Learning to select data for transfer learning with Bayesian Optimization

Ruder, Sebastian; Plank, Barbara

doi:10.18653/v1/d17-1038

Cited by 146 publications

(151 citation statements)

References 29 publications

Supporting

Mentioning

140

Contrasting

Unclassified

Order By: Relevance

“…Ammar et al (2016) and Although not for cross-lingual transfer, there has been prior work on data selection for training models. Tsvetkov et al (2016a) and Ruder and Plank (2017) use Bayesian optimization for data selection. van der Wees et al (2017) study the effect of data selection of neural machine translation, as well as propose a dynamic method to select relevant training data that improves translation performance.…”

Section: Related Workmentioning

confidence: 99%

Choosing Transfer Languages for Cross-Lingual Learning

Lin¹,

Chen²,

Lee³

et al. 2019

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics

102

View full text Add to dashboard Cite

Cross-lingual transfer, where a high-resource transfer language is used to improve the accuracy of a low-resource task language, is now an invaluable tool for improving performance of natural language processing (NLP) on lowresource languages. However, given a particular task language, it is not clear which language to transfer from, and the standard strategy is to select languages based on ad hoc criteria, usually the intuition of the experimenter. Since a large number of features contribute to the success of cross-lingual transfer (including phylogenetic similarity, typological properties, lexical overlap, or size of available data), even the most enlightened experimenter rarely considers all these factors for the particular task at hand. In this paper, we consider this task of automatically selecting optimal transfer languages as a ranking problem, and build models that consider the aforementioned features to perform this prediction. In experiments on representative NLP tasks, we demonstrate that our model predicts good transfer languages much better than ad hoc baselines considering single features in isolation, and glean insights on what features are most informative for each different NLP tasks, which may inform future ad hoc selection even without use of our method. 1 * Equal contribution 1 Code, data, and pre-trained models are available at

show abstract

Section: Related Workmentioning

confidence: 99%

Choosing Transfer Languages for Cross-Lingual Learning

Lin¹,

Chen²,

Lee³

et al. 2019

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics

102

View full text Add to dashboard Cite

show abstract

“…However, we focus on supervised sequence labeling domain adaptation, where huge improvement can be achieved by utilizing only small-scale annotated data from the target domain. Previous works in domain adaptation often try to find a subset of source domain data to align with the target domain data (Chopra et al, 2013;Ruder and Plank, 2017) which realizes a kind of source data sample or construct a common feature space, while those methods may wash out informative characteristics of target domain samples. Instance-based domain adaptation (Jiang and Zhai, 2007;Zhang and Xiong, 2018) implement the source sample weighting by assigning higher weights to source domain samples which are more similar to the target domain.…”

Section: Related Workmentioning

confidence: 99%

Fine-grained Knowledge Fusion for Sequence Labeling Domain Adaptation

Yang¹,

Huang²,

Dai³

et al. 2019

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conferen

View full text Add to dashboard Cite

In sequence labeling, previous domain adaptation methods focus on the adaptation from the source domain to the entire target domain without considering the diversity of individual target domain samples, which may lead to negative transfer results for certain samples. Besides, an important characteristic of sequence labeling tasks is that different elements within a given sample may also have diverse domain relevance, which requires further consideration. To take the multi-level domain relevance discrepancy into account, in this paper, we propose a fine-grained knowledge fusion model with the domain relevance modeling scheme to control the balance between learning from the target domain data and learning from the source domain model. Experiments on three sequence labeling tasks show that our fine-grained knowledge fusion model outperforms strong baselines and other stateof-the-art sequence labeling domain adaptation methods.

show abstract

“…The first category makes the assumption that labeled data from both source and target domains are available to us, though the amount may differ [6,43]. While the second category assumes that no labeled data from the target domain is available in addition to the labeled source domain data [27,29]. Our work falls into the first category.…”

Section: Related Workmentioning

confidence: 99%

“…In this case, there are also two categories. The first is the instance based methods, which select or reweight the source domain training samples so that data from the source domain and the target domain would share a similar data distribution [4,11,27]. The second category is feature based methods, which aim to locate a common feature space that can reduce the differences between the source and target domains.…”

Section: Related Workmentioning

confidence: 99%

“…The DNN based TL framework has been proven to be effective in deep text matching tasks for question answering systems [43]. Although various data selection methods [4,11,23,27] were proposed for TL settings, most of them do not fit well with neural transfer models, because the data selector/reweighter is not jointly trained with the TL model. Specifically, the TL task model is considered as a sub-module of the data selection framework.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Learning to Selectively Transfer

Qiu

et al. 2019

Proceedings of the Twelfth ACM International Conference on Web Search and Data Mining

View full text Add to dashboard Cite

Deep text matching approaches have been widely studied for many applications including question answering and information retrieval systems. To deal with a domain that has insufficient labeled data, these approaches can be used in a Transfer Learning (TL) setting to leverage labeled data from a resource-rich source domain. To achieve better performance, source domain data selection is essential in this process to prevent the "negative transfer" problem. However, the emerging deep transfer models do not fit well with most existing data selection methods, because the data selection policy and the transfer learning model are not jointly trained, leading to sub-optimal training efficiency.In this paper, we propose a novel reinforced data selector to select high-quality source domain data to help the TL model. Specifically, the data selector "acts" on the source domain data to find a subset for optimization of the TL model, and the performance of the TL model can provide "rewards" in turn to update the selector. We build the reinforced data selector based on the actor-critic framework and integrate it to a DNN based transfer learning model, resulting in a Reinforced Transfer Learning (RTL) method. We perform a thorough experimental evaluation on two major tasks for text matching, namely, paraphrase identification and natural language inference. Experimental results show the proposed RTL can significantly improve the performance of the TL model. We further investigate different settings of states, rewards, and policy optimization methods to examine the robustness of our method. Last, we conduct a case study on the selected data and find our method is able to select source domain data whose Wasserstein distance is close to the target domain data. This is reasonable and intuitive as * Work done during an internship at Alibaba Group.

show abstract

Learning to select data for transfer learning with Bayesian Optimization

Cited by 146 publications

References 29 publications

Choosing Transfer Languages for Cross-Lingual Learning

Choosing Transfer Languages for Cross-Lingual Learning

Fine-grained Knowledge Fusion for Sequence Labeling Domain Adaptation

Learning to Selectively Transfer

Contact Info

Product

Resources

About