Explored the use of deep learning simulators to approximate gravitational dynamics.
- Point Prediction: Investigated the prediction of terminal positions based on initial data, potentially replacing iterative physics-based solvers. The dynamics were simulated with a graph neural network.
- Simulation: Built a model to generate full simulations of the system, predicting sequences of positions up to the terminal state.: This was done using an LSTM cell to take into consideration the time steps and the relation between each time step.
- Used torch.geometric for data preprocessing