TY - GEN
T1 - Unlearning Bias in Language Models by Partitioning Gradients
AU - Yu, Charles
AU - Jeoung, Sullam
AU - Kasi, Anish
AU - Yu, Pengfei
AU - Ji, Heng
N1 - Publisher Copyright:
© 2023 Association for Computational Linguistics.
PY - 2023
Y1 - 2023
N2 - Recent research has shown that large-scale pretrained language models, specifically transformers, tend to exhibit issues relating to racism, sexism, religion bias, and toxicity in general. Unfortunately, these pretrained language models are used almost universally in downstream tasks, and natural language processing is often applied to make real-world predictions. Thus, debiasing these language models as early in development as possible is increasingly crucial for preventing unintentional harms caused by natural language systems. To this end, we propose a new technique called partitioned contrastive gradient unlearning (PCGU), a gray-box method for debiasing pretrained masked language models. PCGU aims to optimize only the weights that contribute most to a specific domain of bias, doing so by computing a first-order approximation based on the gradients of contrastive sentence pairs. Our experiments show that PCGU is both low-cost and seems particularly effective at pinpointing the sources of implicit social bias in large pretrained transformers. Although we train using PCGU in the gender-profession domain only, we find that doing so can also partially mitigate bias across other domains. All code for our implementation and experiments can be found at https://github.com/CharlesYu2000/PCGU-UnlearningBias.
AB - Recent research has shown that large-scale pretrained language models, specifically transformers, tend to exhibit issues relating to racism, sexism, religion bias, and toxicity in general. Unfortunately, these pretrained language models are used almost universally in downstream tasks, and natural language processing is often applied to make real-world predictions. Thus, debiasing these language models as early in development as possible is increasingly crucial for preventing unintentional harms caused by natural language systems. To this end, we propose a new technique called partitioned contrastive gradient unlearning (PCGU), a gray-box method for debiasing pretrained masked language models. PCGU aims to optimize only the weights that contribute most to a specific domain of bias, doing so by computing a first-order approximation based on the gradients of contrastive sentence pairs. Our experiments show that PCGU is both low-cost and seems particularly effective at pinpointing the sources of implicit social bias in large pretrained transformers. Although we train using PCGU in the gender-profession domain only, we find that doing so can also partially mitigate bias across other domains. All code for our implementation and experiments can be found at https://github.com/CharlesYu2000/PCGU-UnlearningBias.
UR - http://www.scopus.com/inward/record.url?scp=85162828358&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85162828358&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:85162828358
T3 - Proceedings of the Annual Meeting of the Association for Computational Linguistics
SP - 6032
EP - 6048
BT - Findings of the Association for Computational Linguistics, ACL 2023
PB - Association for Computational Linguistics (ACL)
T2 - 61st Annual Meeting of the Association for Computational Linguistics, ACL 2023
Y2 - 9 July 2023 through 14 July 2023
ER -