#sequence-modeling

Selective-Update RNNs Match Transformers While Using Less Memory

A new RNN architecture learns when to update internal state, preserving memory across long sequences and reducing computational waste on redundant input.