Tag
论文提出 Optimal Hydra，一种面向 LLM 微调的模型异构联邦学习框架，通过全局层级选择与基于客户端数据分布的注意力头选择构建客户端专属子网络（Hydra-nets），在三个数据集上显著优于基线方法并降低本地资源需求。
CALM proposes a method for decentralized federated learning that uses class-wise agreement and label-gated disagreement modulation to handle non-IID data, improving teacher weighting and distillation trust without additional communication overhead.
This paper investigates evidence utility in retrieval-augmented generation, finding that reader-specific ordinal preferences are stable but do not reliably predict intervention outcomes across different readers.
This paper presents a multi-agent reasoning framework where multiple foundation models collaborate through structured critique and aggregation, demonstrating that model heterogeneity significantly improves step-wise reasoning accuracy and reduces variance across domains.
This paper proposes HASA, a heterogeneity-aware subnet allocation method for model-heterogeneous federated learning that assigns subnet widths based on client heterogeneity scores under a fixed compute budget, improving mean and worst-client accuracy.