Subtracting the weight difference between a model trained on stereotyped text and a pre-trained model reduces measured bias on SEAT by about 0.18 effect-size points without hurting GLUE scores at scale factor one.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Bias Vector: Mitigating Biases in Language Models with Task Arithmetic Approach
Subtracting the weight difference between a model trained on stereotyped text and a pre-trained model reduces measured bias on SEAT by about 0.18 effect-size points without hurting GLUE scores at scale factor one.