prospects. However, traditional neural networks have disadvantages such as high energy consumption and high latency (slow computation speed). This paper presents recent research on signal classification and disease diagnosis based on a third-generation neural network, the spiking neuron network, using medical data including EEG signals, ECG signals, EMG signals and MRI images. The advantages and disadvantages of pulsed neural networks compared with traditional networks are summarized and its development orientation in the future is prospected.

cs.NE↗

arXiv2023

The Gas Morphology of Nearby Star-Forming Galaxies

The morphology of a galaxy stems from secular and environmental processes during its evolutionary history. Thus galaxy morphologies have been a long used tool to gain insights on galaxy evolution. We visually classify morphologies on cloud-scales based on the molecular gas distribution of a large sample of 79 nearby main-sequence galaxies, using 1'' resolution CO(2-1) ALMA observations taken as part of the PHANGS survey. To do so, we devise a morphology classification scheme for different types of bars, spiral arms (grand-design, flocculent, multi-arm and smooth), rings (central and non-central rings) similar to the well-established optical ones, and further introduce bar lane classes. In general, our cold gas based morphologies agree well with the ones based on stellar light. Both our bars as well as grand-design spiral arms are preferentially found at the higher mass end of our sample. Our gas-based classification indicates a potential for misidentification of unbarred galaxies in the optical when massive star formation is present. Central or nuclear rings are present in a third of the sample with a strong preferences for barred galaxies (59%). As stellar bars are present in 45$\pm$5% of our sample galaxies, we explore the utility of molecular gas as tracer of bar lane properties. We find that more curved bar lanes have a shorter radial extent in molecular gas and reside in galaxies with lower molecular to stellar mass ratios than those with straighter geometries. Galaxies display a wide range of CO morphology, and this work provides a catalogue of morphological features in a representative sample of nearby galaxies.

astro-ph.GA↗

arXiv2023

NIKA2 Cosmological Legacy Survey: Survey Description and Galaxy Number Counts

Aims. Deep millimeter surveys are necessary to probe the dust-obscured galaxies at high redshift. We conducted a large observing program at 1.2 and 2 mm with the NIKA2 camera installed on the IRAM 30-meter telescope. This NIKA2 Cosmological Legacy Survey (N2CLS) covers two emblematic fields: GOODS-N and COSMOS. We introduce the N2CLS survey and present new 1.2 and 2 mm number count measurements based on the tiered N2CLS observations from October 2017 to May 2021. Methods. We develop an end-to-end simulation that combines an input sky model with the instrument noise and data reduction pipeline artifacts. This simulation is used to compute the sample purity, flux boosting, pipeline transfer function, completeness, and effective area of the survey. We used the 117 deg$^2$ SIDES simulations as the sky model, which include the galaxy clustering. Our formalism allows us to correct the source number counts to obtain galaxy number counts, the difference between the two being due to resolution effects caused by the blending of several galaxies inside the large beam of single-dish instruments. Results. The N2CLS-May2021 survey reaches an average 1-$σ$ noise level of 0.17 and 0.048 mJy on GOODS-N over 159 arcmin$^2$, and 0.46 and 0.14 mJy on COSMOS over 1010 arcmin$^2$, at 1.2 and 2 mm, respectively. For a purity threshold of 80%, we detect 120 and 67 sources in GOODS-N and 195 and 76 sources in COSMOS, at 1.2 and 2 mm, respectively. Our measurement connects the bright single-dish to the deep interferometric number counts. After correcting for resolution effects, our results reconcile the single-dish and interferometric number counts and are further accurately compared with model predictions.

