arXiv subjects

Huan Lin

Publications and source records attributed to Huan Lin.

At least 37 records · Page 2Linked to original sources

arXiv2023

From Statistical Methods to Deep Learning, Automatic Keyphrase Prediction: A Survey

Keyphrase prediction aims to generate phrases (keyphrases) that highly summarizes a given document. Recently, researchers have conducted in-depth studies on this task from various perspectives. In this paper, we comprehensively summarize representative studies from the perspectives of dominant models, datasets and evaluation metrics. Our work analyzes up to 167 previous works, achieving greater coverage of this task than previous surveys. Particularly, we focus highly on deep learning-based keyphrase prediction, which attracts increasing attention of this task in recent years. Afterwards, we conduct several groups of experiments to carefully compare representative models. To the best of our knowledge, our work is the first attempt to compare these models using the identical commonly-used datasets and evaluation metric, facilitating in-depth analyses of their disadvantages and advantages. Finally, we discuss the possible research directions of this task in the future.

cs.CL↗

arXiv2023

WR-ONE2SET: Towards Well-Calibrated Keyphrase Generation

Keyphrase generation aims to automatically generate short phrases summarizing an input document. The recently emerged ONE2SET paradigm (Ye et al., 2021) generates keyphrases as a set and has achieved competitive performance. Nevertheless, we observe serious calibration errors outputted by ONE2SET, especially in the over-estimation of $\varnothing$ token (means "no corresponding keyphrase"). In this paper, we deeply analyze this limitation and identify two main reasons behind: 1) the parallel generation has to introduce excessive $\varnothing$ as padding tokens into training instances; and 2) the training mechanism assigning target to each slot is unstable and further aggravates the $\varnothing$ token over-estimation. To make the model well-calibrated, we propose WR-ONE2SET which extends ONE2SET with an adaptive instance-level cost Weighting strategy and a target Re-assignment mechanism. The former dynamically penalizes the over-estimated slots for different instances thus smoothing the uneven training distribution. The latter refines the original inappropriate assignment and reduces the supervisory signals of over-estimated slots. Experimental results on commonly-used datasets demonstrate the effectiveness and generality of our proposed paradigm.

cs.CL↗

arXiv2022

CKD-TransBTS: Clinical Knowledge-Driven Hybrid Transformer with Modality-Correlated Cross-Attention for Brain Tumor Segmentation

Brain tumor segmentation (BTS) in magnetic resonance image (MRI) is crucial for brain tumor diagnosis, cancer management and research purposes. With the great success of the ten-year BraTS challenges as well as the advances of CNN and Transformer algorithms, a lot of outstanding BTS models have been proposed to tackle the difficulties of BTS in different technical aspects. However, existing studies hardly consider how to fuse the multi-modality images in a reasonable manner. In this paper, we leverage the clinical knowledge of how radiologists diagnose brain tumors from multiple MRI modalities and propose a clinical knowledge-driven brain tumor segmentation model, called CKD-TransBTS. Instead of directly concatenating all the modalities, we re-organize the input modalities by separating them into two groups according to the imaging principle of MRI. A dual-branch hybrid encoder with the proposed modality-correlated cross-attention block (MCCA) is designed to extract the multi-modality image features. The proposed model inherits the strengths from both Transformer and CNN with the local feature representation ability for precise lesion boundaries and long-range feature extraction for 3D volumetric images. To bridge the gap between Transformer and CNN features, we propose a Trans&CNN Feature Calibration block (TCFC) in the decoder. We compare the proposed model with five CNN-based models and six transformer-based models on the BraTS 2021 challenge dataset. Extensive experiments demonstrate that the proposed model achieves state-of-the-art brain tumor segmentation performance compared with all the competitors.

eess.IV↗

arXiv2022

SOAR/Goodman Spectroscopic Assessment of Candidate Counterparts of the LIGO-Virgo Event GW190814

