English(EN) Is Fairness Truly Fair? Towards Reliable Lipschitz Fairness in Multi-Task Learning via Fixed-\texorpdfstring{$δ$}{delta} Alignment

新框架ReLiF改进了多任务学习中的公平性评估

作者 PulseAugur 编辑部 · [2 个来源] · 2026-06-09 09:36

研究人员开发了一个名为ReLiF的新框架，以解决多任务学习（MTL）中Lipschitz公平性评估的问题。该框架引入了固定delta审计，它使用共享的参考容差来跨不同算法进行一致的比较。在临床和密集预测基准上的实验表明，ReLiF可以揭示可能被依赖于方法的阈值所掩盖的效用-公平性权衡。 AI

影响引入了一种更可靠的评估AI模型公平性的方法，有望带来更公平的AI系统。

排序理由学术论文，详细介绍了新框架和实验结果。

AI 生成摘要 · Google Gemini · 来自 2 个来源。我们如何撰写摘要 →

报道来源 [2]

arXiv cs.AI TIER_1 English(EN) · Junbo Ding, Xin Zang, Chenchen Pan, Donghao Song, Jiaxin Zhu, Danhuai Guo · 2026-06-10 04:00

Is Fairness Truly Fair? Towards Reliable Lipschitz Fairness in Multi-Task Learning via Fixed-\texorpdfstring{$\delta$}{delta} Alignment

arXiv:2606.10632v1 Announce Type: cross Abstract: Lipschitz-style individual fairness formalizes the idea that semantically similar examples should receive similar predictions, but its evaluation in multi-task learning (MTL) can be confounded by method-induced representation scal…
arXiv cs.AI TIER_1 English(EN) · Danhuai Guo · 2026-06-09 09:36

公平真的公平吗？通过固定δ对齐实现多任务学习中可靠的Lipschitz公平性

Lipschitz-style individual fairness formalizes the idea that semantically similar examples should receive similar predictions, but its evaluation in multi-task learning (MTL) can be confounded by method-induced representation scales. This paper identifies threshold confounding: w…