astro-ph.GA↗

arXiv2023

The Gas Mass Reservoir of Quiescent Galaxies at Cosmic Noon

We present a 1.1mm stacking analysis of moderately massive (log($M_{*}$/$M_{\odot}$) = 10.7 $\pm$ 0.2) quiescent galaxies (QGs) at $\langle z\rangle \sim1.5$, searching for cold dust continuum emission, an excellent tracer of dust and gas mass. Using both the recent GOODS-ALMA survey as well as the full suite of ALMA Band-6 ancillary data in the GOODS-S field, we report the tentative detection of dust continuum equivalent of dust mass log($M_{dust}$/$M_{\odot}$) = 7.47 $\pm$ 0.13 and gas mass log($M_{gas}$/$M_{\odot}$) = 9.42 $\pm$ 0.14. The emerging gas fraction is $f_{gas}$ = 5.3 $\pm$ 1.8%, consistent with the results of previous stacking analyses based on lower resolution sub(mm) observations. Our results support the scenario where high-z QGs have an order of magnitude larger $f_{gas}$ compared to their local counterparts and have experienced quenching with a non negligible gas reservoir in their interstellar medium - i.e. with gas retention. Subsequent analysis yields an anti-correlation between the $f_{gas}$ and the stellar mass of QGs, especially in the high mass end where galaxies reside in the most massive haloes. The $f_{gas}$ - $M_{*}$ anti-correlation promotes the selection bias as a possible solution to the tension between the stacking results pointing towards gas retention in high-z QGs of moderate $M_{*}$ and the studies of individual targets that favour a fully depleted ISM in massive (log($M_{*}$/$M_{\odot}$) high-z QGs.

astro-ph.GA↗

arXiv2022

A super-linear "radio-AGN main sequence'' links mean radio-AGN power and galaxy stellar mass since z$\sim$3

Mapping the average AGN luminosity across galaxy populations and over time encapsulates important clues on the interplay between supermassive black hole (SMBH) and galaxy growth. This paper presents the demography, mean power and cosmic evolution of radio AGN across star-forming galaxies (SFGs) of different stellar masses (${M_{*}}$). We exploit deep VLA-COSMOS 3 GHz data to build the rest-frame 1.4 GHz AGN luminosity functions at 0.1$\leq$$z$$\leq$4.5 hosted in SFGs. Splitting the AGN luminosity function into different ${M_{*}}$ bins reveals that, at all redshifts, radio AGN are both more frequent and more luminous in higher ${M_*}$ than in lower ${M_*}$ galaxies. The cumulative kinetic luminosity density exerted by radio AGN in SFGs peaks at $z$$\sim$2, and it is mostly driven by galaxies with 10.5$\leq$$\log$(${M_{*}}$/${M_{\odot}}$)$<$11. Averaging the cumulative radio AGN activity across all SFGs at each (${M_{*}}$,$z$) results in a "radio-AGN main sequence" that links the time-averaged radio-AGN power $\langle$$L_{1.4}^{AGN}$$\rangle$ and galaxy stellar mass, in the form: $\log$$\langle$[$L_{1.4}^{AGN}$/ W Hz$^{-1}]\rangle$ = (20.97$\pm$0.16) + (2.51$\pm$0.34)$\cdot$$\log$(1+$z$) + (1.41$\pm$0.09)$\cdot$($\log$[${M_{*}}$/${M_{\odot}}$] -10). The super-linear dependence on ${M_{*}}$, at fixed redshift, suggests enhanced radio-AGN activity in more massive SFGs, as compared to star formation. We ascribe this enhancement to both a higher radio AGN duty cycle and a brighter radio-AGN phase in more massive SFGs. A remarkably consistent ${M_{*}}$ dependence is seen for the evolving X-ray AGN population in SFGs. This similarity is interpreted as possibly driven by secular cold gas accretion fueling both radio and X-ray AGN activity in a similar fashion over the galaxy's lifetime.

astro-ph.GA↗

arXiv2022

RC100: Rotation Curves of 100 Massive Star-Forming Galaxies at z=0.6-2.5 Reveal Little Dark Matter on Galactic Scales

