arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2505.02433cs.LGcs.AI
收藏

FairPO: 为公平多标签学习的鲁棒偏好优化

FairPO: Robust Preference Optimization for Fair Multi-Label Learning

IIT Bombay(印度理工学院班加罗尔)

Soumen Kumar Mondal, Prateek Chanda, Akshit Varmora, Ganesh Ramakrishnan

AI总结:

FairPO通过结合偏好损失和群体鲁棒优化,提升多标签学习中不同标签的公平性,通过针对性改进和平衡目标缓解偏见问题。

摘要:

多标签分类(MLC)常在不同标签间出现性能差异。我们提出FairPO,一种结合基于偏好的损失和群体鲁棒优化的框架,通过针对表现不佳的标签来提高公平性。FairPO将标签划分为一个特权集以进行针对性改进和一个非特权集以维持基础性能。对于特权标签,一种受DPO启发的偏好损失通过纠正真实标签与其混淆对应物之间的排名错误来处理困难示例。一个受约束的目标保持非特权标签的性能,同时一种群体鲁棒偏好优化(GRPO)形式自适应地平衡两个目标以缓解偏见。我们还通过使用对比(CPO)和简单(SimPO)偏好优化的无参考变体展示了FairPO的通用性。

英文摘要:

Multi-label classification (MLC) often suffers from performance disparities across labels. We propose \textbf{FairPO}, a framework combining preference-based loss and group-robust optimization to improve fairness by targeting underperforming labels. FairPO partitions labels into a \textit{privileged} set for targeted improvement and a \textit{non-privileged} set to maintain baseline performance. For privileged labels, a DPO-inspired preference loss addresses hard examples by correcting ranking errors between true labels and their confusing counterparts. A constrained objective maintains performance for non-privileged labels, while a Group Robust Preference Optimization (GRPO) formulation adaptively balances both objectives to mitigate bias. We also demonstrate FairPO's versatility with reference-free variants using Contrastive (CPO) and Simple (SimPO) Preference Optimization.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