arXiv subjects

Lei Bai

Publications and source records attributed to Lei Bai.

At least 307 records · Page 17Linked to original sources

arXiv2020

Adaptive Graph Convolutional Recurrent Network for Traffic Forecasting

Modeling complex spatial and temporal correlations in the correlated time series data is indispensable for understanding the traffic dynamics and predicting the future status of an evolving traffic system. Recent works focus on designing complicated graph neural network architectures to capture shared patterns with the help of pre-defined graphs. In this paper, we argue that learning node-specific patterns is essential for traffic forecasting while the pre-defined graph is avoidable. To this end, we propose two adaptive modules for enhancing Graph Convolutional Network (GCN) with new capabilities: 1) a Node Adaptive Parameter Learning (NAPL) module to capture node-specific patterns; 2) a Data Adaptive Graph Generation (DAGG) module to infer the inter-dependencies among different traffic series automatically. We further propose an Adaptive Graph Convolutional Recurrent Network (AGCRN) to capture fine-grained spatial and temporal correlations in traffic series automatically based on the two modules and recurrent networks. Our experiments on two real-world traffic datasets show AGCRN outperforms state-of-the-art by a significant margin without pre-defined graphs about spatial connections.

cs.LG↗

arXiv2020

Knowledge Adaption for Demand Prediction based on Multi-task Memory Neural Network

Accurate demand forecasting of different public transport modes(e.g., buses and light rails) is essential for public service operation.However, the development level of various modes often varies sig-nificantly, which makes it hard to predict the demand of the modeswith insufficient knowledge and sparse station distribution (i.e.,station-sparse mode). Intuitively, different public transit modes mayexhibit shared demand patterns temporally and spatially in a city.As such, we propose to enhance the demand prediction of station-sparse modes with the data from station-intensive mode and designaMemory-Augmented Multi-taskRecurrent Network (MATURE)to derive the transferable demand patterns from each mode andboost the prediction of station-sparse modes through adaptingthe relevant patterns from the station-intensive mode. Specifically,MATUREcomprises three components: 1) a memory-augmentedrecurrent network for strengthening the ability to capture the long-short term information and storing temporal knowledge of eachtransit mode; 2) a knowledge adaption module to adapt the rele-vant knowledge from a station-intensive source to station-sparsesources; 3) a multi-task learning framework to incorporate all theinformation and forecast the demand of multiple modes jointly.The experimental results on a real-world dataset covering four pub-lic transport modes demonstrate that our model can promote thedemand forecasting performance for the station-sparse modes.

cs.AI↗

arXiv2020

Face to Purchase: Predicting Consumer Choices with Structured Facial and Behavioral Traits Embedding

Predicting consumers' purchasing behaviors is critical for targeted advertisement and sales promotion in e-commerce. Human faces are an invaluable source of information for gaining insights into consumer personality and behavioral traits. However, consumer's faces are largely unexplored in previous research, and the existing face-related studies focus on high-level features such as personality traits while neglecting the business significance of learning from facial data. We propose to predict consumers' purchases based on their facial features and purchasing histories. We design a semi-supervised model based on a hierarchical embedding network to extract high-level features of consumers and to predict the top-$N$ purchase destinations of a consumer. Our experimental results on a real-world dataset demonstrate the positive effect of incorporating facial information in predicting consumers' purchasing behaviors.

cs.CV↗

arXiv2020

Spectrum-Guided Adversarial Disparity Learning

It has been a significant challenge to portray intraclass disparity precisely in the area of activity recognition, as it requires a robust representation of the correlation between subject-specific variation for each activity class. In this work, we propose a novel end-to-end knowledge directed adversarial learning framework, which portrays the class-conditioned intraclass disparity using two competitive encoding distributions and learns the purified latent codes by denoising learned disparity. Furthermore, the domain knowledge is incorporated in an unsupervised manner to guide the optimization and further boosts the performance. The experiments on four HAR benchmark datasets demonstrate the robustness and generalization of our proposed methods over a set of state-of-the-art. We further prove the effectiveness of automatic domain knowledge incorporation in performance enhancement.

cs.LG↗

arXiv2020

Are You A Risk Taker? Adversarial Learning of Asymmetric Cross-Domain Alignment for Risk Tolerance Prediction