We analyze Ha or CO rotation curves (RCs) extending out to several galaxy effective radii for 100 massive, large, star-forming disk galaxies (SFGs) across the peak of cosmic galaxy star formation (z~0.6-2.5), more than doubling the previous sample presented by Genzel et al. (2020) and Price et al. (2021). The observations were taken with SINFONI and KMOS integral-field spectrographs at ESO-VLT, LUCI at LBT, NOEMA at IRAM, and ALMA. We fit the major axis kinematics with beam-convolved, forward models of turbulent rotating disks with bulges embedded in dark matter (DM) halos, including the effects of pressure support. The fraction of dark to total matter within the disk effective radius ($R_e ~ 5 kpc$), $f_DM (R_e)=V_{DM}^2 (R_e)/V_{circ}^2 (R_e)$, decreases with redshift: At z~1 (z~2) the median DM fraction is $0.38\pm 0.23$ ($0.27\pm 0.18$), and a third (half) of all galaxies are "maximal" disks with $f_{DM} (R_e)<0.28$. Dark matter fractions correlate inversely with the baryonic surface density, and the low DM fractions require a flattened, or cored, inner DM density distribution. At z~2 there is ~40% less dark matter mass on average within $R_e$ compared to expected values based on cosmological stellar-mass halo-mass relations. The DM deficit is more evident at high star formation rate (SFR) surface densities ($Σ_{SFR}>2.5 M_{\odot} yr^{-1} kpc^{-2}$) and galaxies with massive bulges ($M_{bulge}>10^{10} M_{\odot}$). A combination of stellar or active galactic nucleus (AGN) feedback, and/or heating due to dynamical friction, either from satellite accretion or clump migration, may drive the DM from cuspy into cored mass distributions. The observations plausibly indicate an efficient build-up of massive bulges and central black holes at z~2 SFGs.

astro-ph.GA↗

arXiv2022

COSMOS2020: UV selected galaxies at $z\geq7.5$

This paper presents a new search for $z\geq7.5$ galaxies using the COSMOS2020 photometric catalogues. Finding galaxies at the reionization epoch through deep imaging surveys remains observationally challenging. The larger area covered by ground-based surveys like COSMOS enables the discovery of the brightest galaxies at these high redshifts. Covering $1.4$deg$^2$, our COSMOS catalogues were constructed from the latest UltraVISTA data release (DR4) combined with the final Spitzer/IRAC COSMOS images and the Hyper-Suprime-Cam Subaru Strategic Program DR2 release. We identify $17$ new $7.5<z<10$ candidate sources, and confirm $15$ previously published candidates. Using deblended photometry extracted by fitting surface brightness models on multi-band images, we select four candidates which would be rejected using fixed aperture photometry. We test the robustness of all our candidates by comparing six different photometric redshift estimates. Finally, we compute the galaxy UV luminosity function in three redshift bins centred at $z=8,9,10$. We find no clear evolution of the number density of the brightest galaxies $M_\text{UV}<-21.5$, in agreement with previous works. Rapid changes in the quenching efficiency or attenuation by dust could explain such lack of evolution between $z\sim 8$ and $z\sim 9. A spectroscopic confirmation of the redshifts, already planned with JWST and the Keck telescopes, will be essential to confirm our results.

astro-ph.GA↗

arXiv2022

Variations in the $Σ_{\rm SFR} {-} Σ_{\rm mol} {-} Σ_{\rm \star}$ plane across galactic environments in PHANGS galaxies

There exists some consensus that stellar mass surface density ($Σ_{*}$) and molecular gas mass surface density ($Σ_{\rm mol}$) are the main quantities responsible for locally setting the star formation rate. This regulation is inferred from locally resolved scaling relations between these two quantities and the star formation rate surface density ($Σ_{\rm SFR}$). However, the universality of these relations is debated. Here, we probe the interplay between these three quantities across different galactic environments at a spatial resolution of 150 pc. We perform a hierarchical Bayesian linear regression to find the best set of parameters $C_{*}$, $C_{\rm mol}$, and $C_{\rm norm}$ that describe the star-forming plane conformed by these quantities, such that $\log Σ_{\rm SFR} = C_{*} \log Σ_{*} + C_{\rm mol} \log Σ_{\rm mol} + C_{\rm norm}$, and explore variations in the determined parameters across galactic environments, focusing our analysis on the $C_{*}$ and $C_{\rm mol}$ slopes. We find signs of variations in the posterior distributions of $C_{*}$ and $C_{\rm mol}$ across different galactic environments. Bars show the most negative value of $C_{*}$, a sign of longer depletion times, while spiral arms show the highest $C_{*}$ among all environments. We conclude that systematic variations in the interplay of $Σ_{*}$, $Σ_{\rm mol}$ and $Σ_{\rm SFR}$ across galactic environments exist at a spatial resolution of 150 pc, and we interpret these variations as produced by an additional mechanism regulating the formation of stars that is not captured by either $Σ_{*}$ or $Σ_{\rm mol}$. We find that these variations correlate with changes in the star formation efficiency across environments, which could be linked to the dynamical state of the gas that prevents it from collapsing and forming stars, or to changes in the molecular gas fraction.

astro-ph.GA↗

arXiv2022

Scintillator Tile Batch Test of CEPC AHCAL

Hadron calorimeter (HCAL) is an essential sub-detector of the baseline detector system for Circular Electron Positron Collider (CEPC). We plan to build an Analog Hadron CALorimeter (AHCAL) prototype based on the Particle Flow Algorithm (PFA). The AHCAL of CEPC uses steel as absorber and scintillator tiles read out by Silicon Photo-Multipliers (SiPMs) as sensitive medium. The energy linearity and resolution of the calorimeter depends on the light yield uniformity of sensitive medium. It is essential to qualify the entire detector production in order to select scintillator tiles with light yield uniform within 10\%. An automated batch test platform has been designed with 144 channels, an automated 3D servo motor. The paper summarizes the tests performed on more than 15000 scintillator tiles. The measured light yield, corrected for the set-up response non-uniformity, is around 12.9 p.e. . About 91.6\% of scintillators (14219 pieces) are qualified within 10\% of light yield window.

physics.ins-det↗

arXiv2022

Evidence for Cold-stream to Hot-accretion Transition as Traced by Lyα Emission from Groups and Clusters at 2 < z < 3.3

We present Keck Cosmic Web Imager (KCWI) observations of giant Lya halos surrounding 9 galaxy groups and clusters at 2<z<3.3, including five new detections and one upper limit. We find observational evidence for the cold-stream to hot-accretion transition predicted by theory by measuring a decrease in the ratio between the spatially extended Lya luminosity and the expected baryonic accretion rate (BAR), with increasing elongation above the transition mass Mstream). This implies a modulation of the share of BAR that remains cold diminishing quasi-linearly (logarithmic slope of 0.97+-0.19, 5 sigma significance) with the halo to Mstream mass ratio. The integrated star-formation rates (SFRs) and AGN bolometric luminosities display a potentially consistent decrease, albeit significant only at 2.6 sigma and 1.3 sigma, respectively. The higher scatter in these tracers suggests the Lya emission might be mostly a direct product of cold accretion in these structures rather than indirect, mediated by outflows and photo-ionization from SFR and AGNs; this is also supported by energetics considerations. Below Mstream (cold-stream regime) we measure LLya/BAR=10^{40.51+-0.16}~erg/s/Msun*yr, consistent with predictions, and SFR/BAR=10^{-0.54+-0.23}: on average 30_{-10}^{+20}% of the cold streams go into stars. Above Mstream (hot-accretion regime), LLya is set by Mstream (within 0.2~dex scatter in our sample), independent of the halo mass but rising tenfold from z=2 to 3.

astro-ph.CO↗

arXiv2021

Star formation scaling relations at ~100 pc from PHANGS: Impact of completeness and spatial scale

Aims: The complexity of star formation at the physical scale of molecular clouds is not yet fully understood. We investigate the mechanisms regulating the formation of stars in different environments within nearby star-forming galaxies from the PHANGS sample. Methods: Integral field spectroscopic data and radio-interferometric observations of 18 galaxies were combined to explore the existence of the resolved star formation main sequence (rSFMS), resolved Kennicutt-Schmidt relation (rKS), and resolved molecular gas main sequence (rMGMS), and we derived their slope and scatter at spatial resolutions from 100 pc to 1 kpc (under various assumptions). Results: All three relations were recovered at the highest spatial resolution (100 pc). Furthermore, significant variations in these scaling relations were observed across different galactic environments. The exclusion of non-detections has a systematic impact on the inferred slope as a function of the spatial scale. Finally, the scatter of the $Σ_\mathrm{mol. gas + stellar}$ versus $Σ_\mathrm{SFR}$ correlation is smaller than that of the rSFMS, but higher than that found for the rKS. Conclusions: The rMGMS has the tightest relation at a spatial scale of 100 pc (scatter of 0.34 dex), followed by the rKS (0.41 dex) and then the rSFMS (0.51 dex). This is consistent with expectations from the timescales involved in the evolutionary cycle of molecular clouds. Surprisingly, the rKS shows the least variation across galaxies and environments, suggesting a tight link between molecular gas and subsequent star formation. The scatter of the three relations decreases at lower spatial resolutions, with the rKS being the tightest (0.27 dex) at a spatial scale of 1 kpc. Variation in the slope of the rSFMS among galaxies is partially due to different detection fractions of $Σ_\mathrm{SFR}$ with respect to $Σ_\mathrm{stellar}$.

astro-ph.GA↗

arXiv2021

Dense molecular gas properties on 100 pc scales across the disc of NGC 3627

It is still poorly constrained how the densest phase of the interstellar medium varies across galactic environment. A large observing time is required to recover significant emission from dense molecular gas at high spatial resolution, and to cover a large dynamic range of extragalactic disc environments. We present new NOrthern Extended Millimeter Array (NOEMA) observations of a range of high critical density molecular tracers (HCN, HNC, HCO+) and CO isotopologues (13CO, C18O) towards the nearby (11.3 Mpc), strongly barred galaxy NGC 3627. These observations represent the current highest angular resolution (1.85"; 100 pc) map of dense gas tracers across a disc of a nearby spiral galaxy, which we use here to assess the properties of the dense molecular gas, and their variation as a function of galactocentric radius, molecular gas, and star formation. We find that the HCN(1-0)/CO(2-1) integrated intensity ratio does not correlate with the amount of recent star formation. Instead, the HCN(1-0)/CO(2-1) ratio depends on the galactic environment, with differences between the galaxy centre, bar, and bar end regions. The dense gas in the central 600 pc appears to produce stars less efficiently despite containing a higher fraction of dense molecular gas than the bar ends where the star formation is enhanced. In assessing the dynamics of the dense gas, we find the HCN(1-0) and HCO+(1-0) emission lines showing multiple components towards regions in the bar ends that correspond to previously identified features in CO emission. These features are co-spatial with peaks of Halpha emission, which highlights that the complex dynamics of this bar end region could be linked to local enhancements in the star formation.

astro-ph.GA↗

arXiv2021

New Constraints on the $^{12}$CO(2-1)/(1-0) Line Ratio Across Nearby Disc Galaxies

Both the CO(2-1) and CO(1-0) lines are used to trace the mass of molecular gas in galaxies. Translating the molecular gas mass estimates between studies using different lines requires a good understanding of the behaviour of the CO(2-1)-to-CO(1-0) ratio, $R_{21}$. We compare new, high quality CO(1-0) data from the IRAM 30-m EMPIRE survey to the latest available CO(2-1) maps from HERACLES, PHANGS-ALMA, and a new IRAM 30-m M51 Large Program. This allows us to measure $R_{21}$ across the full star-forming disc of nine nearby, massive, star-forming spiral galaxies at 27" (${\sim} 1{-}2$ kpc) resolution. We find an average $R_{21} = 0.64\pm0.09$ when we take the luminosity-weighted mean of all individual galaxies. This result is consistent with the mean ratio for disc galaxies that we derive from single-pointing measurements in the literature, $R_{\rm 21, lit}~=~0.59^{+0.18}_{-0.09}$. The ratio shows weak radial variations compared to the point-to-point scatter in the data. In six out of nine targets the central enhancement in $R_{21}$ with respect to the galaxy-wide mean is of order $\sim 10{-}20\%$. We estimate an azimuthal scatter of $\sim$20% in $R_{21}$ at fixed galactocentric radius but this measurement is limited by our comparatively coarse resolution of 1.5 kpc. We find mild correlations between $R_{21}$ and CO brightness temperature, IR intensity, 70-to-160$ μ$m ratio, and IR-to-CO ratio. All correlations indicate that $R_{21}$ increases with gas surface density, star formation rate surface density, and the interstellar radiation field.