On 2019 August 14 at 21:10:39 UTC, the LIGO/Virgo Collaboration (LVC) detected a possible neutron star-black hole merger (NSBH), the first ever identified. An extensive search for an optical counterpart of this event, designated GW190814, was undertaken using the Dark Energy Camera (DECam) on the 4m Victor M. Blanco Telescope at the Cerro Tololo Inter-American Observatory. Target of Opportunity interrupts were issued on 8 separate nights to observe 11 candidates using the 4.1m Southern Astrophysical Research (SOAR) telescope's Goodman High Throughput Spectrograph in order to assess whether any of these transients was likely to be an optical counterpart of the possible NSBH merger. Here, we describe the process of observing with SOAR, the analysis of our spectra, our spectroscopic typing methodology, and our resultant conclusion that none of the candidates corresponded to the gravitational wave merger event but were all instead other transients. Finally, we describe the lessons learned from this effort. Application of these lessons will be critical for a successful community spectroscopic follow-up program for LVC observing run 4 (O4) and beyond.

astro-ph.HE↗

arXiv2022

WSSS4LUAD: Grand Challenge on Weakly-supervised Tissue Semantic Segmentation for Lung Adenocarcinoma

Lung cancer is the leading cause of cancer death worldwide, and adenocarcinoma (LUAD) is the most common subtype. Exploiting the potential value of the histopathology images can promote precision medicine in oncology. Tissue segmentation is the basic upstream task of histopathology image analysis. Existing deep learning models have achieved superior segmentation performance but require sufficient pixel-level annotations, which is time-consuming and expensive. To enrich the label resources of LUAD and to alleviate the annotation efforts, we organize this challenge WSSS4LUAD to call for the outstanding weakly-supervised semantic segmentation (WSSS) techniques for histopathology images of LUAD. Participants have to design the algorithm to segment tumor epithelial, tumor-associated stroma and normal tissue with only patch-level labels. This challenge includes 10,091 patch-level annotations (the training set) and over 130 million labeled pixels (the validation and test sets), from 87 WSIs (67 from GDPH, 20 from TCGA). All the labels were generated by a pathologist-in-the-loop pipeline with the help of AI models and checked by the label review board. Among 532 registrations, 28 teams submitted the results in the test phase with over 1,000 submissions. Finally, the first place team achieved mIoU of 0.8413 (tumor: 0.8389, stroma: 0.7931, normal: 0.8919). According to the technical reports of the top-tier teams, CAM is still the most popular approach in WSSS. Cutmix data augmentation has been widely adopted to generate more reliable samples. With the success of this challenge, we believe that WSSS approaches with patch-level annotations can be a complement to the traditional pixel annotations while reducing the annotation efforts. The entire dataset has been released to encourage more researches on computational pathology in LUAD and more novel WSSS techniques.

eess.IV↗

arXiv2021

Multi-Layer Pseudo-Supervision for Histopathology Tissue Semantic Segmentation using Patch-level Classification Labels

Tissue-level semantic segmentation is a vital step in computational pathology. Fully-supervised models have already achieved outstanding performance with dense pixel-level annotations. However, drawing such labels on the giga-pixel whole slide images is extremely expensive and time-consuming. In this paper, we use only patch-level classification labels to achieve tissue semantic segmentation on histopathology images, finally reducing the annotation efforts. We proposed a two-step model including a classification and a segmentation phases. In the classification phase, we proposed a CAM-based model to generate pseudo masks by patch-level labels. In the segmentation phase, we achieved tissue semantic segmentation by our proposed Multi-Layer Pseudo-Supervision. Several technical novelties have been proposed to reduce the information gap between pixel-level and patch-level annotations. As a part of this paper, we introduced a new weakly-supervised semantic segmentation (WSSS) dataset for lung adenocarcinoma (LUAD-HistoSeg). We conducted several experiments to evaluate our proposed model on two datasets. Our proposed model outperforms two state-of-the-art WSSS approaches. Note that we can achieve comparable quantitative and qualitative results with the fully-supervised model, with only around a 2\% gap for MIoU and FwIoU. By comparing with manual labeling, our model can greatly save the annotation time from hours to minutes. The source code is available at: \url{https://github.com/ChuHan89/WSSS-Tissue}.