Most current studies on survey analysis and risk tolerance modelling lack professional knowledge and domain-specific models. Given the effectiveness of generative adversarial learning in cross-domain information, we design an Asymmetric cross-Domain Generative Adversarial Network (ADGAN) for domain scale inequality. ADGAN utilizes the information-sufficient domain to provide extra information to improve the representation learning on the information-insufficient domain via domain alignment. We provide data analysis and user model on two data sources: Consumer Consumption Information and Survey Information. We further test ADGAN on a real-world dataset with view embedding structures and show ADGAN can better deal with the class imbalance and unqualified data space than state-of-the-art, demonstrating the effectiveness of leveraging asymmetrical domain information.

cs.LG↗

arXiv2019

STG2Seq: Spatial-temporal Graph to Sequence Model for Multi-step Passenger Demand Forecasting

Multi-step passenger demand forecasting is a crucial task in on-demand vehicle sharing services. However, predicting passenger demand over multiple time horizons is generally challenging due to the nonlinear and dynamic spatial-temporal dependencies. In this work, we propose to model multi-step citywide passenger demand prediction based on a graph and use a hierarchical graph convolutional structure to capture both spatial and temporal correlations simultaneously. Our model consists of three parts: 1) a long-term encoder to encode historical passenger demands; 2) a short-term encoder to derive the next-step prediction for generating multi-step prediction; 3) an attention-based output module to model the dynamic temporal and channel-wise information. Experiments on three real-world datasets show that our model consistently outperforms many baseline methods and state-of-the-art models.

cs.LG↗

arXiv2018

Automatic Device Classification from Network Traffic Streams of Internet of Things

With the widespread adoption of Internet of Things (IoT), billions of everyday objects are being connected to the Internet. Effective management of these devices to support reliable, secure and high quality applications becomes challenging due to the scale. As one of the key cornerstones of IoT device management, automatic cross-device classification aims to identify the semantic type of a device by analyzing its network traffic. It has the potential to underpin a broad range of novel features such as enhanced security (by imposing the appropriate rules for constraining the communications of certain types of devices) or context-awareness (by the utilization and interoperability of IoT devices and their high-level semantics) of IoT applications. We propose an automatic IoT device classification method to identify new and unseen devices. The method uses the rich information carried by the traffic flows of IoT networks to characterize the attributes of various devices. We first specify a set of discriminating features from raw network traffic flows, and then propose a LSTM-CNN cascade model to automatically identify the semantic type of a device. Our experimental results using a real-world IoT dataset demonstrate that our proposed method is capable of delivering satisfactory performance. We also present interesting insights and discuss the potential extensions and applications.

cs.NI↗

arXiv2014

The Inside-out Growth of the Most Massive Galaxies at 0.3<z<0.9

We study the surface brightness profiles of a sample of brightest cluster galaxies (BCGs) with 0.3<z<0.9. The BCGs are selected from the first Red-sequence Cluster Survey and an X-ray cluster survey. The surface brightness profiles of the BCGs are measured using HST ACS images, and most of them can be well modeled by a Sersic profile with index ~ 6 and half-light radius ~ 30 kpc. Although the single Sersic model fits the profiles well, we argue that the systematics in the sky background measurement and the coupling between the model parameters make the comparison of the model parameters ambiguous. Direct comparison of the BCG profiles, on the other hand, has revealed an inside-out growth for these most massive galaxies: as the mass of a BCG increases, the central mass density of the galaxy increases slowly ($ρ_{1kpc} \propto M*^{0.39}$), while the slope of the outer profile grows continuously shallower ($α\propto M*^{-2.5}$). Such a fashion of growth continues down to the less massive early-type galaxies (ETGs), without apparent distinction between BCGs and non-BCGs. For the very massive ETGs and BCGs, the slope of the Kormendy relation starts to trace the slope of the surface brightness profiles and becomes insensitive to subtle profile evolution. These results are generally consistent with dry mergers being the major driver of the mass growth for BCGs and massive ETGs. We also find strong correlations between the richness of clusters and the properties of BCGs: the more massive the clusters are, the more massive the BCGs ($M*_{bcg} \propto M_{clusters}^{0.6}$) and the shallower their surface brightness profiles. After taking into account the bias in the cluster samples, we find the masses of the BCGs have grown by at least a factor of 1.5 from z=0.5 to z=0, in contrast to the previous findings of no evolution. Such an evolution validates the expectation from the LCDM model.

astro-ph.GA↗

arXiv2013

The IMACS Cluster Building Survey. I. Description of the Survey and Analysis Methods

