Research Interest

"Two things fill the mind with ever new and increasing admiration and awe, the more often and steadily we reflect upon them: the starry heavens above me and the moral law within me." - Immanuel Kant

Our research group is dedicated to advancing the field of astronomy by harnessing the power of modern deep learning techniques. Our primary focus is on expanding the scope of inferences and solving inverse problems to investigate a wide range of topics across all cosmic scales. From exploring the dynamical evolution of planetary systems to constructing precise emulations of 3D stellar atmospheres and their emergent spectra, our work encompasses a diverse array of subjects. We also study the billion-year evolution of our own galaxy, the Milky Way, by studying the billion of stars that we can now observe. Furthermore, we aim to understand the fate of the universe, including investigating whether the distribution of galaxies exhibits any parity violation. These complex and non-linear problems were previously difficult to address, but the advent of deep learning approaches has provided us with the tools to tackle these big questions and make strides on these topics.

Our research is grounded in the extensive amount of survey data available, and we are actively involved in analyzing data from various surveys across different domains. These include spectroscopy surveys such as SDSS-V, DESI, and 4MOST, astrometric data from Gaia, photometric surveys like Euclid, Roman, and CSST, and time-series data from LSST, TESS, and PLATO.

In addition to statistical inference and inverse problems, our group is at the forefront of developing Large Language Models specifically tailored for astronomy. As the principal investigator of the AstroLLaMA collaboration (formerly part of UniverseTBD), I work closely with the Oak Ridge National Laboratory and the Astrophysics Data System database to build ambitious large language models. By extensively pretraining and fine-tuning these models on the vast corpus of astronomical literature, we aim to gain insights into the mechanisms behind creativity and scientific breakthroughs. Moreover, we are working on a series of projects that will lead to end-to-end autonomous research, known as AI x Scientist, in astronomy. Our goal is to leverage AI to accelerate research in astronomy, enabling more efficient and effective exploration of the universe.


Galaxy Evolution


Stellar Astrophysics


Cosmology /Reionization


Planetary Systems


Large Language Models


Simulation-Based Inference



Research Projects

"One of the principal objects of theoretical research is to find the point of view from which the subject appears in the greatest simplicity." - Josiah Willard Gibbs

Reimagining How Science Can Be Done with AI

The emergence of large language models has ushered in a new era of scientific research. Our research group leads the AstroLLaMA collaboration, a vibrant consortium of astronomers and computer scientists united by a shared vision: to harness the immense potential of these models and propel astronomical research to new heights. In close partnership with the Oak Ridge National Laboratory, the Astrophysics Data Systems (ADS) and Microsoft, the primary database relied upon by astronomers worldwide, we specialize in pretraining large language models using an extensive corpus of astronomical literature. We further refine these models through meticulously curated tasks, drawing upon the wealth of domain knowledge within our field. Our research has demonstrated that such specialized large language models in astronomy lead to a more sophisticated and nuanced understanding of the astronomical literature, enabling the generation of novel scientific hypotheses. Our research pursues two primary objectives. Firstly, we strive to unravel the intricate processes of knowledge accumulation, shedding light on the complex dynamics that drive scientific progress. Secondly, we are dedicated to developing an end-to-end astronomy AI scientist capable of accelerating research in this captivating field.

Deciphering Our Galactic Backyard

One of the key areas of focus in our research group is unraveling the mysteries of our cosmic home, the Milky Way galaxy. We leverage cutting-edge data from the Gaia satellite, which probes the motion of a billion stars, and actively participate in major spectroscopic surveys such as SDSS-V and 4MOST. Currently, our research efforts revolve around two main themes. Firstly, we aim to untangle the intricate three-body dynamical dance between the Milky Way and its nearest neighbors, the Large Magellanic Cloud and the Small Magellanic Clouds, to shed light on how these cosmic companions have shaped our galaxy. Secondly, we study the ancient history of the Milky Way by studying the chemical evolution of its oldest components, seeking to understand the formation of the primordial Milky Way during the first few billion years of its existence. In addition to our Milky Way studies, we are also part of a team that will be exploring our neighboring galaxy, Andromeda, using the groundbreaking capabilities of the James Webb Space Telescope. We develop advanced spectroscopic inference tools to unravel the secrets of our sister galaxy, pushing the boundaries of our understanding of galaxy formation and evolution by using these nearby galaxies as cosmic laboratories.