eess.IV↗

arXiv2021

Towards User-Driven Neural Machine Translation

A good translation should not only translate the original content semantically, but also incarnate personal traits of the original text. For a real-world neural machine translation (NMT) system, these user traits (e.g., topic preference, stylistic characteristics and expression habits) can be preserved in user behavior (e.g., historical inputs). However, current NMT systems marginally consider the user behavior due to: 1) the difficulty of modeling user portraits in zero-shot scenarios, and 2) the lack of user-behavior annotated parallel dataset. To fill this gap, we introduce a novel framework called user-driven NMT. Specifically, a cache-based module and a user-driven contrastive learning method are proposed to offer NMT the ability to capture potential user traits from their historical inputs under a zero-shot learning fashion. Furthermore, we contribute the first Chinese-English parallel corpus annotated with user behavior called UDT-Corpus. Experimental results confirm that the proposed user-driven NMT can generate user-specific translations.

cs.CL↗

arXiv2020

Dynamic Context-guided Capsule Network for Multimodal Machine Translation

Multimodal machine translation (MMT), which mainly focuses on enhancing text-only translation with visual features, has attracted considerable attention from both computer vision and natural language processing communities. Most current MMT models resort to attention mechanism, global context modeling or multimodal joint representation learning to utilize visual features. However, the attention mechanism lacks sufficient semantic interactions between modalities while the other two provide fixed visual context, which is unsuitable for modeling the observed variability when generating translation. To address the above issues, in this paper, we propose a novel Dynamic Context-guided Capsule Network (DCCN) for MMT. Specifically, at each timestep of decoding, we first employ the conventional source-target attention to produce a timestep-specific source-side context vector. Next, DCCN takes this vector as input and uses it to guide the iterative extraction of related visual features via a context-guided dynamic routing mechanism. Particularly, we represent the input image with global and regional visual features, we introduce two parallel DCCNs to model multimodal context vectors with visual features at different granularities. Finally, we obtain two multimodal context vectors, which are fused and incorporated into the decoder for the prediction of the target word. Experimental results on the Multi30K dataset of English-to-German and English-to-French translation demonstrate the superiority of DCCN. Our code is available on https://github.com/DeepLearnXMU/MM-DCCN.

cs.CL↗

arXiv2020

New models for symbolic data analysis

Symbolic data analysis (SDA) is an emerging area of statistics concerned with understanding and modelling data that takes distributional form (i.e. symbols), such as random lists, intervals and histograms. It was developed under the premise that the statistical unit of interest is the symbol, and that inference is required at this level. Here we consider a different perspective, which opens a new research direction in the field of SDA. We assume that, as with a standard statistical analysis, inference is required at the level of individual-level data. However, the individual-level data are aggregated into symbols - group-based distributional-valued summaries - prior to the analysis. In this way, large and complex datasets can be reduced to a smaller number of distributional summaries, that may be analysed more efficiently than the original dataset. As such, we develop SDA techniques as a new approach for the analysis of big data. In particular we introduce a new general method for constructing likelihood functions for symbolic data based on a desired probability model for the underlying measurement-level data, while only observing the distributional summaries. This approach opens the door for new classes of symbol design and construction, in addition to developing SDA as a viable tool to enable and improve upon classical data analyses, particularly for very large and complex datasets. We illustrate this new direction for SDA research through several real and simulated data analyses.

stat.CO↗

arXiv2020

Dark Energy Survey Year 1 Results: Cosmological Constraints from Cluster Abundances and Weak Lensing

We perform a joint analysis of the counts and weak lensing signal of redMaPPer clusters selected from the Dark Energy Survey (DES) Year 1 dataset. Our analysis uses the same shear and source photometric redshifts estimates as were used in the DES combined probes analysis. Our analysis results in surprisingly low values for $S_8 =σ_8(Ω_{\rm m}/0.3)^{0.5}= 0.65\pm 0.04$, driven by a low matter density parameter, $Ω_{\rm m}=0.179^{+0.031}_{-0.038}$, with $σ_8-Ω_{\rm m}$ posteriors in $2.4σ$ tension with the DES Y1 3x2pt results, and in $5.6σ$ with the Planck CMB analysis. These results include the impact of post-unblinding changes to the analysis, which did not improve the level of consistency with other data sets compared to the results obtained at the unblinding. The fact that multiple cosmological probes (supernovae, baryon acoustic oscillations, cosmic shear, galaxy clustering and CMB anisotropies), and other galaxy cluster analyses all favor significantly higher matter densities suggests the presence of systematic errors in the data or an incomplete modeling of the relevant physics. Cross checks with X-ray and microwave data, as well as independent constraints on the observable--mass relation from SZ selected clusters, suggest that the discrepancy resides in our modeling of the weak lensing signal rather than the cluster abundance. Repeating our analysis using a higher richness threshold ($λ\ge 30$) significantly reduces the tension with other probes, and points to one or more richness-dependent effects not captured by our model.