The IMACS Cluster Building Survey uses the wide field spectroscopic capabilities of the IMACS spectrograph on the 6.5m Baade Telescope to survey the large-scale environment surrounding rich intermediate-redshift clusters of galaxies. The goal is to understand the processes which may be transforming star-forming field galaxies into quiescent cluster members as groups and individual galaxies fall into the cluster from the surrounding supercluster. This first paper describes the survey: the data taking and reduction methods. We provide new calibrations of star formation rates derived from optical and infrared spectroscopy and photometry. We demonstrate that there is a tight relation between the observed star formation rate per unit B luminosity, and the ratio of the extinctions of the stellar continuum and the optical emission lines. With this, we can obtain accurate extinction-corrected colors of galaxies. Using these colors as well as other spectral measures, we determine new criteria for the existence of ongoing and recent starbursts in galaxies.

astro-ph.CO↗

arXiv2012

The Suppression of Star Formation and the Effect of Galaxy Environment in Low-Redshift Galaxy Groups

Understanding the interaction between galaxies and their surroundings is central to building a coherent picture of galaxy evolution. Here we use GALEX imaging of a statistically representative sample of 23 galaxy groups at z=0.06 to explore how local and global group environment affect the UV properties and dust-corrected star formation rates of their member galaxies. The data provide star formation rates out to beyond 2R_200 in all groups, down to a completeness limit and limiting galaxy stellar mass of 0.06 M_sun/yr and 10^8 M_sun, respectively. At fixed galaxy stellar mass, we find that the fraction of star-forming group members is suppressed relative to the field out to an average radius of R ~ 1.5 Mpc ~ 2R_200, mirroring results for massive clusters. For the first time we also report a similar suppression of the specific star formation rate within such galaxies, on average by 40% relative to the field, thus directly revealing the impact of the group environment in quenching star formation within infalling galaxies. At fixed galaxy density and stellar mass, this suppression is stronger in more massive groups, implying that both local and global group environment play a role in quenching. The results favor an average quenching timescale of at least ~2 Gyr and strongly suggest that a combination of tidal interactions and starvation is responsible. Despite their past and ongoing quenching, galaxy groups with more than four members still account for at least ~25% of the total UV output in the nearby universe.

astro-ph.CO↗

arXiv2010

Witnessing the Formation of a Brightest Cluster Galaxy in a Nearby X-ray Cluster

The central dominant galaxies in galaxy clusters constitute the most massive and luminous galaxies in the Universe. Despite this, the formation of these brightest cluster galaxies (BCGs) and the impact of this on the surrounding cluster environment remain poorly understood. Here we present multi-wavelength observations of the nearby poor X-ray cluster MZ 10451, in which both processes can be studied in unprecedented detail. Chandra observations of the intracluster medium (ICM) in the cluster core, which harbors two optically bright early-type galaxies in the process of merging, show that the system has retained a cool core and a central metal excess. This suggests that any merger-induced ICM heating and mixing remain modest at this stage. Tidally stripped stars seen around either galaxy likely represent an emerging intracluster light component, and the central ICM abundance enhancement may have a prominent contribution from in situ enrichment provided by these stars. The smaller of the merging galaxies shows evidence for having retained a hot gas halo, along with tentative evidence for some obscured star formation, suggesting that not all BCG major mergers at low redshift are completely dissipationless. Both galaxies are slightly offset from the peak of the ICM emission, with all three lying on an axis that roughly coincides with the large-scale elongation of the ICM. Our data are consistent with a picture in which central BCGs are built up by mergers close to the cluster core, by galaxies infalling on radial orbits aligned with the cosmological filaments feeding the cluster.

