Стратегия тонкой настройки
method
ID: finetuningstrategy
Related Papers
BERxiT: Early Exiting for BERT with Better Fine-Tuning and Extension to Regression
AbstractThe slow speed of BERT has motivated much research on accelerating its inference, and the early exiting idea has been proposed to make trade-offs between model quality and efficiency. This …...
Reducing Disambiguation Biases in NMT by Leveraging Explicit Word Sense Information
AbstractRecent studies have shed some light on a common pitfall of Neural Machine Translation (NMT) models, stemming from their struggle to disambiguate polysemous words without lapsing into their most frequently …...
Guide the Learner: Controlling Product of Experts Debiasing Method Based on Token Attribution Similarities
AbstractSeveral proposals have been put forward in recent years for improving out-of-distribution (OOD) performance through mitigating dataset biases. A popular workaround is to train a robust model by re-weighting training …...