Mask More and Mask Later: Efficient Pre-training of Masked Language Models by Disentangling the [MASK] Token

Liao, Baohao (Corresponding author); Thulke, David (Corresponding author); Hewavitharana, Sanjika (Corresponding author); Ney, Hermann (Corresponding author); Monz, Christof (Corresponding author)

Abu Dhabi (2022)
Beitrag zu einem Tagungsband

In: Conference on Empirical Methods in Natural Language Processing

Einrichtungen

• Fachgruppe Informatik [120000]
• Lehrstuhl für Informatik 6 (Maschinelles Lernen) [122010]