Active Sentence Learning by Adversarial Uncertainty Sampling in Discrete Space

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 4908–4917 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 4908 Active Sentence Learning by Adversarial Uncertainty Sampling in Discrete Space Dongyu Ru†,‡, Jiangtao Feng†, Lin Qiu‡, Hao Zhou†, Mingxuan Wang†, Weinan Zhang‡, Yong Yu‡, Lei Li† †ByteDance AI Lab {fengjiangtao,zhouhao.nlp,wangmingxuan.89,lileilab}@bytedance.com ‡Shanghai Jiao Tong University {maxru,lqiu,wnzhang,yyu}@apex.sjtu.edu.cn Abstract Active learning for sentence understanding aims at discovering informative unlabeled data for annotation and therefore reducing the de- mand for labeled data. We argue that the typical uncertainty sampling method for ac- tive learning is time-consuming and can hardly work in real-time, which may lead to inef- fective sample selection. We propose adver- sarial uncertainty sampling in discrete space (AUSDS) to retrieve informative unlabeled samples more efficiently. AUSDS maps sen- tences into latent space generated by the popu- lar pre-trained language models, and discover informative unlabeled text samples for annota- tion via adversarial attack. The proposed ap- proach is extremely efficient compared with traditional uncertainty sampling with more than 10x speedup. Experimental results on five datasets show that AUSDS outperforms strong baselines on effectiveness. 1 Introduction Deep neural models become popular in natural lan- guage processing (Peters et al., 2018; Radford et al., 2018; Devlin et al., 2018). Neural models usually consume massive labeled data, which requires a huge quantity of human labors. But data are not born equal, where informative data with high un- certainty are decisive to decision boundary and are worth labeling. Thus selecting such worth-labeling data from unlabeled text corpus for annotation is an effective way to reduce the human labors and to obtain informative data. Active learning approaches are a straightfor- ward choice to reduce such human labors. Pre- vious works, such as uncertainty sampling (Lewis and Gale, 1994), needs to traverse all unlabeled data to find informative unlabeled samples, which are always near the decision boundary with large entropy. However, the traverse process is very time-consuming, thus cannot be executed fre- quently (Settles and Craven, 2008). A common choice is to perform the sampling process after every specific period, and it samples and labels in- formative unlabeled data then trains the model until convergence (Deng et al., 2018). We argue that infrequently performing uncer- tainty sampling may lead to the “ineffective sam- pling” problem. Because in the early phase of train- ing, the decision boundary changes quickly, which makes previously collected samples less effective after several updates of the model. Ideally, uncer- tainty sampling should be performed frequently in the early phase of model training. In this paper, we propose the adversarial uncer- tainty sampling in discrete space (AUSDS) to ad- dress the ineffective sampling problem for active sentence learning by introducing more frequent sampling with significantly lower costs. Specif- ically, we propose to leverage the adversarial at- tack (Goodfellow et al., 2014; Kurakin et al., 2016) to the selecting of informative samples with high uncertainty, which significantly narrows down the search space. Fig. 1 shows the difference between uncertainty sampling and AUSDS. The typical un- …

Content truncated to 500 words