Pit One Against Many: Leveraging Attention-head Embeddings for Parameter-efficient Multi-head Attention.

Huiyin Xue, Nikolaos Aletras

Browse the full EMNLP paper archive.