astro-ph.CO↗

arXiv2010

The First Mid-IR View of the Star-forming Properties of Nearby Galaxy Groups

We present the first mid-IR study of galaxy groups in the nearby Universe based on Spitzer MIPS observations of a sample of nine redshift-selected groups from the XMM-IMACS (XI) project, at z=0.06. We find that on average the star-forming (SF) galaxy fraction in the groups is about 30% lower than the value in the field and 30% higher than in clusters. The SF fractions do not show any systematic dependence on group velocity dispersion, total stellar mass, or the presence of an X-ray emitting intragroup medium, but a weak anti-correlation is seen between SF fraction and projected galaxy density. However, even in the densest regions, the SF fraction in groups is still higher than that in cluster outskirts, suggesting that preprocessing of galaxies in group environments is not sufficient to explain the much lower SF fraction in clusters. The typical specific star formation rates (SFR/M*) of SF galaxies in groups are similar to those in the field across a wide range of stellar mass (M*>10^9.6 msun), favoring a quickly acting mechanism that suppresses star formation to explain the overall smaller fraction of SF galaxies in groups. If galaxy-galaxy interactions are responsible, then the extremely low starburst galaxy fraction (<1%) implies a short timescale (~0.1 Gyr) for any merger-induced starburst stage. Comparison to two rich clusters shows that clusters contain a population of massive SF galaxies with very low SFR (14% of all the galaxies with M*>10^10 Msun), possibly as a consequence of ram pressure stripping being less efficient in removing gas from more massive galaxies.

astro-ph.CO↗

arXiv2009

A Spectroscopically Confirmed Excess of 24 micron Sources in a Super Galaxy Group at z=0.37: Enhanced Dusty Star Formation Relative to the Cluster and Field Environment

To trace how dust-obscured star formation varies with environment, we compare the fraction of 24 micron sources in a super galaxy group to the field and a rich galaxy cluster at z~0.35. We draw on multi-wavelength observations that combine Hubble, Chandra, and Spitzer imaging with extensive optical spectroscopy (>1800 redshifts) to isolate galaxies in each environment and thus ensure a uniform analysis. We focus on the four galaxy groups in supergroup 1120-12 that will merge to form a galaxy cluster comparable in mass to Coma. We find that 1) the fraction of supergroup galaxies with SFR(IR)>3 Msun/yr is four times higher than in the cluster (32% vs. 7%); 2) the supergroup's infrared luminosity function confirms that it has a higher density of IR members compared to the cluster and includes bright IR sources not found in galaxy clusters at z<0.35; and 3) there is a strong trend of decreasing IR fraction with increasing galaxy density, i.e. an IR-density relation, not observed in the cluster. These dramatic differences are surprising because the early-type fraction in the supergroup is already as high as in clusters, i.e. the timescales for morphological transformation cannot be strongly coupled to when the star formation is completely quenched. The supergroup has a significant fraction (~17%) of luminous, low-mass, IR members that are outside the group cores (R>0.5 Mpc); once their star formation is quenched, most will evolve into faint red galaxies. Our analysis indicates that the supergroup's 24 micron population also differs from that in the field: 1) despite the supergroup having twice the fraction of E/S0s as the field, the fraction of IR galaxies is comparable in both environments, and 2) the supergroup's IR luminosity function has a higher L(IR)* than that previously measured for the field.

astro-ph.CO↗

arXiv2009

Evolution of the Rate and Mode of Star Formation in Galaxies since z=0.7

We present the star formation rate (SFR) and starburst fraction (SBF) for a sample of field galaxies from the ICBS intermediate-redshift cluster survey. We use [O II] and Spitzer 24 micron fluxes to measure SFRs, and 24 micron fluxes and H-delta absorption to measure of SBFs, for both our sample and a present-epoch field sample from the Sloan Digital Sky Survey (SDSS) and Spitzer Wide-area Infrared Extragalactic (SWIRE) survey. We find a precipitous decline in the SFR since z=1, in agreement with other studies, as well as a corresponding rapid decline in the fraction of galaxies undergoing long-duration moderate-amplitude starbursts. We suggest that the change in both the rate and mode of star formation could result from the strong decrease since z=1 of gas available for star formation.

astro-ph.CO↗

arXiv2008

The IR Luminosity Functions of Rich Clusters

We present MIPS observations of the cluster A3266. About 100 spectroscopic cluster members have been detected at 24 micron. The IR luminosity function in A3266 is very similar to that in the Coma cluster down to the detection limit L_IR~10^43 ergs/s, suggesting a universal form of the bright end IR LF for local rich clusters with M~10^15 M_sun. The shape of the bright end of the A3266-Coma composite IR LF is not significantly different from that of nearby field galaxies, but the fraction of IR-bright galaxies (SFR > 0.2M_sun/yr) in both clusters increases with cluster-centric radius. The decrease of the blue galaxy fraction toward the high density cores only accounts for part of the trend; the fraction of red galaxies with moderate SFRs (0.2 < SFR < 1 M_sun/yr) also decreases with increasing galaxy density. These results suggest that for the IR bright galaxies, nearby rich clusters are distinguished from the field by a lower star-forming galaxy fraction, but not by a change in L*_IR. The composite IR LF of Coma and A3266 shows strong evolution when compared with the composite IR LF of two z~0.8 clusters, MS 1054 and RX J0152, with L*_IR \propto (1+z)^{3.2+/-0.7},Phi*_IR \propto (1+z)^{1.7+/-1.0}. This L*_IR evolution is indistinguishable from that in the field, and the Phi*_IR evolution is stronger, but still consistent with that in the field. The similarity of the evolution of bright-end IR LF in very different cluster and field environments suggests either this evolution is driven by the mechanism that works in both environments, or clusters continually replenish their star-forming galaxies from the field, yielding an evolution in the IR LF that is similar to the field. The mass-normalized integrated star formation rates (SFRs) of clusters within 0.5R_200 also evolve strongly with redshift, as (1+z)^5.3.