Foundational Models for Spectra & Time Series

The rapid advancement of large language models has been largely propelled by the emergence of foundation models, which allow for efficient fine-tuning across a wide range of tasks. However, these foundational models are often limited to either language or image domains. Our research group is at the forefront of developing cutting-edge foundational models specifically tailored for spectra and time series data, which are ubiquitous in the field of astronomy. These highly versatile models are designed to be easily fine-tuned for a diverse array of tasks involving spectra and time-series analysis, ranging from inferring stellar properties to detecting anomalies and outliers. The training of our models benefits from our in-house state-of-the-art 3D simulations of stellar atmospheres, further refined using first-principles-based radiative transfer calculations for unrivaled accuracy. In addition to the models themselves, we are also developing a comprehensive suite of tools to facilitate the community in effortlessly fine-tuning these foundational models for their own research purposes. Our ultimate goal is to democratize the utilization of these powerful foundational models within the astronomy community, empowering astronomers to harness the full potential of deep learning in their own scientific endeavors.

Most Stars Lose Their Planets, But Why?

Our research, recently published in Nature, has unveiled that at least one in every twelve stars has consumed its own planets. This discovery was made possible through the C3PO program, spearheaded by our group. By meticulously studying twin stars using some of the world's most largest telescopes, including Magellan, VLT, and Keck, and analyzing the differences in their chemical compositions, we have found that planetary systems are frequently unstable, leading to the ejection of planets, with some ultimately being devoured by their host stars. This finding is not entirely unexpected, as a planetary system is essentially an N-body dynamical system, and it is well-known that three-body systems can exhibit instability (as recently popularized by a Netflix show). However, the precise timing and underlying reasons for these instabilities remain intriguing subjects of ongoing research. Understanding the intricacies of planetary dynamics provides valuable insights into the very formation of planetary systems, like our Solar system. Our current research efforts, in addition to further observational investigations of these systems through the C3PO program, also involve the development of cutting-edge deep learning models to aid us in better tackling the notorious three (or more) body problems.

Binary Ballet: Dancing with Newton and Einstein

The study of binary star systems has long been a pillar of astronomy. In recent years, the field has undergone a revolution, propelled by the wealth of astrometric data from the Gaia mission and the abundance of spectroscopic data. Our research group has been at the forefront of exploring the fascinating research avenues opened up by the study of binaries, which includes (a) leveraging binary systems to deepen our understanding of star formation and fundamental stellar astrophysics; (b) harnessing the power of binaries to hunt for captivating phenomena, such as the elusive stellar-mass black holes lurking in the Milky Way; and (c) putting Einstein's theory of gravity to the test using these cosmic laboratories. One notable output of our reserach is the use of a vast number of binary systems to calibrate the masses of stars across different evolutionary stages on the Herzsprung-Russell diagram. We have pioneered innovative techniques to detect binary systems using only single-epoch astrometric and spectroscopic data, enabling us to characterize their orbits and eccentricities with precision. Furthermore, we are pushing the boundaries of deep learning models to identify rare compact objects, such as stellar-mass black holes, while minimizing the false-positive detections.

Is the Universe the Same as Its Mirror Image?

The long-held assumption that the Universe is identical to its mirror image has recently been called into question. In other words, the Universe may not be statistically the same if we were to "flip" it. While parity violation itself is not entirely unexpected—after all, it has led to some of the most groundbreaking discoveries in physics, such as the revelation of the weak force—the notion of parity violation on a cosmic scale remains a topic of intense debate. The challenge lies in quantifying such detections using classical statistical methods and relying heavily on simulated Universes. Since the exact nature of the Universe's parity violation remains unknown, the current detections' heavy reliance on simulations has been a major shortcoming. Our research group has been innovating various methods to quantify these detections without the need for simulations. We have been exploring a wide range of deep learning techniques, including the scattering transform, which we pioneered in its application to cosmology, as well as graph neural networks and neural radiance fields, to quantify these detections. We are also applying them to the latest data from DESI to quantify parity violation, gain a deeper understanding of the Universe's current state and push the boundaries of our understanding of the Universe's fundamental symmetries.

AI generated music summarizing our research projects.


"The limits of my language mean the limits of my world." - Ludwig Wittgenstein

Research Group

"We are all in the gutter, but some of us are looking at the stars." - Oscar Wilde

Public Outreach

"The reward of the young scientist is the emotional thrill of being the first person in the history of the world to see something or to understand something." - Cecilia Payne-Gaposchkin

TED: How do we study the stars [0.8M views]

TED: How to measure distances [ 3.4M views ]

TEDx 2024: Seeing Humanity through Dystopian AI

TEDx Podcast: Not Your Typical Astronomer - Yuan-Sen's Optimization Mindset



Interactive Modules: Interstellar Absorption and the Lyman Alpha Forest (full screen)

Popular Press Columns

"As we look out into the Universe and identify the many accidents of physics and astronomy that have worked together to our benefit, it almost seems as if the Universe must in some sense have known that we were coming." - Freeman Dyson


