FairSteer: Inference Time Debiasing for LLMs with Dynamic Activation Steering

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Li, Yichen, Fan, Zhiting, Chen, Ruizhe, Gai, Xiaotang, Gong, Luqi, Zhang, Yan, Liu, Zuozhu
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916828440166400
author Li, Yichen
Fan, Zhiting
Chen, Ruizhe
Gai, Xiaotang
Gong, Luqi
Zhang, Yan
Liu, Zuozhu
author_facet Li, Yichen
Fan, Zhiting
Chen, Ruizhe
Gai, Xiaotang
Gong, Luqi
Zhang, Yan
Liu, Zuozhu
contents Large language models (LLMs) are prone to capturing biases from training corpus, leading to potential negative social impacts. Existing prompt-based debiasing methods exhibit instability due to their sensitivity to prompt changes, while fine-tuning-based techniques incur substantial computational overhead and catastrophic forgetting. In this paper, we propose FairSteer, a novel inference-time debiasing framework without requiring customized prompt design or model retraining. Motivated by the linear representation hypothesis, our preliminary investigation demonstrates that fairness-related features can be encoded into separable directions in the hidden activation space. FairSteer operates in three steps: biased activation detection, debiasing steering vector (DSV) computation, and dynamic activation steering. Specifically, it first trains a lightweight linear classifier to detect bias signatures in activations, and then computes DSVs as intervention directions derived from small contrastive prompt pairs. Subsequently, it performs debiasing by adjusting activations with DSVs in the inference stage. Comprehensive evaluation with six LLMs demonstrates the superiority of FairSteer across question-answering, counterfactual input evaluation and open-ended text generation tasks. Code will be released.
format Preprint
id arxiv_https___arxiv_org_abs_2504_14492
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle FairSteer: Inference Time Debiasing for LLMs with Dynamic Activation Steering
Li, Yichen
Fan, Zhiting
Chen, Ruizhe
Gai, Xiaotang
Gong, Luqi
Zhang, Yan
Liu, Zuozhu
Computation and Language
Large language models (LLMs) are prone to capturing biases from training corpus, leading to potential negative social impacts. Existing prompt-based debiasing methods exhibit instability due to their sensitivity to prompt changes, while fine-tuning-based techniques incur substantial computational overhead and catastrophic forgetting. In this paper, we propose FairSteer, a novel inference-time debiasing framework without requiring customized prompt design or model retraining. Motivated by the linear representation hypothesis, our preliminary investigation demonstrates that fairness-related features can be encoded into separable directions in the hidden activation space. FairSteer operates in three steps: biased activation detection, debiasing steering vector (DSV) computation, and dynamic activation steering. Specifically, it first trains a lightweight linear classifier to detect bias signatures in activations, and then computes DSVs as intervention directions derived from small contrastive prompt pairs. Subsequently, it performs debiasing by adjusting activations with DSVs in the inference stage. Comprehensive evaluation with six LLMs demonstrates the superiority of FairSteer across question-answering, counterfactual input evaluation and open-ended text generation tasks. Code will be released.
title FairSteer: Inference Time Debiasing for LLMs with Dynamic Activation Steering
topic Computation and Language
url https://arxiv.org/abs/2504.14492