Aug 31, 2018 · 1 min read
Nice concept.
Reading your article, I was thinking that your dataset would be well suited to recurrent neural networks where the success criteria would be how close (or how far ahead) the RNN is capable of predicting the next positions of the planets knowing past positions.
Also, it looks like neural network architectures are increasingly specialized, so an architecture well suited to recognizing or predicting planet orbits might not be that great at recognizing cats or whatever the latest trend in AI application is. Do you already have in mind an application where you think NN architectures optimized for planet orbit recognition would perform well ?