1 paper · 1 filter
Korbinian Pöppel, Maximilian Beck, Sepp Hochreiter
While Transformers and other sequence-parallelizable neural network architectures seem like the current state of the art in sequence modeling, they specifically lack state-tracking…