@inproceedings{32a5d533fcf44787ac1c7a5abef0b87f,
title = "Let the Model Decide its Curriculum for Multitask Learning",
abstract = "Curriculum learning strategies in prior multitask learning approaches arrange datasets in a difficulty hierarchy either based on human perception or by exhaustively searching the optimal arrangement. However, human perception of difficulty may not always correlate well with machine interpretation leading to poor performance and exhaustive search is computationally expensive. Addressing these concerns, we propose two classes of techniques to arrange training instances into a learning curriculum based on difficulty scores computed via model-based approaches. The two classes i.e Dataset-level and Instance-level differ in granularity of arrangement. Through comprehensive experiments with 12 datasets, we show that instance-level and dataset-level techniques result in strong representations as they lead to an average performance improvement of 4.17% and 3.15% over their respective baselines. Furthermore, we find that most of this improvement comes from correctly answering the difficult instances, implying a greater efficacy of our techniques on difficult tasks.",
author = "Neeraj Varshney and Swaroop Mishra and Chitta Baral",
note = "Funding Information: We thank the anonymous reviewers for their insightful feedback. This research was supported by DARPA SAIL-ON program. Publisher Copyright: {\textcopyright} 2022 Association for Computational Linguistics.; 3rd Workshop on Deep Learning Approaches for Low-Resource NLP, DeepLo 2022 ; Conference date: 14-07-2022",
year = "2022",
language = "English (US)",
series = "DeepLo 2022 - 3rd Workshop on Deep Learning Approaches for Low-Resource NLP, Proceedings of the DeepLo Workshop",
publisher = "Association for Computational Linguistics (ACL)",
pages = "117--125",
editor = "Colin Cherry and Angela Fan and George Foster and Gholamreza Haffari and Shahram Khadivi and Nanyun Peng and Xiang Ren and Ehsan Shareghi and Swabha Swayamdipta",
booktitle = "DeepLo 2022 - 3rd Workshop on Deep Learning Approaches for Low-Resource NLP, Proceedings of the DeepLo Workshop",
}