The Gaps between Pre-train and Downstream Settings in Bias Evaluation and Debiasing
Kaneko, Masahiro, Bollegala, Danushka, Baldwin, Timothy
–arXiv.org Artificial Intelligence
The output tendencies of Pre-trained Language Models (PLM) vary markedly before and after Fine-Tuning (FT) due to the updates to the model parameters. These divergences in output tendencies result in a gap in the social biases of PLMs. For example, there exits a low correlation between intrinsic bias scores of a PLM and its extrinsic bias scores under FT-based debiasing methods. Additionally, applying FT-based debiasing methods to a PLM leads to a decline in performance in downstream tasks. On the other hand, PLMs trained on large datasets can learn without parameter updates via In-Context Learning (ICL) using prompts. ICL induces smaller changes to PLMs compared to FT-based debiasing methods. Therefore, we hypothesize that the gap observed in pre-trained and FT models does not hold true for debiasing methods that use ICL. In this study, we demonstrate that ICL-based debiasing methods show a higher correlation between intrinsic and extrinsic bias scores compared to FT-based methods. Moreover, the performance degradation due to debiasing is also lower in the ICL case compared to that in the FT case.
arXiv.org Artificial Intelligence
Jan-16-2024
- Country:
- North America
- United States
- Washington > King County
- Seattle (0.04)
- Minnesota > Hennepin County
- Minneapolis (0.14)
- Louisiana > Orleans Parish
- New Orleans (0.04)
- Washington > King County
- Canada > Ontario
- Toronto (0.04)
- United States
- Europe
- Spain > Catalonia
- Barcelona Province > Barcelona (0.04)
- Italy > Tuscany
- Florence (0.04)
- Ireland > Leinster
- County Dublin > Dublin (0.05)
- Denmark > Capital Region
- Copenhagen (0.04)
- Croatia > Dubrovnik-Neretva County
- Dubrovnik (0.04)
- Bulgaria > Sofia City Province
- Sofia (0.04)
- Spain > Catalonia
- Asia
- North America
- Genre:
- Research Report > New Finding (0.66)
- Industry:
- Education (0.68)
- Technology: