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

Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu, Jiatao Gu, Michael Auli
Proceedings of the 39th International Conference on Machine Learning, PMLR 162:1298-1312, 2022.

Abstract

While the general idea of self-supervised learning is identical across modalities, the actual algorithms and objectives differ widely because they were developed with a single modality in mind. To get us closer to general self-supervised learning, we present data2vec, a framework that uses the same learning method for either speech, NLP or computer vision. The core idea is to predict latent representations of the full input data based on a masked view of the input in a self-distillation setup using a standard Transformer architecture. Instead of predicting modality-specific targets such as words, visual tokens or units of human speech which are local in nature, data2vec predicts contextualized latent representations that contain information from the entire input. Experiments on the major benchmarks of speech recognition, image classification, and natural language understanding demonstrate a new state of the art or competitive performance to predominant approaches.

Cite this Paper

BibTeX
@InProceedings{pmlr-v162-baevski22a, title = {data2vec: A General Framework for Self-supervised Learning in Speech, Vision and Language}, author = {Baevski, Alexei and Hsu, Wei-Ning and Xu, Qiantong and Babu, Arun and Gu, Jiatao and Auli, Michael}, booktitle = {Proceedings of the 39th International Conference on Machine Learning}, pages = {1298--1312}, year = {2022}, editor = {Chaudhuri, Kamalika and Jegelka, Stefanie and Song, Le and Szepesvari, Csaba and Niu, Gang and Sabato, Sivan}, volume = {162}, series = {Proceedings of Machine Learning Research}, month = {17--23 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v162/baevski22a/baevski22a.pdf}, url = {https://proceedings.mlr.press/v162/baevski22a.html}, abstract = {While the general idea of self-supervised learning is identical across modalities, the actual algorithms and objectives differ widely because they were developed with a single modality in mind. To get us closer to general self-supervised learning, we present data2vec, a framework that uses the same learning method for either speech, NLP or computer vision. The core idea is to predict latent representations of the full input data based on a masked view of the input in a self-distillation setup using a standard Transformer architecture. Instead of predicting modality-specific targets such as words, visual tokens or units of human speech which are local in nature, data2vec predicts contextualized latent representations that contain information from the entire input. Experiments on the major benchmarks of speech recognition, image classification, and natural language understanding demonstrate a new state of the art or competitive performance to predominant approaches.} }
Endnote
%0 Conference Paper %T data2vec: A General Framework for Self-supervised Learning in Speech, Vision and Language %A Alexei Baevski %A Wei-Ning Hsu %A Qiantong Xu %A Arun Babu %A Jiatao Gu %A Michael Auli %B Proceedings of the 39th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2022 %E Kamalika Chaudhuri %E Stefanie Jegelka %E Le Song %E Csaba Szepesvari %E Gang Niu %E Sivan Sabato %F pmlr-v162-baevski22a %I PMLR %P 1298--1312 %U https://proceedings.mlr.press/v162/baevski22a.html %V 162 %X While the general idea of self-supervised learning is identical across modalities, the actual algorithms and objectives differ widely because they were developed with a single modality in mind. To get us closer to general self-supervised learning, we present data2vec, a framework that uses the same learning method for either speech, NLP or computer vision. The core idea is to predict latent representations of the full input data based on a masked view of the input in a self-distillation setup using a standard Transformer architecture. Instead of predicting modality-specific targets such as words, visual tokens or units of human speech which are local in nature, data2vec predicts contextualized latent representations that contain information from the entire input. Experiments on the major benchmarks of speech recognition, image classification, and natural language understanding demonstrate a new state of the art or competitive performance to predominant approaches.
APA
Baevski, A., Hsu, W., Xu, Q., Babu, A., Gu, J. & Auli, M.. (2022). data2vec: A General Framework for Self-supervised Learning in Speech, Vision and Language. Proceedings of the 39th International Conference on Machine Learning, in Proceedings of Machine Learning Research 162:1298-1312 Available from https://proceedings.mlr.press/v162/baevski22a.html.

Related Material