arXiv subjects

Xiaokai Chen

Publications and source records attributed to Xiaokai Chen.

23 records · Page 2Linked to original sources

arXiv2022

Strong Conformity and Assembly Bias: Towards a Physical Understanding of the Galaxy-Halo Connection in SDSS Clusters

Understanding the physical connection between cluster galaxies and massive haloes is key to mitigating systematic uncertainties in next-generation cluster cosmology. We develop a novel method to infer the level of conformity between the stellar mass of the brightest central galaxies~(BCGs) $M_*^{BCG}$ and the satellite richness $λ$, defined as their correlation coefficient $ρ_{cc}$ at fixed halo mass, using the abundance and weak lensing of SDSS clusters as functions of $M_*^{BCG}$ and $λ$. We detect a halo mass-dependent conformity as $ρ_{cc}{=}0.60{+}0.08\ln(M_h/3{\times}10^{14}M_{\odot}/h)$. The strong conformity successfully resolves the "halo mass equality" conundrum discovered in Zu et al. 2021 -- when split by $M_*^{BCG}$ at fixed $λ$, the low and high-$M_*^{BCG}$ clusters have the same average halo mass despite having a $0.34$ dex discrepancy in average $M_*^{BCG}$. On top of the best-fitting conformity model, we develop a cluster assembly bias~(AB) prescription calibrated against the CosmicGrowth simulation, and build a conformity+AB model for the cluster weak lensing measurements. Our model predicts that with a ${\sim}20\%$ lower halo concentration $c$, the low-$M_*^{BCG}$ clusters are ${\sim}10\%$ more biased than the high-$M_*^{BCG}$ systems, in excellent agreement with the observations. We also show that the observed conformity and assembly bias are unlikely due to projection effects. Finally, we build a toy model to argue that while the early-time BCG-halo co-evolution drives the $M_*^{BCG}$-$c$ correlation, the late-time dry merger-induced BCG growth naturally produces the $M_*^{BCG}$-$λ$ conformity despite the well-known anti-correlation between $λ$ and $c$. Our method paves the path towards simultaneously constraining cosmology and cluster formation with future cluster surveys.

astro-ph.CO↗

arXiv2021

Does Concentration Drive the Scatter in the Stellar-to-Halo Mass Relation of Galaxy Clusters?

Concentration is one of the key dark matter halo properties that could drive the scatter in the stellar-to-halo mass relation of massive clusters. We derive robust photometric stellar masses for a sample of brightest central galaxies (BCGs) in SDSS redMaPPer clusters at $0.17<z<0.3$, and split the clusters into two equal-halo mass subsamples by their BCG stellar mass $M_*$. The weak lensing profiles $ΔΣ$ of the two cluster subsamples exhibit different slopes on scales below 1 M$pc/h$. To interpret such discrepancy, we perform a comprehensive Bayesian modelling of the two $ΔΣ$ profiles by including different levels of miscentring effects between the two subsamples as informed by X-ray observations. We find that the two subsamples have the same average halo mass of $1.74 \times 10^{14} M_{\odot}/h$, but the concentration of the low-$M_*$ clusters is $5.87_{-0.60}^{+0.77}$, ${\sim}1.5σ$ smaller than that of their high-$M_*$ counterparts~($6.95_{-0.66}^{+0.78}$). Furthermore, both cluster weak lensing and cluster-galaxy cross-correlations indicate that the large-scale bias of the low-$M_*$, low-concentration clusters are ${\sim}10\%$ higher than that of the high-$M_*$, high-concentration systems, hence possible evidence of the cluster assembly bias effect. Our results reveal a remarkable physical connection between the stellar mass within 20{-}30 k$pc/h$, the dark matter mass within ${\sim}$ 200 k$pc/h$, and the cosmic overdensity on scales above 10 M$pc/h$, enabling a key observational test of theories of co-evolution between massive clusters and their central galaxies.

astro-ph.CO↗

arXiv2021

Boosting Share Routing for Multi-task Learning

Multi-task learning (MTL) aims to make full use of the knowledge contained in multi-task supervision signals to improve the overall performance. How to make the knowledge of multiple tasks shared appropriately is an open problem for MTL. Most existing deep MTL models are based on parameter sharing. However, suitable sharing mechanism is hard to design as the relationship among tasks is complicated. In this paper, we propose a general framework called Multi-Task Neural Architecture Search (MTNAS) to efficiently find a suitable sharing route for a given MTL problem. MTNAS modularizes the sharing part into multiple layers of sub-networks. It allows sparse connection among these sub-networks and soft sharing based on gating is enabled for a certain route. Benefiting from such setting, each candidate architecture in our search space defines a dynamic sparse sharing route which is more flexible compared with full-sharing in previous approaches. We show that existing typical sharing approaches are sub-graphs in our search space. Extensive experiments on three real-world recommendation datasets demonstrate MTANS achieves consistent improvement compared with single-task models and typical multi-task methods while maintaining high computation efficiency. Furthermore, in-depth experiments demonstrates that MTNAS can learn suitable sparse route to mitigate negative transfer.

cs.LG↗

arXiv2019

Not All Words are Equal: Video-specific Information Loss for Video Captioning

An ideal description for a given video should fix its gaze on salient and representative content, which is capable of distinguishing this video from others. However, the distribution of different words is unbalanced in video captioning datasets, where distinctive words for describing video-specific salient objects are far less than common words such as 'a' 'the' and 'person'. The dataset bias often results in recognition error or detail deficiency of salient but unusual objects. To address this issue, we propose a novel learning strategy called Information Loss, which focuses on the relationship between the video-specific visual content and corresponding representative words. Moreover, a framework with hierarchical visual representations and an optimized hierarchical attention mechanism is established to capture the most salient spatial-temporal visual information, which fully exploits the potential strength of the proposed learning strategy. Extensive experiments demonstrate that the ingenious guidance strategy together with the optimized architecture outperforms state-of-the-art video captioning methods on MSVD with CIDEr score 87.5, and achieves superior CIDEr score 47.7 on MSR-VTT. We also show that our Information Loss is generic which improves various models by significant margins.

cs.CV↗

arXiv2018

DenseImage Network: Video Spatial-Temporal Evolution Encoding and Understanding

Many of the leading approaches for video understanding are data-hungry and time-consuming, failing to capture the gist of spatial-temporal evolution in an efficient manner. The latest research shows that CNN network can reason about static relation of entities in images. To further exploit its capacity in dynamic evolution reasoning, we introduce a novel network module called DenseImage Network(DIN) with two main contributions. 1) A novel compact representation of video which distills its significant spatial-temporal evolution into a matrix called DenseImage, primed for efficient video encoding. 2) A simple yet powerful learning strategy based on DenseImage and a temporal-order-preserving CNN network is proposed for video understanding, which contains a local temporal correlation constraint capturing temporal evolution at multiple time scales with different filter widths. Extensive experiments on two recent challenging benchmarks demonstrate that our DenseImage Network can accurately capture the common spatial-temporal evolution between similar actions, even with enormous visual variations or different time scales. Moreover, we obtain the state-of-the-art results in action and gesture recognition with much less time-and-memory cost, indicating its immense potential in video representing and understanding.

cs.CV↗