Parallel-in-Time Training of Recurrent Neural Networks for Dynamical Systems Reconstruction

Paper proposes parallel-in-time training for RNNs in dynamical systems reconstruction, leveraging recent algorithms that achieve O(log T) time complexity along sequence length T instead of classical O(T) backpropagation through time, enabling longer sequence modeling.