[25] data2vec: A General Framework for Self-supervised Learning in Speech, Vision and Language

[Dongwoo-Im]

Links

• Paper : https://arxiv.org/abs/2202.03555
• Github : https://github.com/facebookresearch/fairseq/tree/main/examples/data2vec

한 줄 요약

• Vision, speech, language 도메인에 masked prediction task를 활용한 self-supervised learning objective 제안

선택 이유

• Masked prediction objective를 다양한 modality에 어떻게 적용하였는지 궁금해서 선정

[Dongwoo-Im]

https://dongwoo-im.github.io/papers/review/2023-03-20-data2vec/