用 LetX 一起写 LaTeX。

LetX 是为学生和研究团队打造的免费在线协作 LaTeX 编辑器。几秒钟即可编译，实时协同写作，未发表的内容始终保持私密。

快速编译 1000+ 个模板协作者人数不限无需安装

免费开始写作探索模板

所属计划

公式

实时预览

e^iπ + 1 = 0

∫₀^∞e^−x² dx =√π2

main.tex

\documentclass{article}
\usepackage{amsmath}
\begin{document}
\begin{align*}
  e^{i\pi} + 1 &= 0 \\[1ex]
  \int_{0}^{\infty} e^{-x^2} dx &= \frac{\sqrt{\pi}}{2}
\end{align*}
\end{document}

真实编辑过程

邀请团队成员，实时共同编辑，几秒内编译生成 PDF。这就是 LetX 的实际使用体验。

letx.app/editor/exoplanet-thesis

Exoplanet Detection Thesis刚刚编辑

main.tex

1% Exoplanet & Star Detection, main.tex
2\documentclass[11pt]{article}
3\usepackage{graphicx, amsmath, wrapfig}
4
5\title{Synergistic ML \& CV for Exoplanet Detection}
6\author{Shihab Shahriar Antor \and Ashraful Kabir Alif}
7
8\begin{document}
9\maketitle
10
11\begin{abstract}
12  We detect faint exoplanets in JWST/MIRI imagery.
13\end{abstract}
14
15\section{Introduction}Alif
16Direct imaging is limited by \emph{stellar glare}.Shihab
17\begin{wrapfigure}{r}{0.45\linewidth}
18  \includegraphics[width=\linewidth]{jwst-miri}
19  \caption{JWST/MIRI direct imaging.}
20\end{wrapfigure}
21
22\section{Method}
23We train a contrastive objective:
24\begin{equation}
25  \mathcal{L} = -\log \frac{e^{s^+}}{\sum e^{s_i}}
26\end{equation}
27\section{Results}
28\end{document}

无问题行 16，列 5465 个字符LaTeX

pdfLaTeX1 / 4100%

Synergistic Application of Machine Learning and Computer Vision for the Detection of Exoplanets and Stars

Shihab Shahriar Antor Ashraful Kabir Alif

Department of Computer Science, BRAC University

Abstract. We present a framework that detects faint exoplanets in JWST/MIRI coronagraph imagery using contrastive and few-shot learning, separating planetary signals from bright host stars under low signal-to-noise. On held-out fields it outperforms CNN and GNN baselines while using far fewer labels.

Keywords: exoplanets, direct imaging, contrastive learning, few-shot learning, JWST

1. Introduction

Figure 1: JWST/MIRI direct imaging of a faint exoplanet (b) beside its host star.

Direct imaging of exoplanets is fundamentally limited by the overwhelming glare of the host star. Modern coronagraphs on JWST/MIRI suppress this stellar light, yet faint planetary signals remain buried in structured noise. As shown in Figure 1, the planet sits within the diffraction pattern of its star, making detection a needle-in-a-haystack problem. We cast this as a representation-learning task and show that contrastive objectives recover signals that pixel-wise classifiers miss.

2. Method

We learn a discriminative embedding with a contrastive objective, pulling together augmented views of one source and pushing apart star and planet candidates:

𝓛= −loge^s⁺Σ e^s_i(1)

A few-shot head then adapts to new instruments from a handful of labelled detections, which is critical given the scarcity of confirmed direct-imaging targets.

3. Results

On held-out fields the model recovers 92% of injected planets at a false-positive rate below 0.1 per frame, outperforming CNN and GNN baselines by 14 and 9 points of recall respectively.

Preprint, drafted collaboratively on LetXJune 2026

Exoplanet Thesis

main.tex

Synergistic Application of Machine Learning and Computer Vision for the Detection of Exoplanets and Stars

Shihab Shahriar Antor Ashraful Kabir Alif

Department of Computer Science, BRAC University

Keywords: exoplanets, direct imaging, contrastive learning, few-shot learning, JWST

1. Introduction

Figure 1: JWST/MIRI direct imaging of a faint exoplanet (b) beside its host star.

