arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2103.05247cs.LGcs.AI
收藏

作为通用计算引擎的预训练Transformers

Pretrained Transformers as Universal Computation Engines

UC Berkeley(加州大学伯克利分校);Facebook AI Research(脸书人工智能研究院);Google Brain(谷歌大脑)

Kevin Lu, Aditya Grover, Pieter Abbeel, Igor Mordatch

摘要:

我们研究了在自然语言上预训练的Transformer以最小的微调推广到其他模态的能力-特别是,没有微调残差块的自注意和前馈层。我们考虑这样一个模型,我们称之为冻结预训练Transformer(FPT),并研究在各种序列分类任务上对其进行微调,这些任务包括数值计算,视觉和蛋白质折叠预测。与之前研究在与预训练数据集相同的模态上进行微调的工作相比,我们表明,在自然语言上进行预训练可以提高非语言下游任务的性能和计算效率。此外,我们对架构进行了分析,比较了随机初始化的Transformer与随机LSTM的性能。结合这两种见解,我们发现语言预训练的Transformers可以在各种非语言任务上获得强大的性能。

英文摘要:

We investigate the capability of a transformer pretrained on natural language to generalize to other modalities with minimal finetuning -- in particular, without finetuning of the self-attention and feedforward layers of the residual blocks. We consider such a model, which we call a Frozen Pretrained Transformer (FPT), and study finetuning it on a variety of sequence classification tasks spanning numerical computation, vision, and protein fold prediction. In contrast to prior works which investigate finetuning on the same modality as the pretraining dataset, we show that pretraining on natural language can improve performance and compute efficiency on non-language downstream tasks. Additionally, we perform an analysis of the architecture, comparing the performance of a random initialized transformer to a random LSTM. Combining the two insights, we find language-pretrained transformers can obtain strong performance on a variety of non-language tasks.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