Active learning in NLP主动学习的流程可以分为初始化和循环查询两个阶段。在初始化阶段,先随机的从无标签数据集中选取一小部分样本由标注者完成标注,并将这一小部分标注样本作为初始训练集,建立初始的机器学习模型。主动学习的循环阶段有重新训练机器学习模型的步骤,重新训练模型一种方式是用全部语料重新训练模型参数,另一种方式是在已有的模型参数的基础上做模型参数的fine-tuning。对自然语言处理Google发布的BERT新模型就在在11项NLP任务中获得了不错的结果。
EMNLP 2020, pp.4380-4391, (2020)
We propose an expert-in-the-loop training framework Active Learning with Contrastive Explanations to utilize contrastive natural language explanations to improve a learning algorithm’s data efficiency
Cited by1BibtexViews48DOI
0
0
EMNLP, (2018): 2904-2909
This paper provides a largescale empirical study of deep active learning, addressing multiple tasks and, for each, multiple datasets, multiple models, and a full suite of acquisition functions
Cited by35BibtexViews47DOI
0
0
Stanislav Peshterliev, John Kearney,Abhyuday Jagannatha,Imre Kiss,Spyros Matsoukas
arXiv: Computation and Language, (2018)
Our proposed Majority-CRF algorithm leads to statistically significant performance gains over standard active learning and random sampling methods while working with a limited annotation budget
Cited by3BibtexViews43
0
0
international conference on learning representations, (2017)
We proposed deep active learning algorithms for named entity recognition and empirically demonstrated that they achieve state-of-the-art performance with much less data than models trained in the standard supervised fashion
Cited by167BibtexViews48
0
0
empirical methods in natural language processing, (2017): 606-616
We design an active learning algorithm as a policy based on deep reinforcement learning
Cited by83BibtexViews46DOI
0
0
Jesse Thomason, Aishwarya Padmakumar,Jivko Sinapov, Justin Hart,Peter Stone,Raymond J. Mooney
CoRL, pp.67-76, (2017)
We introduce opportunistic active learning, where a system engaged in a task makes use of active learning metrics to query for labels potentially useful for future tasks
Cited by42BibtexViews36
0
0
ACL (Short Papers), pp.6-10, (2011)
We introduced a novel approach to seeding Active learning, in which the seeds are selected from the examples with low Language Model probability
Cited by22BibtexViews3
0
0
ACL, pp.1552-1561, (2010)
This paper presents BAGEL, a statistical language generator which uses dynamic Bayesian networks to learn from semantically-aligned data produced by 42 untrained annotators
Cited by91BibtexViews39
0
0
(2009)
This report is a survey of the literature relevant to active machine learning in the context of natural language processing
Cited by208BibtexViews28
0
0
EMNLP '09 Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing: Vo..., pp.296-305, (2009)
Through actual annotation experiments that control for several factors, we have evaluated the potential of incorporating active learning and label suggestions to speed up morpheme glossing in a realistic language documentation context
Cited by57BibtexViews5
0
0
EMNLP, pp.1070-1079, (2008)
Our large-scale empirical evaluation demonstrates that some of these newly proposed methods advance the state of the art in active learning with sequence models
Cited by716BibtexViews101
0
0
(2008)
We have presented an extensive empirical study of annotation costs in four real-world text and image domains
Cited by218BibtexViews12
0
0
ACM Multimedia 2001, pp.892-899, (2004)
Compared to other approaches for automatic image annotation, the proposed model takes an advantage of word-to-word correlation
Cited by167BibtexViews17DOI
0
0
ACL '02 Proceedings of the 40th Annual Meeting on Association for Computational Linguistics, pp.120-127, (2002)
This paper presents a method to reduce this demand using active learning, which selects what samples to annotate, instead of annotating blindly the whole training corpus
Cited by191BibtexViews23DOI
0
0
ICASSP, pp.3904-3907, (2002)
For active learning in automatic speech recognition, we trained language and acoustic models using the initial set of 4,000 utterances
Cited by135BibtexViews11DOI
0
0
ICML, pp.406-414, (1999)
Active learning is a new area of machine learning that has been almost exclusively applied to classification tasks
Cited by413BibtexViews42
0
0