Collective Bias Mitigation via Model Routing and Collaboration
Collective Bias Mitigation routes multiple LLMs, and a committee cut age bias from 0.25 to 0.10.
The paper introduces Collective Bias Mitigation, a framework that learns fine-grained model behavior and shares knowledge across diverse LLMs to reduce bias beyond single-model self-debiasing. In a top-7 setting, the Committee topology lowered an age-bias score from 0.25 to 0.10. Debating and Committee topologies both substantially reduce bias, with Committee offering a better balance of effectiveness and inference cost.
- CBM selects and organizes multiple LLMs to reduce bias.
- Committee cut an age-bias score from 0.25 to 0.10.
- Committee balances mitigation effectiveness against inference cost.
- Single-model self-debiasing is treated as insufficient for deep stereotypes.
From the original article
1 min read at huggingface.co
Large language models (LLMs) are increasingly deployed in public health, finance, and governance, requiring both accuracy and societal value alignment. Despite recent advances, LLMs often perpetuate or amplify bias embedded in their training data, posing challenges to fairness.
By Mingzhe Du, Luu Anh Tuan, Xiaobao Wu · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.