node

A TensorFlow (V2) implementation of nerual ordinary differential equations.

TODO

• stable MLP NODE for MNIST dataset.
• Stable CNN NODE for MNIST dataset.
• Generic function that computes the trajectory.
• Visualization utils (How to visualize high dimensional trajectory).
• Generalizing the input type from tf.Tensor to Nest[tf.Tensor].
• Study the fix point for the trained stable NODE.
• Adaptive numeric precision in solver.
• CNF.
• Convert to TF code style.
• Add adaptive solvers.
• Re-write ALBERT with NODE.
• Time series forecasting via ALBERT.
• Precise definitions for phase point, phase vector field, e.t.c.
• Adaptive time interval.
• Dynamical convergence in Hopfield.