"The struggle itself towards the heights is enough to fill a man's heart. One must imagine Sisyphus happy." - Albert Camus

Download Resume

Professional Appointment


The Ohio State University — Associate Professor


Australian National University — Associate Professor (tenured)


Australian National University — Assistant Professor (tenured)


Institute for Advanced Study, Princeton — NASA Hubble Fellow


Princeton University — Carnegie-Princeton Fellow

Visiting Appointment


Max Planck Institute for Astronomy


Tsinghua University - Institute for Advanced Study


Universiti Malaya - Department of Physics


Johns Hopkins University - Department of Physics



Harvard University — A.M., Ph.D., Astrophysics and Astronomy


National University of Singapore; B.Sc., M.Sc., Physics, minor in Mathematics


École Polytechnique; Engineer's Degree (equivalent to B.S.E. + M.S.E.), concurrent with the NUS degrees

Short Bio

Hello! I'm Yuan-Sen, an Associate Professor of astronomy and computer science at the Australian National University.

I earned my Ph.D. in Astrophysics and Astronomy from Harvard University. As a native of Malaysia, I've had the opportunity to live in seven countries and learn six languages along the way, making me a true global citizen. I completed my concurrent Bachelor's and Master's degrees at the National University of Singapore and École Polytechnique in France, supported by the Eiffel scholarship. During this time, I was honored to receive the Institute for Physics Medal and the National Academy of Science Award in Singapore, and I was also awarded a NASA Earth and Space Science Fellowship to pursue my studies at Harvard.

After completing my Ph.D., I was granted a NASA Hubble Fellowship, Carnegie-Princeton Fellowship, and the IAS Fellowship, which allowed me to conduct postdoctoral research at the Institute for Advanced Study at Princeton before joining the Australian National University as a tenured faculty member. Throughout my career, I've been fortunate to receive recognition such as the Humboldt Fellowship, the CCAPP Price Prize, and the ARC DECRA Fellowship.

My research focuses on advancing data modeling and statistical inference in astronomy through the lens of deep learning, unraveling the mysteries of the cosmos. I'm honored to co-chair the NASA Cosmic Programs Stars Science Interest Group and the IEEE "Deep Vision in Space" Task Force, and I've been actively involved in shaping future spectroscopic studies, such as the MUST and FOBOS surveys, as a science group leader.

One of the projects I'm particularly excited about is AstroLLaMA, where I am leading the development of the first specialized Large Language Models for astronomy in collaboration with Oak Ridge National Laboratory. This project is also supported by Microsoft.

As someone who grew up in Malaysia, where scientific education is still developing, I'm deeply passionate about promoting science in my home country. I've had the opportunity to organize the two major astronomy conferences, including an IAU Symposium, the first in the region since 1990s, and a summer school in Malaysia, and I regularly contribute columns to Sin Chew Jit Poh, a prominent Chinese newspaper with over a million readers. I've also given a TEDx talk in Malaysia and created two TED-Ed videos that have collectively garnered over four million views. I was also the first Malaysian astronomer to join the Malaysian Olympiad on Astronomy & Astrophysics Council, lecturing and selecting a team representing Malaysia.

Beyond academia, I've had the opportunity to apply my machine learning expertise to the art world, serving as a chief science officer on a project aimed at detecting art forgeries. And in my younger days, I even made a mark in the gaming community as a top Night-Elf player in Warcraft 3 in Malaysia!

NUS / École Polytechnique
Harvard (PhD)
Priceton / IAS / Carnegie

Services and Leadership Role



Alexander von Humboldt Research Award

Australian Research Council DECRA Fellowship

AURA Future Leader

NASA Hubble Fellowship

Carnegie-Princeton Fellowship

Institute for Advanced Study Fellowship

CCAPP Price Prize in Cosmology and AstroParticle Physics

NASA Earth and Space Science Fellowship

Selected to attend the Lindau Meeting of Nobel Laureates

Malaysian Perdana Scholar Award

National Academy of Science Award, Singapore

Institute for Physics Medal, Singapore

French Eiffel's Scholarship

NUS Jurong Book Prize

Australian Mathematics Competition, Gold Medal

Research Milestones


Refereed Publications


First/Supervising Author


Second/Third Author








"We cannot work without hoping that others will advance further than we have. In principle, this progress goes on ad infinitum. " - Max Weber

At the Australian National University, I teach courses in both computer science and astronomy. In computer science, I teach an advanced machine learning course called Statistical Machine Learning. It's all about the math behind machine learning techniques. Over in astronomy, I teach a course called Astronomical Computing, which covers the machine learning and stats methods that astronomers use all the time. Right now, I'm working on a textbook for the astronomy course, and I've put it online for anyone to check out. Just take a look below.

Contact Me

Have a project you'd like to discuss?