arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2107.07445cs.CLcs.LG
收藏

AutoBERT-Zero:从无到有不断发展的BERT骨干

AutoBERT-Zero: Evolving BERT Backbone from Scratch

The University of Hong Kong(香港大学);Hong Kong University of Science and Technology(香港科技大学);The Education University of Hong Kong(香港教育大学);Sun Yat-sen University(中山大学)

Jiahui Gao, Hang Xu, Han Shi, Xiaozhe Ren, Philip L. H. Yu, Xiaodan Liang, Xin Jiang, Zhenguo Li

摘要:

基于transformer的预训练语言模型,如BERT及其变体,最近在各种自然语言处理(NLP)任务中取得了令人满意的性能。然而,传统的范式构建的骨干,纯粹堆叠的手动设计的全球自我注意层,引入归纳偏差,从而导致次优。在这项工作中,我们首次尝试在灵活的搜索空间上自动发现新的预训练语言模型(PLM)骨干,其中包含从头开始的最基本的操作。具体来说,我们提出了一个精心设计的搜索空间,它(i)包含层内层次的原始数学运算,以探索新的注意力结构,(ii)利用卷积块作为层间层次注意力的补充,以更好地学习局部依赖性。为了提高寻找有前途的架构的效率,我们提出了一个操作优先神经架构搜索(OP-NAS)算法,优化搜索算法和候选模型的评估。具体来说,我们提出了操作优先级(OP)的演化策略,以促进模型搜索,通过平衡探索和开发。此外,我们设计了一个双分支权重共享(BIWS)训练策略,用于快速模型评估。大量的实验表明,搜索的架构(命名为AutoBERT-Zero)显着优于BERT及其变体的不同模型的能力,在各种下游任务,证明了该架构的转移和缩放能力。值得注意的是,AutoBERT-Zero-base在GLUE测试集上的得分比RoBERTa-base(使用更多数据)和BERT-large(模型大小更大)高出2.4和1.4。

英文摘要:

Transformer-based pre-trained language models like BERT and its variants have recently achieved promising performance in various natural language processing (NLP) tasks. However, the conventional paradigm constructs the backbone by purely stacking the manually designed global self-attention layers, introducing inductive bias and thus leads to sub-optimal. In this work, we make the first attempt to automatically discover novel pre-trained language model (PLM) backbone on a flexible search space containing the most fundamental operations from scratch. Specifically, we propose a well-designed search space which (i) contains primitive math operations in the intra-layer level to explore novel attention structures, and (ii) leverages convolution blocks to be the supplementary for attentions in the inter-layer level to better learn local dependency. To enhance the efficiency for finding promising architectures, we propose an Operation-Priority Neural Architecture Search (OP-NAS) algorithm, which optimizes both the search algorithm and evaluation of candidate models. Specifically, we propose Operation-Priority (OP) evolution strategy to facilitate model search via balancing exploration and exploitation. Furthermore, we design a Bi-branch Weight-Sharing (BIWS) training strategy for fast model evaluation. Extensive experiments show that the searched architecture (named AutoBERT-Zero) significantly outperforms BERT and its variants of different model capacities in various downstream tasks, proving the architecture's transfer and scaling abilities. Remarkably, AutoBERT-Zero-base outperforms RoBERTa-base (using much more data) and BERT-large (with much larger model size) by 2.4 and 1.4 higher score on GLUE test set.

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。

↑