astro-ph.CO↗

arXiv2018

DES Science Portal: Computing Photometric Redshifts

A significant challenge facing photometric surveys for cosmological purposes is the need to produce reliable redshift estimates. The estimation of photometric redshifts (photo-zs) has been consolidated as the standard strategy to bypass the high production costs and incompleteness of spectroscopic redshift samples. Training-based photo-z methods require the preparation of a high-quality list of spectroscopic redshifts, which needs to be constantly updated. The photo-z training, validation, and estimation must be performed in a consistent and reproducible way in order to accomplish the scientific requirements. To meet this purpose, we developed an integrated web-based data interface that not only provides the framework to carry out the above steps in a systematic way, enabling the ease testing and comparison of different algorithms, but also addresses the processing requirements by parallelizing the calculation in a transparent way for the user. This framework called the Science Portal (hereafter Portal) was developed in the context the Dark Energy Survey (DES) to facilitate scientific analysis. In this paper, we show how the Portal can provide a reliable environment to access vast data sets, provide validation algorithms and metrics, even in the case of multiple photo-zs methods. It is possible to maintain the provenance between the steps of a chain of workflows while ensuring reproducibility of the results. We illustrate how the Portal can be used to provide photo-z estimates using the DES first year (Y1A1) data. While the DES collaboration is still developing techniques to obtain more precise photo-zs, having a structured framework like the one presented here is critical for the systematic vetting of DES algorithmic improvements and the consistent production of photo-zs in the future DES releases.

astro-ph.GA↗

arXiv2018

Weak-lensing calibration of a stellar mass-based mass proxy for redMaPPer and Voronoi Tessellation clusters in SDSS Stripe 82

We present the first weak lensing calibration of $μ_{\star}$, a new galaxy cluster mass proxy corresponding to the total stellar mass of red and blue members, in two cluster samples selected from the SDSS Stripe 82 data: 230 redMaPPer clusters at redshift $0.1\leq z<0.33$ and 136 Voronoi Tessellation (VT) clusters at $0.1 \leq z < 0.6$. We use the CS82 shear catalog and stack the clusters in $μ_{\star}$ bins to measure a mass-observable power law relation. For redMaPPer clusters we obtain $M_0 = (1.77 \pm 0.36) \times 10^{14}h^{-1} M_{\odot}$, $α= 1.74 \pm 0.62$. For VT clusters, we find $M_0 = (4.31 \pm 0.89) \times 10^{14}h^{-1} M_{\odot}$, $α= 0.59 \pm 0.54$ and $M_0 = (3.67 \pm 0.56) \times 10^{14}h^{-1} M_{\odot}$, $α= 0.68 \pm 0.49$ for a low and a high redshift bin, respectively. Our results are consistent, internally and with the literature, indicating that our method can be applied to any cluster finding algorithm. In particular, we recommend that $μ_{\star}$ be used as the mass proxy for VT clusters. Catalogs including $μ_{\star}$ measurements will enable its use in studies of galaxy evolution in clusters and cluster cosmology.

astro-ph.CO↗

arXiv2017

Estimating global species richness using symbolic data meta-analysis

Global species richness is a key biodiversity metric. Despite recent efforts to estimate global species richness, the resulting estimates have been highly uncertain and often logically inconsistent. Estimates lower down either the taxonomic or geographic hierarchies are often larger than those above. Further, these estimates have been represented in a wide variety of forms, including intervals (a, b), point estimates with no uncertainty, and point estimates with either symmetrical or asymmetrical bounds, making it difficult to combine information across different estimates. Here, we develop a Bayesian hierarchical approach to estimate the global species richness from published studies. It allows us to recover interval estimates at each level of the hierarchy, even when data are partially or wholly unobserved, while respecting logical constraints, and to determine the effects of estimation on the whole hierarchy of obtaining future estimates anywhere within it

stat.AP↗

arXiv2017

Core or cusps: The central dark matter profile of a redshift one strong lensing cluster with a bright central image

