#RNN
2 related articles
·3 min
Information-Theoretic Audit: Dissociating Transient State from Durable Parameter Memory in Recurrent Architectures
An information-theoretic audit of recurrent neural networks dissociates transient information in hidden states from durable parameter memory consolidated after training, exploring implications for interpretability, continual learning, and architecture design.
Read more →
·2 min
100x Faster RNN Training: Parallel-in-Time Methods Tackle Chaotic System Reconstruction
NeurIPS 2026 Spotlight paper proposes a parallel-in-time training method combining DEER and Generalized Teacher Forcing (GTF) to speed up RNN training on chaotic dynamical systems by over 100x, outperforming Mamba and other state space models.
Read more →