Uncertainty sampling based active learning with diversity constraint by sparse selection

Wang, Gaoang; Hwang, Jenq‐Neng; Rose, Craig S.; Wallace, Farron

doi:10.1109/mmsp.2017.8122269

Cited by 17 publications

(10 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…5) as in (Hwa, 2004;Joshi et al, 2009). Given that the difference in degree of certainty for similar examples can be small, uncertainty selections are prone to return similar examples (Wang et al, 2017). To address this issue, some works incorporate measures to exploit the diversity information of the examples in the selection process (Sener and Savarese, 2017;Wang et al, 2017;Sinha et al, 2019).…”

Section: Active Learningmentioning

confidence: 99%

“…Given that the difference in degree of certainty for similar examples can be small, uncertainty selections are prone to return similar examples (Wang et al, 2017). To address this issue, some works incorporate measures to exploit the diversity information of the examples in the selection process (Sener and Savarese, 2017;Wang et al, 2017;Sinha et al, 2019). Finally, expected-model change selects examples that would cause the greatest change to a model's output if their labels were known (Freytag et al, 2014;Roy and McCallum, 2001;.…”

Section: Active Learningmentioning

confidence: 99%

See 1 more Smart Citation

Textual Data Augmentation for Efficient Active Learning on Tiny Datasets

Quteineh¹,

Samothrakis²,

Sutcliffe³

2020

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)

View full text Add to dashboard Cite

In this paper we propose a novel data augmentation approach where guided outputs of a language generation model, e.g. GPT-2, when labeled, can improve the performance of text classifiers through an active learning process. We transform the data generation task into an optimization problem which maximizes the usefulness of the generated output, using Monte Carlo Tree Search (MCTS) as the optimization strategy and incorporating entropy as one of the optimization criteria. We test our approach against a Non-Guided Data Generation (NGDG) process that does not optimize for a reward function. Starting with a small set of data, our results show an increased performance with MCTS of 26% on the TREC-6 Questions dataset, and 10% on the Stanford Sentiment Treebank SST-2 dataset. Compared with NGDG, we are able to achieve increases of 3% and 5% on TREC-6 and SST-2.

show abstract

Section: Active Learningmentioning

confidence: 99%

Section: Active Learningmentioning

confidence: 99%

Textual Data Augmentation for Efficient Active Learning on Tiny Datasets

Quteineh¹,

Samothrakis²,

Sutcliffe³

2020

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)

View full text Add to dashboard Cite

show abstract

“…Diversity constraint sampling (Diversity-AL). Diversity sampling methods have been used in information retrieval (Xu et al, 2007) and image classification (Wang et al, 2017b). The core idea is that samples that are highly similar to each other typically yield little new information and thus low performance.…”

Section: Active Learningmentioning

confidence: 99%

“…For text compression, we suggest AL strategies to maximize the model's coverage and the diversity of the samples. To this end, we build upon work in uncertainty sampling by (Peris and Casacuberta, 2018;Wang et al, 2017b) and propose a new strategy to predict the sample diversity at a structural level.…”

Section: Neural Seq2seq Text Compressionmentioning

confidence: 99%

Data-efficient Neural Text Compression with Interactive Learning

P.V.S¹,

Meyer

2019

Proceedings of the 2019 Conference of the North

View full text Add to dashboard Cite

Neural sequence-to-sequence models have been successfully applied to text compression. However, these models were trained on huge automatically induced parallel corpora, which are only available for a few domains and tasks. In this paper, we propose a novel interactive setup to neural text compression that enables transferring a model to new domains and compression tasks with minimal human supervision. This is achieved by employing active learning, which intelligently samples from a large pool of unlabeled data. Using this setup, we can successfully adapt a model trained on small data of 40k samples for a headline generation task to a general text compression dataset at an acceptable compression quality with just 500 sampled instances annotated by a human.

show abstract

“…Du et al proposed a robust multi-label active learning algorithm which introduced maximum correntropy criterion as the measure of uncertainty [15]. In [16], [17], the sparse modeling was incorporated into this samples selection to address the problem of redundant information between uncertain samples. In query-by-committee method, the query is formulated according to the criterion of minimal agreement.…”

Section: Introductionmentioning

confidence: 99%

Active Informative Pairwise Constraint Formulation Algorithm for Constraint-Based Clustering

Zhong

Deng

2019

IEEE Access

View full text Add to dashboard Cite

Constraint-based clustering utilizes pairwise constraints to improve clustering performance. In this paper, we propose a novel formulation algorithm to generate more informative pairwise constraints from limited queries for the constraint-based clustering. Our method consists of two phases: pre-clustering and marking. The pre-clustering phase introduces the fuzzy c-means clustering (FCM) to generate the cluster knowledge that is composed of the membership degree and the cluster centers. In the marking phase, we first propose the weak sample with the larger uncertainty expressed by the entropy of the membership degree. Then, we study the strong sample that contains less uncertainty and should be closest to its cluster center. Finally, given weak samples in descending order of entropy, we formulate informative pairs with strong samples and seek answers using the second minimal symmetric relative entropy priority principle, which leads to more efficient queries. Making use of the pairwise constraint k-means clustering (PCKM) as the underlying constraint-based clustering algorithm, further data experiments are conducted in several datasets to verify the improvement of our method.INDEX TERMS Constraint-based clustering, pairwise constraint, weak sample, strong sample, symmetric relative entropy.

show abstract

Uncertainty sampling based active learning with diversity constraint by sparse selection

Cited by 17 publications

References 12 publications

Textual Data Augmentation for Efficient Active Learning on Tiny Datasets

Textual Data Augmentation for Efficient Active Learning on Tiny Datasets

Data-efficient Neural Text Compression with Interactive Learning

Active Informative Pairwise Constraint Formulation Algorithm for Constraint-Based Clustering

Contact Info

Product

Resources

About