Can I read Overwriting Pretrained Bias with Finetuning Data on EtoBox?

Overwriting Pretrained Bias with Finetuning Data by Wang, Angelina; Russakovsky, Olga is a scholarly article available to read on EtoBox.

What is Overwriting Pretrained Bias with Finetuning Data about?

Transfer learning is beneficial by allowing the expressive features of models pretrained on large-scale datasets to be finetuned for the target task of smaller, more domain-specific datasets. However, there is a concern that these pretrained models may come with their own biases which would propagate into the finetuned model. In this work, we investigate bias when conceptualized as both spurious correlations between the target task and a sensitive attribute as well as underrepresentation of a particular group in the dataset. Under both notions of bias, we find that (1) models finetuned on top of pretrained models can indeed inherit their biases, but (2) this bias can be corrected for through relatively minor interventions to the finetuning dataset, and often with a negligible impact to performance. Our findings imply that careful curation of the finetuning dataset is important for reducing biases on a downstream task, and doing so can even compensate for bias in the pretrained model.

Author

Wang, Angelina; Russakovsky, Olga

Published

2023

Language

EN