astro-ph.GA↗

arXiv2021

Privacy Preserving Distributed Machine Learning with Federated Learning

Edge computing and distributed machine learning have advanced to a level that can revolutionize a particular organization. Distributed devices such as the Internet of Things (IoT) often produce a large amount of data, eventually resulting in big data that can be vital in uncovering hidden patterns, and other insights in numerous fields such as healthcare, banking, and policing. Data related to areas such as healthcare and banking can contain potentially sensitive data that can become public if they are not appropriately sanitized. Federated learning (FedML) is a recently developed distributed machine learning (DML) approach that tries to preserve privacy by bringing the learning of an ML model to data owners'. However, literature shows different attack methods such as membership inference that exploit the vulnerabilities of ML models as well as the coordinating servers to retrieve private data. Hence, FedML needs additional measures to guarantee data privacy. Furthermore, big data often requires more resources than available in a standard computer. This paper addresses these issues by proposing a distributed perturbation algorithm named as DISTPAB, for privacy preservation of horizontally partitioned data. DISTPAB alleviates computational bottlenecks by distributing the task of privacy preservation utilizing the asymmetry of resources of a distributed environment, which can have resource-constrained devices as well as high-performance computers. Experiments show that DISTPAB provides high accuracy, high efficiency, high scalability, and high attack resistance. Further experiments on privacy-preserving FedML show that DISTPAB is an excellent solution to stop privacy leaks in DML while preserving high data utility.

cs.DB↗

arXiv2021

Three Lyman-alpha emitting filaments converging to a massive galaxy group at z=2.91: discussing the case for cold gas infall

We have discovered a 300kpc-wide giant Lya nebula centered on the massive galaxy group RO-1001 at z=2.91 in the COSMOS field. Keck Cosmic Web Imager observations reveal three cold gas filaments converging into the center of the potential well of its ~4x10^13Msun dark matter halo, hosting 1200Msun/yr of star formation as probed by ALMA and NOEMA observations. The nebula morphological and kinematics properties and the prevalence of blueshifted components in the Lya spectra are consistent with a scenario of gas accretion. The upper limits on AGN activity and overall energetics favor gravity as the primary Lya powering source and infall as the main source of gas flows to the system. Although interpretational difficulties remain, with outflows and likely also photoionization with ensuing recombination still playing a role, this finding provides arguably an ideal environment to quantitatively test models of cold gas accretion and galaxy feeding inside an actively star-forming massive halo at high redshift.

astro-ph.GA↗

arXiv2021

Content-Aware Speaker Embeddings for Speaker Diarisation

Recent speaker diarisation systems often convert variable length speech segments into fixed-length vector representations for speaker clustering, which are known as speaker embeddings. In this paper, the content-aware speaker embeddings (CASE) approach is proposed, which extends the input of the speaker classifier to include not only acoustic features but also their corresponding speech content, via phone, character, and word embeddings. Compared to alternative methods that leverage similar information, such as multitask or adversarial training, CASE factorises automatic speech recognition (ASR) from speaker recognition to focus on modelling speaker characteristics and correlations with the corresponding content units to derive more expressive representations. CASE is evaluated for speaker re-clustering with a realistic speaker diarisation setup using the AMI meeting transcription dataset, where the content information is obtained by performing ASR based on an automatic segmentation. Experimental results showed that CASE achieved a 17.8% relative speaker error rate reduction over conventional methods.

cs.SD↗