We report on SPT-CLJ2011-5228, a giant system of arcs created by a cluster at $z=1.06$. The arc system is notable for the presence of a bright central image. The source is a Lyman Break galaxy at $z_s=2.39$ and the mass enclosed within the 14 arc second radius Einstein ring is $10^{14.2}$ solar masses. We perform a full light profile reconstruction of the lensed images to precisely infer the parameters of the mass distribution. The brightness of the central image demands that the central total density profile of the lens be shallow. By fitting the dark matter as a generalized Navarro-Frenk-White profile---with a free parameter for the inner density slope---we find that the break radius is $270^{+48}_{-76}$ kpc, and that the inner density falls with radius to the power $-0.38\pm0.04$ at 68 percent confidence. Such a shallow profile is in strong tension with our understanding of relaxed cold dark matter halos; dark matter only simulations predict the inner density should fall as $r^{-1}$. The tension can be alleviated if this cluster is in fact a merger; a two halo model can also reconstruct the data, with both clumps (density going as $r^{-0.8}$ and $r^{-1.0}$) much more consistent with predictions from dark matter only simulations. At the resolution of our Dark Energy Survey imaging, we are unable to choose between these two models, but we make predictions for forthcoming Hubble Space Telescope imaging that will decisively distinguish between them.

astro-ph.CO↗

arXiv2017

Models of the strongly lensed quasar DES J0408-5354

We present gravitational lens models of the multiply imaged quasar DES J0408-5354, recently discovered in the Dark Energy Survey (DES) footprint, with the aim of interpreting its remarkable quad-like configuration. We first model the DES single-epoch $grizY$ images as a superposition of a lens galaxy and four point-like objects, obtaining spectral energy distributions (SEDs) and relative positions for the objects. Three of the point sources (A,B,D) have SEDs compatible with the discovery quasar spectra, while the faintest point-like image (G2/C) shows significant reddening and a `grey' dimming of $\approx0.8$mag. In order to understand the lens configuration, we fit different models to the relative positions of A,B,D. Models with just a single deflector predict a fourth image at the location of G2/C but considerably brighter and bluer. The addition of a small satellite galaxy ($R_{\rm E}\approx0.2$") in the lens plane near the position of G2/C suppresses the flux of the fourth image and can explain both the reddening and grey dimming. All models predict a main deflector with Einstein radius between $1.7"$ and $2.0",$ velocity dispersion $267-280$km/s and enclosed mass $\approx 6\times10^{11}M_{\odot},$ even though higher resolution imaging data are needed to break residual degeneracies in model parameters. The longest time-delay (B-A) is estimated as $\approx 85$ (resp. $\approx125$) days by models with (resp. without) a perturber near G2/C. The configuration and predicted time-delays of J0408-5354 make it an excellent target for follow-up aimed at understanding the source quasar host galaxy and substructure in the lens, and measuring cosmological parameters. We also discuss some lessons learnt from J0408-5354 on lensed quasar finding strategies, due to its chromaticity and morphology.

astro-ph.GA↗

arXiv2016

VDES J2325-5229 a z=2.7 gravitationally lensed quasar discovered using morphology independent supervised machine learning