2. Method

We learn a discriminative embedding with a contrastive objective, pulling together augmented views of one source and pushing apart star and planet candidates:

𝓛= −loge^s⁺Σ e^s_i(1)

A few-shot head then adapts to new instruments from a handful of labelled detections, which is critical given the scarcity of confirmed direct-imaging targets.

3. Results

On held-out fields the model recovers 92% of injected planets at a false-positive rate below 0.1 per frame, outperforming CNN and GNN baselines by 14 and 9 points of recall respectively.

Preprint, drafted collaboratively on LetXJune 2026

无问题行 16，列 5465 个字符LaTeX

文件

代码

PDF

已连接

编辑器中实时同步

邀请协作编译

完成，图也显示出来了。随时可以重新编译。

全球高校的学生和教师都在使用

BRAC 大学、慕尼黑工业大学、汉堡大学、巴塞罗那加泰罗尼亚理工大学、渥太华大学，以及 75+ 个国家的 730+ 所高校与实验室，都有学生和研究人员在用 LetX 写作。

BRAC UniversityBangladesh Universitas Negeri SurabayaIndonesia UNAMMexico UC San DiegoUnited States HCMC Univ. of TechnologyVietnam Univ. Nacional de San AgustínPeru Portland State UniversityUnited States University of WindsorCanada University of PretoriaSouth Africa De La Salle UniversityPhilippines Universitas PadjadjaranIndonesia BITS PilaniIndia Idaho State UniversityUnited States Universidad de Buenos AiresArgentina Universidad Nacional de ColombiaColombia Universidad Nacional de IngenieríaPeru Univ. Nacional Mayor de San MarcosPeru Carleton UniversityCanada Hogeschool RotterdamNetherlands RajagiriIndia PUCIT, University of the PunjabPakistan Universidad de GuadalajaraMexico Universidad Diego PortalesChile San José State UniversityUnited States Universidade de São PauloBrazil Universidad de ConcepciónChile Technical University of IașiRomania Islamic University of TechnologyBangladesh CUETBangladesh University of AlbertaCanada University College CorkIreland Univ. Federal de Juiz de ForaBrazil FAST-NUCESPakistan GITAM UniversityIndia PUC ValparaísoChile Instituto Superior TécnicoPortugal Universidad de La LagunaSpain

实时协作

实时协作写作。编辑会无冲突地合并，不必再通过邮件来回发送 .tex 文件或覆盖他人的修改，版本历史还能让你回到之前的草稿。

无冲突编辑（Yjs CRDT）
协作者不限：免费
版本历史

2 人在线

Direct Imaging of Exoplanets

Coronagraphs on JWST/MIRI suppress the host star’s glare, yet faint planetary signals stay buried in structured stellar noise.

We recover them with contrastive and few-shot learning, separating planet candidates from the star even at very low signal-to-noise.

已保存 · 版本历史

能编辑文档的助手

用日常语言提出需求。Lexi 会阅读项目、提出修改、通过编译检查能否成功构建，并在当前编辑的位置应用修改。

这就是编辑器

你的文件，你的源码，你的 PDF。

画面里有文件树、LaTeX 源码、实时 PDF，还有和你一起编辑文档的其他人。Lexi 在编辑器底部待命，等你开口提需求。

read_filemain.tex

make_table3 列，2 行

propose_editmain.tex

compile无溢出盒子

此外list_filessearch_textformatlintbib_hygieneask_user

LResidual baselines

SRAK

反馈问题导出分享S

文件大纲

ROOT

main.tex

refs.bib

figures

sections

main.tex

Lexi历史

1
2
3
4
5
6
7
8
9
10
11
12
13
14
\documentclass[11pt]{article}
\usepackage{graphicx}
\begin{document}
\section{Results}
Our model reaches 94.2\% accuracy on the
held-out set; the baseline reaches 89.7\%.
F1 was 0.94 against 0.88.
Training converged after 40 epochs.
\end{document}