AstroAI with Perforated AI

Inspiration

NASA's Kepler and TESS missions have discovered thousands of exoplanets using transit photometry. We wanted to combine this exciting field with Perforated AI's cutting-edge dendritic optimization to improve detection accuracy.

What it does

AstroAI generates synthetic exoplanet transit light curves and uses neural networks to classify whether a transit signal is present. The project integrates Perforated AI's dendritic optimization to enhance model performance compared to baseline PyTorch models.

How we built it

• Python with PyTorch for neural network models (MLP and CNN architectures)
• Astropy for realistic light curve simulation with noise modeling
• Perforated AI library for dendritic optimization integration
• Streamlit for interactive demo interface

Challenges we ran into

• Modeling realistic noise patterns in synthetic light curves
• Integrating Perforated AI's API with custom model architectures
• Balancing model complexity with training efficiency on CPU

Accomplishments that we're proud of

• Achieved 87% baseline accuracy on transit detection
• Successfully integrated Perforated AI dendritic optimization
• Built complete training pipeline with comparison metrics

What we learned

• How dendritic optimization can improve neural network performance
• Time-series analysis techniques for astronomical data
• The importance of proper validation in scientific ML applications

What's next for AstroAI

• Integration with real NASA Kepler/TESS data
• Multi-planet detection capabilities
• Deployment as a web service for citizen science

Built With

• machine-learning
• neural-networks
• numpy
• perforatedai
• python
• pytorch
• streamlit