📄 中文摘要
AI 驱动的无线接入网（AI-RANs）旨在为异构用户提供共享边缘资源下的时变学习任务。为了确保用户之间的推理性能公平性，需要适应性和公正的学习机制。提出了一种在线内嵌在线的公平多任务学习框架（OWO-FMTL），以确保用户之间的长期公平性。该方法结合了两个学习循环：外循环在每轮中更新共享模型，内循环在每轮内通过轻量级的原始-对偶更新重新平衡用户优先级。公平性通过广义α-公平性进行量化，允许效率与公平之间的权衡。该框架保证了性能差距的逐渐减小。
公平的多任务学习在 AI 无线接入网中的应用
出处: Equitable Multi-Task Learning for AI-RANs
发布: 2026年3月11日
📄 English Summary
Equitable Multi-Task Learning for AI-RANs
AI-enabled Radio Access Networks (AI-RANs) aim to serve heterogeneous users with time-varying learning tasks over shared edge resources. Ensuring equitable inference performance across these users necessitates adaptive and fair learning mechanisms. An online-within-online fair multi-task learning (OWO-FMTL) framework is proposed to ensure long-term equity among users. This method integrates two learning loops: an outer loop that updates the shared model across rounds and an inner loop that rebalances user priorities within each round through a lightweight primal-dual update. Equity is quantified using generalized alpha-fairness, allowing for a trade-off between efficiency and fairness. The framework guarantees a diminishing performance disparity.