GalaxAI

The universe is calling to you. Since the dawn of humanity, eyes have stared up at the stars—united by wonder. What do the skies hide? Today, everything changes. With GalaxAI, the answer awaits.

Comment

Inspiration

Hope Roundcount (they/them) is an assistant researcher in astrophysics with a deep love for the universe. They are passionate about making the cosmos more accessible, intuitive, and understandable for both seasoned professionals and curious amateur stargazers. This passion is what led them to take on a space-focused project this year, where they are grappling with the complex and age-old question of distinguishing active galactic nuclei from quiescent galaxies. This issue is particularly relevant to their mentors at UMass, to the astrophysics community at large, and to Hope's own research.

Alan Khoultchaev (he/him), a computer science major, joined the team because of his love for solving problems and his drive to contribute to something meaningful. As the programmer behind the user interface of the project, Alan valiantly faced a relentless battle with bugs (which made Hope glad to have a programming master on the team), treating each one like an enemy in a video game where he's on his last life—always determined to make the system work smoothly!

What it does

GalaxAI is hosted on a website with a domain registered through GoDaddy, utilizing GitHub as the original repository and deploying the application through CloudFlare for seamless access and performance. The homepage is designed with user-friendly navigation, featuring four primary sections that lead to essential resources for astronomers and data scientists alike.

In the top left, an image directs users to a detailed tutorial that explains the core concepts of spectroscopy, how to properly analyze spectroscopic data, and how to access and utilize a comprehensive database of astronomical data. To facilitate the process, a suite of Python functions is provided for plotting, debugging, and summarizing the data, offering a more intuitive and accessible approach to navigating the complexities of astronomical observations. This empowers astronomers, both seasoned professionals and enthusiastic amateurs, to explore the mysteries of the skies with greater precision and ease.

In the top right corner, users can click the image to access the FITS datasets download page, where they can obtain the spectroscopic continuums used to train GalaxAI’s machine learning models. These datasets, sourced from reputable databases, are all open-source, and each dataset is accompanied by proper citations, ensuring transparency and academic integrity. The page allows users to download the data for their own research and exploration, providing a solid foundation for anyone looking to further investigate the spectra of active galactic nuclei (AGNs) and quiescent galaxies.

The bottom left button takes users to the heart of GalaxAI: the Machine Learning analyzer page. This is where the main challenge of the project is tackled: distinguishing between the spectral data of AGNs—supermassive black holes—and quiescent galaxies, whose spectra often appear remarkably similar. This overlap has been a significant issue in astrophysics, as highlighted by a paper from UMass led by Prof. Kate Whitaker, which proposed the discovery of "dead" galaxies that could disrupt our understanding of the universe’s age and the process of galaxy evolution. GalaxAI uses machine learning, specifically scikit-learn, to analyze spectroscopic data and provide a statistically informed prediction of whether an object is more likely to be an AGN or a quiescent galaxy. The model has been trained on a diverse set of spectra, enabling it to differentiate between the two with a high degree of accuracy. This approach allows for an additional layer of verification before crucial papers are published, ensuring that astronomers have a reliable tool to support their conclusions. With GalaxAI, the stakes of misidentifying these objects—such as the potential for undermining foundational scientific theories—are mitigated, providing more precise data analysis at a lower cost.

Finally, the bottom right of the homepage links to a bibliography, where all the datasets, images, and resources used in the development of GalaxAI are properly credited, ensuring that all contributors and sources are acknowledged and respected.

(As a minor addition, you can also click on "About" in the bottom right corner to get a simple summary of the project and to get in contact with us.)

How we built it

Initially, we utilized Replit to share files, but limitations on collaboration prompted us to seek an alternative solution. To address this, Alan Khoultchaev created a GitHub repository and added Hope as a collaborator, where all of our HTML, Python, JavaScript, CSS files, and other project assets were securely stored. We registered the domain name http://galaxai.wiki through a third-party service and linked it to the repository’s files. The majority of the user interface was developed using HTML and CSS, while basic web functionalities were implemented with JavaScript. For the data analysis component, we leveraged Python, utilizing scikit-learn for machine learning to streamline the AI development process and avoid redundant work.

Challenges we ran into

The biggest challenge we ran into was finding a way to make the tool publicly accessible. First, we planned to use Replit, but then we found that there was an issue of collaborator limits. We moved it to Github Pages, and then used Flask to incorporate the Python functionality. However, Github only supports static pages, so Flask is not supported. We then tried Render by importing the Github repository, but Render did not have correctly updated Python, which made it impossible for us to use scikit-learn, a necessity for our project. Finally, we moved to Cloudflare, where we were finally able to deploy our project but still faced issues of errors, directory mismanagement, and more. Ironically, the ML was a lot easier to set up in this project than the simple website. However, we adapted and worked hard to ensure this project would be a success, and we are so excited to bring GalaxAI to you now!

Accomplishments that we're proud of

We are both proud to have successfully created an AI tool, as we both previously have absolutely no experience with AI at all. Hope in particular is proud of the way this tool will be used, as they are confident when they attend their meeting with their research mentor this week that she will be blown away by this project. Alan, on the other hand, did a stellar (get it?) job of setting up the entire user interface of the project!

Overall, this was a very difficult endeavor. Additionally, with only two people on a team trying to learn so many new skills, there were times when we discussed whether it would have been better to have a larger team so that not every new skill had to fall on our shoulders. However, the things we learned from this hackathon were invaluable, and in the end, the lessons will be something we continue to use in work and life from here on out.

What we learned

That Github won't work with Flask, for one. (I know I already mentioned this, but you must understand this one took us probably four hours to finally figure out, so I'm mentioning it again.)

But also, we learned that we are capable of more than we thought. To be honest, we did not think we had the skills for this. Coming into this project, Alan claimed to not know much about UI design, and Hope was...intimidated by the thought of building an AI, to say the least. Even more than that, Hope had never calculated their own redshifts, Alan had never worked with such intricate layers of different programming languages all converging into HTML, and neither of them have ever completed a project with this level of sophistication and practicality in the real world.

(Seriously. Hope is bringing this to their professor on Tuesday and very excited to hear surprised yet impressed feedback. You have no idea how much of a pain this sort of thing has been to try to solve in astro research up until now.)

What's next for Black Hole Project

We would love to continue feeding the machine more data in order to provide an even more accurate algorithm! In addition to this, we are exploring the possibility of expanding the scope of the project to include the identification of other celestial objects. Potential areas of extension include distinguishing between spiral and elliptical galaxies, identifying exoplanets, classifying quasars, detecting supernovae, and more. These advancements would significantly broaden the project's applicability and contribute to a deeper understanding of the universe.

Share this project:

Updates