题目: Data Augmentation using Pre-trained Transformer Models
简介:
基于语言模型的预训练模型,如BERT,在不同的NLP任务中提供了显著的收益。在本文中,我们研究了不同类型的基于自回归模型(GPT-2)、自编码器模型(BERT)和seq2seq模型(BART)等用于条件数据增强的预训练变压器模型。我们表明,将类标签前置到文本序列提供了一种简单而有效的方法来设置预训练模型的条件,以便进行数据扩充。在三个分类基准上,预先训练的Seq2Seq模型优于其他模型。此外,我们还探讨了不同的基于预训练模型的数据扩充在数据多样性方面是如何不同的,以及这些方法如何很好地保存类标签信息。
There are many sectors which have moved to Cloud and are planning aggressively to move their workloads to Cloud since the world entered Covid-19 pandemic. There are various reasons why Cloud is an essential irresistible technology and serves as an ultimate solution to access IT software and systems. It has become a new essential catalyst for Enterprise Organisations which are looking for Digital Transformation. Remote working is a common phenomenon now across all the IT companies making the services available all the time. Covid-19 has made cloud adoption an immediate priority for Organisation rather than a slowly approached future transformation. The benefits of Cloud lies in the fact that employees rather engineers of an enterprise are no more dependent on the closed hardware-based IT infrastructure and hence eliminates the necessity of working from the networked office premises. This has raised a huge demand for skilled Cloud specialist who can manage and support the systems running on cloud across different regions of the world. In this research, the reasons for growing Cloud adoption after pandemic Covid-19 has been described and the challenges which Organization will face is also explained. This study also details the most used cloud services during the pandemic considering Amazon Web Services as the cloud provider.