Representation Surgery: Theory and Practice of Affine Steering

Shashwat Singh, Shauli Ravfogel, Jonathan Herzig, Roee Aharoni, Ryan Cotterell, Ponnurangam Kumaraguru
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:45663-45680, 2024.

Abstract

Language models often exhibit undesirable behavior, e.g., generating toxic or gender-biased text. In the case of neural language models, an encoding of the undesirable behavior is often present in the model’s representations. Thus, one natural (and common) approach to prevent the model from exhibiting undesirable behavior is to steer the model’s representations in a manner that reduces the probability of it generating undesirable text. This paper investigates the formal and empirical properties of steering functions, i.e., transformation of the neural language model’s representations that alter its behavior. First, we derive two optimal, in the least-squares sense, affine steering functions under different constraints. Our theory provides justification for existing approaches and offers a novel, improved steering approach. Second, we offer a series of experiments that demonstrate the empirical effectiveness of the methods in mitigating bias and reducing toxic generation.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-singh24d, title = {Representation Surgery: Theory and Practice of Affine Steering}, author = {Singh, Shashwat and Ravfogel, Shauli and Herzig, Jonathan and Aharoni, Roee and Cotterell, Ryan and Kumaraguru, Ponnurangam}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {45663--45680}, year = {2024}, editor = {Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix}, volume = {235}, series = {Proceedings of Machine Learning Research}, month = {21--27 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v235/main/assets/singh24d/singh24d.pdf}, url = {https://proceedings.mlr.press/v235/singh24d.html}, abstract = {Language models often exhibit undesirable behavior, e.g., generating toxic or gender-biased text. In the case of neural language models, an encoding of the undesirable behavior is often present in the model’s representations. Thus, one natural (and common) approach to prevent the model from exhibiting undesirable behavior is to steer the model’s representations in a manner that reduces the probability of it generating undesirable text. This paper investigates the formal and empirical properties of steering functions, i.e., transformation of the neural language model’s representations that alter its behavior. First, we derive two optimal, in the least-squares sense, affine steering functions under different constraints. Our theory provides justification for existing approaches and offers a novel, improved steering approach. Second, we offer a series of experiments that demonstrate the empirical effectiveness of the methods in mitigating bias and reducing toxic generation.} }
Endnote
%0 Conference Paper %T Representation Surgery: Theory and Practice of Affine Steering %A Shashwat Singh %A Shauli Ravfogel %A Jonathan Herzig %A Roee Aharoni %A Ryan Cotterell %A Ponnurangam Kumaraguru %B Proceedings of the 41st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2024 %E Ruslan Salakhutdinov %E Zico Kolter %E Katherine Heller %E Adrian Weller %E Nuria Oliver %E Jonathan Scarlett %E Felix Berkenkamp %F pmlr-v235-singh24d %I PMLR %P 45663--45680 %U https://proceedings.mlr.press/v235/singh24d.html %V 235 %X Language models often exhibit undesirable behavior, e.g., generating toxic or gender-biased text. In the case of neural language models, an encoding of the undesirable behavior is often present in the model’s representations. Thus, one natural (and common) approach to prevent the model from exhibiting undesirable behavior is to steer the model’s representations in a manner that reduces the probability of it generating undesirable text. This paper investigates the formal and empirical properties of steering functions, i.e., transformation of the neural language model’s representations that alter its behavior. First, we derive two optimal, in the least-squares sense, affine steering functions under different constraints. Our theory provides justification for existing approaches and offers a novel, improved steering approach. Second, we offer a series of experiments that demonstrate the empirical effectiveness of the methods in mitigating bias and reducing toxic generation.
APA
Singh, S., Ravfogel, S., Herzig, J., Aharoni, R., Cotterell, R. & Kumaraguru, P.. (2024). Representation Surgery: Theory and Practice of Affine Steering. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:45663-45680 Available from https://proceedings.mlr.press/v235/singh24d.html.

Related Material