不可靠建议下的学习增强型在线分配:鲁棒性、曝光公平性与分布偏移
文章背景与核心概要
本文研究了旨在利用预测结果来提升在线决策能力的学习增强型算法。由于不可靠的外部建议可能会对系统的运行效率和公平性造成负面影响,作者针对具有有限候选集、不可逆决策以及曝光约束的在线分配问题,提出了一种兼具鲁棒性与公平性的决策规则。该方法将预测建议与保守的后备机制以及公平性修正进行了有效结合。
在有界误差假设下,本文从理论上证明了该算法的一致性与鲁棒性,表明其性能损失与预测误差保持正比例关系。此外,实验结果表明,该方法在面对对抗性建议时表现出极强的稳定性,并能够显著降低曝光差距,为算法在现实不确定环境中的安全应用提供了坚实支撑。
摘要 (Summary)
This paper investigates learning-augmented algorithms designed to improve online decision-making using predictions. Because unreliable advice can negatively impact efficiency and fairness, the author proposes a robust and fair rule for online allocation problems featuring finite candidate sets, irreversible decisions, and exposure constraints. This method effectively combines predictive advice with a conservative fallback mechanism and a fairness correction. Under bounded-error assumptions, the paper proves both consistency and robustness, showing that performance loss remains proportional to prediction error. Furthermore, experimental results demonstrate strong stability against adversarial advice and significant reductions in exposure disparity.
本文研究了旨在利用预测改善在线决策的学习增强型算法。由于不可靠的建议会损害效率和公平性,作者针对具有有限候选集、不可逆决策和曝光约束的在线分配问题,提出了一种鲁棒且公平的规则。该方法将预测建议与保守的后备机制及公平性修正相结合。在有界误差假设下,本文证明了其一致性和鲁棒性,并证明性能损失与预测误差成正比。此外,实验结果表明,该方法在对抗性建议下具有很强的稳定性,并能显著减少曝光差距。
文档元数据 (Document Metadata)
Metadata Field Details Title Learning-Augmented Online Allocation under Unreliable Advice: Robustness, Exposure Fairness, and Distribution Shift Authors Fredy Pokou (MRE, CRIStAL) Subjects Artificial Intelligence ( cs.AI)arXiv Identifier arXiv:2608.26889 [cs.AI] DOI 10.48550/arXiv.2608.26889 Submission Date August 27, 2026
| 元数据字段 | 详情 |
|---|---|
| 标题 | 不可靠建议下的学习增强型在线分配:鲁棒性、曝光公平性与分布偏移 |
| 作者 | Fredy Pokou (MRE, CRIStAL) |
| 研究领域 | 人工智能 (cs.AI) |
| arXiv 标识符 | arXiv:2608.26889 [cs.AI] |
| DOI | 10.48550/arXiv.2608.26889 |
| 提交日期 | 2026年8月27日 |
摘要原文 (Abstract)
Learning-augmented algorithms improve online decisions using predictions, but unreliable advice may harm efficiency and fairness. We study an online allocation problem with finite candidate sets, irreversible decisions, and exposure constraints. We propose a robust and fair rule combining advice with a conservative fallback and fairness correction. Under bounded-error assumptions, we prove consistency and robustness with loss proportional to prediction error. Experiments show stability under adversarial advice and significant reductions in exposure disparity.
学习增强型算法利用预测来改进在线决策,但不可靠的建议可能会损害效率和公平性。我们研究了一个具有有限候选集、不可逆决策和曝光约束的在线分配问题。我们提出了一种将建议与保守后备和公平性修正相结合的鲁棒且公平的规则。在有界误差假设下,我们证明了一致性和鲁棒性,其损失与预测误差成正比。实验表明,该方法在对抗性建议下具有稳定性,并能显著减少曝光差距。
访问与资源 (Access & Resources)
- Full-Text: View PDF | TeX Source
- Citation Tools: NASA ADS | Google Scholar | Semantic Scholar
- 全文链接: 查看 PDF | TeX 源码
- 引用工具: NASA ADS | Google Scholar | Semantic Scholar