Generative AI Models for Planetary System Formation
Delve into the fascinating world of generative AI models for planetary formation and exoplanet detection. We leverage the power of artificial intelligence to unravel the mysteries of planetary systems and search for Earth twins and potential life in the Universe.
Transformer models
Discover how advances in Large Language Model studies and the Transformer architecture can be used to create a planetary system generative model with excellent performances.
RNN models
Recursive Neural Networks are specially suited to consider samples of variable dimension like sequences. Check-out our generative model, which uses a RNN architecture to predict the probability of planet N in a sequence, knowing the properties of planets 1 to N-1, as a mixture of Gaussians.
GAN models
By training a Generator and a Discriminator to compete one against the other, GAN models can produce samples convincingly similar to training samples, but can suffer from mode collapse. Discover how we apply these techniques to numerical datasets.
Auto-Encoders
Demonising Auto-Encoders and Variational Auto-Encoders are trained to remove noise from corrupted samples, combining an encoder and a decoder. By sampling in their latent space, one can turn the decoder in a generative model. Find out how we leverage these techniques in some un-published experiments.
Ready to explore the possibilities with our generative AI models?
Contact us now to learn more about how we can help you in understanding planetary formation and astronomical data.