arXiv subjects

Zechang Sun

Publications and source records attributed to Zechang Sun.

25 records · Page 2Linked to original sources

arXiv2024

AstroLLaMA-Chat: Scaling AstroLLaMA with Conversational and Diverse Datasets

We explore the potential of enhancing LLM performance in astronomy-focused question-answering through targeted, continual pre-training. By employing a compact 7B-parameter LLaMA-2 model and focusing exclusively on a curated set of astronomy corpora -- comprising abstracts, introductions, and conclusions -- we achieve notable improvements in specialized topic comprehension. While general LLMs like GPT-4 excel in broader question-answering scenarios due to superior reasoning capabilities, our findings suggest that continual pre-training with limited resources can still enhance model performance on specialized topics. Additionally, we present an extension of AstroLLaMA: the fine-tuning of the 7B LLaMA model on a domain-specific conversational dataset, culminating in the release of the chat-enabled AstroLLaMA for community use. Comprehensive quantitative benchmarking is currently in progress and will be detailed in an upcoming full paper. The model, AstroLLaMA-Chat, is now available at https://huggingface.co/universeTBD, providing the first open-source conversational AI tool tailored for the astronomy community.

astro-ph.IM↗

arXiv2023

DESI survey validation data in the COSMOS/HSC field: Cool gas trace main sequence star-forming galaxies at the cosmic noon

We present the first result in exploring the gaseous halo and galaxy correlation using the Dark Energy Spectroscopic Instrument (DESI) survey validation data in the Cosmic Evolution Survey (COSMOS) and Hyper Suprime-Cam (HSC) field. We obtain the multiphase gaseous halo properties in the circumgalactic medium (CGM) by using 115 quasar spectra (S/N > 3). We detect MgII absorption at redshift 0.6 < z < 2.5, CIV absorption at 1.6 < z < 3.6, and HI absorption associated with the MgII and CIV. By cross-matching the COSMOS2020 catalog, we identify the MgII and CIV host galaxies in ten quasar fields at 0.9 < z < 3.1. We find that within the impact parameter of 250 kpc, a tight correlation is seen between strong MgII equivalent width and the host galaxy star formation rate. The covering fraction fc of strong MgII selected galaxies, which is the ratio of absorbing galaxy in a certain galaxy population, shows significant evolution in the main-sequence galaxies and marginal evolution in all the galaxy populations within 250 kpc at 0.9 < z < 2.2. The fc increase in the main-sequence galaxies likely suggests the co-evolution of strong MgII absorbing gas and the main-sequence galaxies at the cosmic noon. Furthermore, several MgII and CIV absorbing gas is detected out of the galaxy virial radius, tentatively indicating the feedback produced by the star formation and/or the environmental effects.

astro-ph.GA↗

arXiv2023

Zephyr : Stitching Heterogeneous Training Data with Normalizing Flows for Photometric Redshift Inference

We present zephyr, a novel method that integrates cutting-edge normalizing flow techniques into a mixture density estimation framework, enabling the effective use of heterogeneous training data for photometric redshift inference. Compared to previous methods, zephyr demonstrates enhanced robustness for both point estimation and distribution reconstruction by leveraging normalizing flows for density estimation and incorporating careful uncertainty quantification. Moreover, zephyr offers unique interpretability by explicitly disentangling contributions from multi-source training data, which can facilitate future weak lensing analysis by providing an additional quality assessment. As probabilistic generative deep learning techniques gain increasing prominence in astronomy, zephyr should become an inspiration for handling heterogeneous training data while remaining interpretable and robustly accounting for observational uncertainties.

astro-ph.IM↗

arXiv2023

Quasar Factor Analysis -- An Unsupervised and Probabilistic Quasar Continuum Prediction Algorithm with Latent Factor Analysis

Since their first discovery, quasars have been essential probes of the distant Universe. However, due to our limited knowledge of its nature, predicting the intrinsic quasar continua has bottlenecked their usage. Existing methods of quasar continuum recovery often rely on a limited number of high-quality quasar spectra, which might not capture the full diversity of the quasar population. In this study, we propose an unsupervised probabilistic model, Quasar Factor Analysis (QFA), which combines factor analysis (FA) with physical priors of the intergalactic medium (IGM) to overcome these limitations. QFA captures the posterior distribution of quasar continua through generatively modeling quasar spectra. We demonstrate that QFA can achieve the state-of-the-art performance, $\sim 2\%$ relative error, for continuum prediction in the Ly$α$ forest region compared to previous methods. We further fit 90,678 $2<\mathrm{z}<3.5$, SNR$>2$ quasar spectra from Sloan Digital Sky Survey Data Release 16 and found that for $\sim 30\%$ quasar spectra where the continua were ill-determined with previous methods, QFA yields visually more plausible continua. QFA also attains $\lesssim 1\%$ error in the 1D Ly$α$ power spectrum measurements at $\mathrm{z}\sim 3$ and $\sim 4\%$ in $\mathrm{z}\sim 2.4$. In addition, QFA determines latent factors representing more physically motivated than PCA. We investigate the evolution of the latent factors and report no significant redshift or luminosity dependency except for the Baldwin effect. The generative nature of QFA also enables outlier detection robustly; we showed that QFA is effective in selecting outlying quasar spectra, including damped Ly$α$ systems and potential Type II quasar spectra.

