Unsupervised knowledge distillation from a pretrained language model to itself, by alternating between its bi- and cross-encoder forms.
Source: Trans-Encoder: Unsupervised sentence-pair modelling through self- and mutual-distillationsPaper | Code | Results | Date | Stars |
---|
Task | Papers | Share |
---|---|---|
Language Modelling | 1 | 25.00% |
Paraphrase Identification | 1 | 25.00% |
Semantic Textual Similarity | 1 | 25.00% |
Sentence | 1 | 25.00% |
Component | Type |
|
---|---|---|
Mirror-BERT
|
Self-Supervised Learning | (optional) |
SimCSE
|
Sentence Embeddings | (optional) |