Multimodal Pretraining Unmasked: A Meta-Analysis and a Unified Framework of Vision-and-Language BERTs

Author:

Bugliarello Emanuele1,Cotterell Ryan23,Okazaki Naoaki4,Elliott Desmond5

Affiliation:

1. University of Copenhagen. emanuele@di.ku.dk

2. University of Cambridge

3. ETH Zürich. rcotterell@inf.ethz.ch

4. Tokyo Institute of Technology. okazaki@c.titech.ac.jp

5. University of Copenhagen. de@di.ku.dk

Abstract

Abstract Large-scale pretraining and task-specific fine- tuning is now the standard methodology for many tasks in computer vision and natural language processing. Recently, a multitude of methods have been proposed for pretraining vision and language BERTs to tackle challenges at the intersection of these two key areas of AI. These models can be categorized into either single-stream or dual-stream encoders. We study the differences between these two categories, and show how they can be unified under a single theoretical framework. We then conduct controlled experiments to discern the empirical differences between five vision and language BERTs. Our experiments show that training data and hyperparameters are responsible for most of the differences between the reported results, but they also reveal that the embedding layer plays a crucial role in these massive models.

Publisher

MIT Press - Journals

Reference53 articles.

1. Bottom-up and top-down attention for image captioning and visual question answering;Anderson,2018

2. VQA: Visual Question Answering;Antol,2015

3. Layer normalization;Ba;arXiv preprint arXiv:1607.06450,2016

4. Redefine statistical significance;Benjamin;Nature Human Behaviour,2018

5. Uniter: Universal image- text representation learning;Chen,2020

Cited by 41 articles. 订阅此论文施引文献 订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献

同舟云学术

1.学者识别学者识别

2.学术分析学术分析

3.人才评估人才评估

"同舟云学术"是以全球学者为主线,采集、加工和组织学术论文而形成的新型学术文献查询和分析系统,可以对全球学者进行文献检索和人才价值评估。用户可以通过关注某些学科领域的顶尖人物而持续追踪该领域的学科进展和研究前沿。经过近期的数据扩容,当前同舟云学术共收录了国内外主流学术期刊6万余种,收集的期刊论文及会议论文总量共计约1.5亿篇,并以每天添加12000余篇中外论文的速度递增。我们也可以为用户提供个性化、定制化的学者数据。欢迎来电咨询!咨询电话:010-8811{复制后删除}0370

www.globalauthorid.com

TOP

Copyright © 2019-2024 北京同舟云网络信息技术有限公司
京公网安备11010802033243号  京ICP备18003416号-3