💬 NLP
Influence-driven Curriculum Learning for Pre-training on Limited Data
This paper demonstrates that replacing conventional human-centered difficulty metrics with model-centric training data influence scores enables curriculum learning to significantly outperform random-order pre-training on limited data, achieving over a 10 percentage point improvement in benchmarks.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.