Definition
A neural network design that mixes convolutions with attention for sequence tasks.
A sequence model architecture combining self-attention with convolutional augmentation, originally for speech, recombined by agents in AIRA-Design's LRA experiments.