astro-ph.CO↗

arXiv2023

The Mass-Metallicity Relation of Dwarf Galaxies at Cosmic Noon from JWST Observations

We present a study of the mass-metallicity relation (MZR) of 51 dwarf galaxies ($M_\star\approx 10^{6.5} - 10^{9.5}~M_\odot$) at $z = 2-3$ from the Abell 2744 and SMACS J0723-3732 galaxy cluster fields. These dwarf galaxies are identified and confirmed by deep JWST/NIRISS imaging and slitless grism spectroscopic observations. By taking advantage of the superior performance of JWST and the gravitational lensing effect, we extend the previous MZR relation at $z=2-3$ to a much lower mass regime down by $\approx$ 2.5 orders of magnitude as compared with previous studies. We find that the MZR has a shallower slope at the low-mass end ($M_\star<10^{9}~M_\odot$), with a slope turnover point of $\approx$ $10^9~M_\odot$. This implies that the dominating feedback processes in dwarf galaxies may be different from that in massive galaxies. From $z=3$ to $z=2$, the metallicity of the dwarf galaxies is enhanced by $\approx0.09$ dex for a given stellar mass, consistent with the mild evolution found in galaxies with higher mass. Furthermore, we confirm the existence of a fundamental metallicity relation (FMR) between the gas-phase metallicity, stellar mass, and star formation rate in dwarf galaxies at $z=2-3$. Our derived FMR, which has no significant redshift evolution, can be used as a benchmark to understand the origin of the anti-correlation between the SFR and metallicity of dwarf galaxies in the high-$z$ Universe.

astro-ph.GA↗

arXiv2022

An Unsupervised Learning Approach for Quasar Continuum Prediction

Modeling quasar spectra is a fundamental task in astrophysics as quasars are the tell-tale sign of cosmic evolution. We introduce a novel unsupervised learning algorithm, Quasar Factor Analysis (QFA), for recovering the intrinsic quasar continua from noisy quasar spectra. QFA assumes that the Ly$α$ forest can be approximated as a Gaussian process, and the continuum can be well described as a latent factor model. We show that QFA can learn, through unsupervised learning and directly from the quasar spectra, the quasar continua and Ly$α$ forest simultaneously. Compared to previous methods, QFA achieves state-of-the-art performance for quasar continuum prediction robustly but without the need for predefined training continua. In addition, the generative and probabilistic nature of QFA paves the way to understanding the evolution of black holes as well as performing out-of-distribution detection and other Bayesian downstream inferences.

astro-ph.CO↗

arXiv2022

Deep Learning of DESI Mock Spectra to Find Damped Lyα Systems

We have updated and applied a convolutional neural network (CNN) machine learning model to discover and characterize damped Ly$α$ systems (DLAs) based on Dark Energy Spectroscopic Instrument (DESI) mock spectra. We have optimized the training process and constructed a CNN model that yields a DLA classification accuracy above 99$\%$ for spectra which have signal-to-noise (S/N) above 5 per pixel. Classification accuracy is the rate of correct classifications. This accuracy remains above 97$\%$ for lower signal-to-noise (S/N) $\approx1$ spectra. This CNN model provides estimations for redshift and HI column density with standard deviations of 0.002 and 0.17 dex for spectra with S/N above 3 per pixel. Also, this DLA finder is able to identify overlapping DLAs and sub-DLAs. Further, the impact of different DLA catalogs on the measurement of Baryon Acoustic Oscillation (BAO) is investigated. The cosmological fitting parameter result for BAO has less than $0.61\%$ difference compared to analysis of the mock results with perfect knowledge of DLAs. This difference is lower than the statistical error for the first year estimated from the mock spectra: above $1.7\%$. We also compared the performance of CNN and Gaussian Process (GP) model. Our improved CNN model has moderately 14$\%$ higher purity and 7$\%$ higher completeness than an older version of GP code, for S/N $>$ 3. Both codes provide good DLA redshift estimates, but the GP produces a better column density estimate by $24\%$ less standard deviation. A credible DLA catalog for DESI main survey can be provided by combining these two algorithms.

astro-ph.GA↗