What exactly does it do?

Our distributed system is designed to assist the astronomy community in the detection and classification of exoplanets using Semi-Supervised Machine Learning. Users (eventually, astronomers working in space observatories) are required to provide the relevant data pertaining to the planet they wish to classify. Subsequently, our application will categorize their planet into one of the following classes: confirmed, candidate, or false positive.

You can find here a basic and static version prototype of what the web app can do : http://www.exowebai.wiki (hosted thanks to GoDaddy free domain and GitHub Pages).

How does it work?

Our system encompasses four primary functions:

Classification: We receive unlabeled data collected from telescope sensors, pre-process it, and apply our machine learning model to make predictions.
Communication: The pseudo-labeled data just classified is sent to the central server and stored in the network's public database before being broadcasted later to the whole network.
Retraining: Initially, the model is trained on the data from the NASA Exoplanet Archive, and subsequently, it undergoes further training using data from the network's public database. (self-supervised learning approach)
Visualization: using a lightweight web application, users could easily check the network's public dataset.

What benefits does it have?

Continuously retraining our machine learning model using the diverse data collected from the observatories enhances the model's generalization capacity, leading to improved accuracy in predictions over time. The model becomes more effective as it is exposed to new and varied data. Furthermore, we prioritize the selection of data with a high probability of being well-classified, typically at a probability threshold of 0.75 or higher that could be configured later by the user. This approach ensures that our model is fed only high-quality data, further enhancing its performance.

What do you hope to achieve?

The search for possible life beyond Earth has long been a primary focus for many space agencies. Considerable efforts, time, and financial resources have been dedicated to the detection and study of exoplanets, with the hope of identifying any signs of life outside our solar system. These ongoing efforts, while commendable, could be even more effective if they were coordinated and combined. By joining forces, sharing experiences, and pooling advancements, space agencies can optimize their resources and shift their focus toward more advanced technologies for space exploration.

This is precisely the objective of our project:

We aim to connect scientists worldwide and create a platform where they can benefit from one another's progress and contribute their own findings, maximize the utilization of available resources and prevent any redundancy or duplication of efforts, ultimately fostering real progress in the field of exoplanet research.
Furthermore, our framework aspires to empower scientists to fully exploit the available data. The exchange of data and findings will enable scientists to comprehend the collective advancements in this field, allowing them to build upon existing achievements instead of starting from scratch. This approach prevents the wasteful expenditure of funds and resources.
Moreover, we intend to simplify and reduce the cost of exoplanet research. Most often, space agencies allocate substantial funds to construct and launch telescopes into space, as well as to maintain them. For example, NASA's expenditure of approximately $87 million on the TESS (Transiting Exoplanet Survey Satellite) mission is a significant investment. This cost can be a barrier to entry for other scientists with significant potential, hindering the much needed progress in this field.

Hence, we have established a collaborative platform where data resources are accessible to all, leveling the playing field and maximizing opportunities for exoplanets survey.

What tools, coding languages, hardware, or software did you use to develop your project?

Back-end languages and libraries:

Python 3, Jupyter Notebook, Flask server, Joblib

AI & Data Science libraries:

Numpy, Pandas, Matplotlib, Seaborn, Scikit-learn

Front-end:

HTML, CSS, JavaScript

Data Source:

NASA Exoplanet Archive

Collaborative coding:

Github, Git

Video demo:

Canva

Our Hackaton experience:

Our NASA Hackathon experience was wonderful and unique in its own way. It was a blend of various emotions, including excitement, enthusiasm, stress, and amusement. This experience taught us how to work effectively as a team, communicate with one another, and clearly convey our ideas to ensure mutual understanding.

Furthermore, it instilled in us the values of commitment and taking responsibility, encouraging us to persevere through challenges rather than giving up at the first hurdle. It would be misleading to claim that it was an easy journey, but it was undeniably a fruitful and enjoyable one.

We learned a lot on this journey, whether it was from the workshops held throughout the week by the local event organizers or our own research in the domain of space, particularly exoplanets. This experience was definitely eye-opening regarding the challenges that the astronomical community faces. It also served as a motivator for us to continue our work and find innovative solutions to address these challenges.

2023 NASA Space Apps Challenge

SpaceArc