Logo image
Back
LLM Bias Detection and Mitigation through the Lens of Desired Distributions
Conference proceeding   Open access

LLM Bias Detection and Mitigation through the Lens of Desired Distributions

Ingroj Shrestha and Padmini Srinivasan
Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pp.1464-1480
01/01/2025
DOI: 10.18653/v1/2025.emnlp-main.76
url
https://doi.org/10.18653/v1/2025.emnlp-main.76View
Published (Version of record) Open Access

Abstract

Although prior work on bias mitigation has focused on promoting social equality and demographic parity, less attention has been given to aligning LLM's outputs to desired distributions. For example, we might want to align a model with real-world distributions to support factual grounding. Thus, we define bias as deviation from a desired distribution, which may be an equal or real-world distribution, depending on application goals. We propose a weighted adaptive loss(1) based fine-tuning method that aligns LLM's gender-profession output distribution with the desired distribution, while preserving language modeling capability. Using 3 profession sets-male-dominated, female-dominated, and gender-balanced-derived from U.S. labor statistics (2024), we assess both our adaptive method for reflecting reality and a non-adaptive variant for equality. Across three masked language models, bias is observed under both distributions. We achieve near-complete mitigation under equality and 30-75% reduction under real-world settings. Autoregressive LLMs show no bias under equality but notable bias under real-world settings, with the Llama Instruct models (3.2-3B, 3.1-8B) achieving a 50-62% reduction.
 Computer Science Social Sciences Technology Computer Science, Artificial Intelligence Computer Science, Information Systems Computer Science, Theory & Methods Linguistics Science & Technology

Details

Metrics

2 Record Views
1 Times Cited - Web of Science
Logo image