2026-10-01 · America/Los_Angeles · 社区动态 · #12

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

Can training of nonlinear RNNs be efficiently parallelized, ensuring fast convergence even on very long time series from chaotic systems? In our #NeurIPS2026 spotlight “Parallel-in-Time Training of Recurrent Neural Networks for Dynamical Systems (DS) Reconstruction (DSR)” (preprint: https://arxiv.org/abs/2605.12683) we speed up training of nonlinear RNNs on time series from chaotic DS by more than 2 orders of magnitude (>100x) by combining DEER with generalized teacher forcing (GTF). DEER (https://openreview.net/forum?id=E34AlVLN0v) solves the RNN forward pass through Newton-type fixed point…

热度 47.4 / 100；排名与评分保留该期记录。