astro-ph↗

arXiv2008

The Late Stellar Assembly of Massive Cluster Galaxies Via Major Merging

We present multi-wavelength observations of the brightest galaxies in four X-ray luminous groups at z~0.37 that will merge to form a cluster comparable in mass to Coma. Ordered by increasing stellar mass, the four brightest group galaxies (BGGs) present a time sequence where BGG-1, 2, and 3 are in merging systems and BGG-4 is a massive remnant [M(stars)=6.7x10^(11) Msun]. BGG-1 and 2 have bright, gravitationally bound companions and BGG-3 has two nuclei separated by only 2.5 kpc, thus merging at z<0.5 increases the BGG mass by >40% (merging timescale<2 Gyr) and V-band luminosity by ~0.4 mag. The BGGs' rest-frame (B-V) colors correspond to stellar ages of >3 Gyr, and their tight scatter in (B-V) color [sigma(BV)=0.032] confirms they formed the bulk of their stars at z>0.9. Optical spectroscopy shows no signs of recent (<1.5 Gyr) or ongoing star formation. Only two BGGs are weakly detected at 24 microns, and X-ray and optical data indicate the emission in BGG-2 is due to an AGN. All four BGGs and their companions are early-type (bulge-dominated) galaxies, and they are embedded in diffuse stellar envelopes up to ~140 kpc across. The four BGG systems must evolve into the massive, red, early-type galaxies dominating local clusters. Our results show that: 1) massive galaxies in groups and clusters form via dissipationless merging; and 2) the group environment is critical for this process.

astro-ph↗

arXiv2008

The Flux Auto- and Cross-Correlation of the Lyman-alpha Forest. I. Spectroscopy of QSO Pairs with Arcminute Separations and Similar Redshifts

The Lyman-alpha forest has opened a new redshift regime for cosmological investigation. At z > 2 it provides a unique probe of cosmic geometry and an independent constraint on dark energy that is not subject to standard candle or ruler assumptions. In Paper I of this series on using the Lyman-alpha forest observed in pairs of QSOs for a new application of the Alcock-Paczynski test, we present and discuss the results of a campaign to obtain moderate-resolution spectroscopy (FWHM ~ 2.5 Angstroms) of the Lyman-alpha forest in pairs of QSOs with small redshift differences (Delta z < 0.25, z > 2.2) and arcminute separations (< 5'). This data set, composed of seven individual QSOs, 35 pairs, and one triplet, is also well-suited for future investigations of the coherence of Lyman-alpha absorbers on ~ 1 Mpc transverse scales and the transverse proximity effect. We note seven revisions for previously published QSO identifications and/or redshifts.

astro-ph↗

arXiv2007

Search for IR Emission from Intracluster Dust in A2029

We have searched for IR emission from the intracluster dust (ICD) in the galaxy cluster A2029. Weak signals of enhanced extended emission in the cluster are detected at both 24 and 70 micron. However, the signals are indistinguishable from the foreground fluctuations. The 24 versus 70 micron color map does not discriminate the dust emission in the cluster from the cirrus emission. After excluding the contamination from the point sources, we obtain upper limits for the extended ICD emission in A2029, 5 x 10^3 Jy/sr at 24 micron and 5 x 10^4 Jy/sr at 70 micron. The upper limits are generally consistent with the expectation from theoretical calculations and support a dust deficiency in the cluster compared to the ISM in our galaxy. Our results suggest that even with the much improved sensitivity of current IR telescopes, a clear detection of the IR emission from ICD may be difficult due to cirrus noise.

astro-ph↗