We present the discovery and preliminary characterization of a gravitationally lensed quasar with a source redshift $z_{s}=2.74$ and image separation of $2.9"$ lensed by a foreground $z_{l}=0.40$ elliptical galaxy. Since the images of gravitationally lensed quasars are the superposition of multiple point sources and a foreground lensing galaxy, we have developed a morphology independent multi-wavelength approach to the photometric selection of lensed quasar candidates based on Gaussian Mixture Models (GMM) supervised machine learning. Using this technique and $gi$ multicolour photometric observations from the Dark Energy Survey (DES), near IR $JK$ photometry from the VISTA Hemisphere Survey (VHS) and WISE mid IR photometry, we have identified a candidate system with two catalogue components with $i_{AB}=18.61$ and $i_{AB}=20.44$ comprised of an elliptical galaxy and two blue point sources. Spectroscopic follow-up with NTT and the use of an archival AAT spectrum show that the point sources can be identified as a lensed quasar with an emission line redshift of $z=2.739\pm0.003$ and a foreground early type galaxy with $z=0.400\pm0.002$. We model the system as a single isothermal ellipsoid and find the Einstein radius $θ_E \sim 1.47"$, enclosed mass $M_{enc} \sim 4 \times 10^{11}$M$_{\odot}$ and a time delay of $\sim$52 days. The relatively wide separation, month scale time delay duration and high redshift make this an ideal system for constraining the expansion rate beyond a redshift of 1.

astro-ph.GA↗

arXiv2015

Discovery of two gravitationally lensed quasars in the Dark Energy Survey

We present spectroscopic confirmation of two new lensed quasars via data obtained at the 6.5m Magellan/Baade Telescope. The lens candidates have been selected from the Dark Energy Survey (DES) and WISE based on their multi-band photometry and extended morphology in DES images. Images of DES J0115-5244 show two blue point sources at either side of a red galaxy. Our long-slit data confirm that both point sources are images of the same quasar at $z_{s}=1.64.$ The Einstein Radius estimated from the DES images is $0.51$". DES J2200+0110 is in the area of overlap between DES and the Sloan Digital Sky Survey (SDSS). Two blue components are visible in the DES and SDSS images. The SDSS fiber spectrum shows a quasar component at $z_{s}=2.38$ and absorption compatible with Mg II and Fe II at $z_{l}=0.799$, which we tentatively associate with the foreground lens galaxy. The long-slit Magellan spectra show that the blue components are resolved images of the same quasar. The Einstein Radius is $0.68$" corresponding to an enclosed mass of $1.6\times10^{11}\,M_{\odot}.$ Three other candidates were observed and rejected, two being low-redshift pairs of starburst galaxies, and one being a quasar behind a blue star. These first confirmation results provide an important empirical validation of the data-mining and model-based selection that is being applied to the entire DES dataset.

astro-ph.GA↗

arXiv2015

Mass Calibration of Galaxy Clusters at Redshift 0.1-1.0 using Weak Lensing in the Sloan Digital Sky Survey Stripe 82 Co-add

We present galaxy cluster mass-richness relations found in the Sloan Digital Sky Survey Stripe 82 co-add using clusters found using a Voronoi tessellation cluster finder. These relations were found using stacked weak lensing shear observed in a large sample of galaxy clusters. These mass-richness relations are presented for four redshift bins, $0.1 < z \leq 0.4$, $0.4 < z \leq 0.7$, $0.7 < z \leq 1.0$ and $0.1 < z \leq 1.0$. We describe the sample of galaxy clusters and explain how these clusters were found using a Voronoi tessellation cluster finder. We fit an NFW profile to the stacked weak lensing shear signal in redshift and richness bins in order to measure virial mass $(M_{200})$. We describe several effects that can bias weak lensing measurements, including photometric redshift bias, the effect of the central BCG, halo miscentering, photometric redshift uncertainty and foreground galaxy contamination. We present mass-richness relations using richness measure $N_{VT}$ with each of these effects considered separately as well as considered altogether. We also examine redshift evolution of the mass-richness relation. As a result we present measurements of the mass coefficient ($M_{200|20}$) and the power law slope ($α$) for power law fits to the mass and richness values in each of the redshift bins. We find values of the mass coefficient of $8.49 \pm 0.526$, $14.1 \pm 1.78$, $30.2 \pm 8.74$ and $9.23 \pm 0.525 \times 10^{13}$ $h^{-1}$ $M_{sun}$ for each of the four redshift bins respectively. We find values of the power law slope of $0.905 \pm 0.0585$, $0.948 \pm 0.100$, $1.33 \pm 0.260$ and $0.883 \pm 0.0500$ respectively.

astro-ph.CO↗