Dynamic Data Selection for Curriculum Learning via Ability Estimation

Dynamic Data Selection for Curriculum Learning via Ability Estimation. Lalor, J. P. & Yu, H. In Findings of the Association for Computational Linguistics: EMNLP 2020, pages 545–555, Online, November, 2020. Association for Computational Linguistics.

Paper abstract bibtex

Curriculum learning methods typically rely on heuristics to estimate the difficulty of training examples or the ability of the model. In this work, we propose replacing difficulty heuristics with learned difficulty parameters. We also propose Dynamic Data selection for Curriculum Learning via Ability Estimation (DDaCLAE), a strategy that probes model ability at each training epoch to select the best training examples at that point. We show that models using learned difficulty and/or ability outperform heuristic-based curriculum learning models on the GLUE classification tasks.

@inproceedings{lalor_dynamic_2020,
	address = {Online},
	title = {Dynamic {Data} {Selection} for {Curriculum} {Learning} via {Ability} {Estimation}},
	url = {https://www.aclweb.org/anthology/2020.findings-emnlp.48},
	abstract = {Curriculum learning methods typically rely on heuristics to estimate the difficulty of training examples or the ability of the model. In this work, we propose replacing difficulty heuristics with learned difficulty parameters. We also propose Dynamic Data selection for Curriculum Learning via Ability Estimation (DDaCLAE), a strategy that probes model ability at each training epoch to select the best training examples at that point. We show that models using learned difficulty and/or ability outperform heuristic-based curriculum learning models on the GLUE classification tasks.},
	urldate = {2020-11-29},
	booktitle = {Findings of the {Association} for {Computational} {Linguistics}: {EMNLP} 2020},
	publisher = {Association for Computational Linguistics},
	author = {Lalor, John P. and Yu, Hong},
	month = nov,
	year = {2020},
	pmid = {33381774 PMCID: PMC7771727},
	pages = {545--555},
}

Downloads: 0

{"_id":"H3rSCmKwzvdGDco7h","bibbaseid":"lalor-yu-dynamicdataselectionforcurriculumlearningviaabilityestimation-2020","author_short":["Lalor, J. P.","Yu, H."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","address":"Online","title":"Dynamic Data Selection for Curriculum Learning via Ability Estimation","url":"https://www.aclweb.org/anthology/2020.findings-emnlp.48","abstract":"Curriculum learning methods typically rely on heuristics to estimate the difficulty of training examples or the ability of the model. In this work, we propose replacing difficulty heuristics with learned difficulty parameters. We also propose Dynamic Data selection for Curriculum Learning via Ability Estimation (DDaCLAE), a strategy that probes model ability at each training epoch to select the best training examples at that point. We show that models using learned difficulty and/or ability outperform heuristic-based curriculum learning models on the GLUE classification tasks.","urldate":"2020-11-29","booktitle":"Findings of the Association for Computational Linguistics: EMNLP 2020","publisher":"Association for Computational Linguistics","author":[{"propositions":[],"lastnames":["Lalor"],"firstnames":["John","P."],"suffixes":[]},{"propositions":[],"lastnames":["Yu"],"firstnames":["Hong"],"suffixes":[]}],"month":"November","year":"2020","pmid":"33381774 PMCID: PMC7771727","pages":"545–555","bibtex":"@inproceedings{lalor_dynamic_2020,\n\taddress = {Online},\n\ttitle = {Dynamic {Data} {Selection} for {Curriculum} {Learning} via {Ability} {Estimation}},\n\turl = {https://www.aclweb.org/anthology/2020.findings-emnlp.48},\n\tabstract = {Curriculum learning methods typically rely on heuristics to estimate the difficulty of training examples or the ability of the model. In this work, we propose replacing difficulty heuristics with learned difficulty parameters. We also propose Dynamic Data selection for Curriculum Learning via Ability Estimation (DDaCLAE), a strategy that probes model ability at each training epoch to select the best training examples at that point. We show that models using learned difficulty and/or ability outperform heuristic-based curriculum learning models on the GLUE classification tasks.},\n\turldate = {2020-11-29},\n\tbooktitle = {Findings of the {Association} for {Computational} {Linguistics}: {EMNLP} 2020},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {Lalor, John P. and Yu, Hong},\n\tmonth = nov,\n\tyear = {2020},\n\tpmid = {33381774 PMCID: PMC7771727},\n\tpages = {545--555},\n}\n\n","author_short":["Lalor, J. P.","Yu, H."],"key":"lalor_dynamic_2020","id":"lalor_dynamic_2020","bibbaseid":"lalor-yu-dynamicdataselectionforcurriculumlearningviaabilityestimation-2020","role":"author","urls":{"Paper":"https://www.aclweb.org/anthology/2020.findings-emnlp.48"},"metadata":{"authorlinks":{}},"html":""},"bibtype":"inproceedings","biburl":"http://fenway.cs.uml.edu/papers/pubs-all.bib","dataSources":["TqaA9miSB65nRfS5H"],"keywords":[],"search_terms":["dynamic","data","selection","curriculum","learning","via","ability","estimation","lalor","yu"],"title":"Dynamic Data Selection for Curriculum Learning via Ability Estimation","year":2020}