Recursive Learning Without Collapse: A Weighting-Based Stabilization Framework
无崩溃的递归学习：基于加权的稳定化框架
机构 * Wang Yanan Institute for Studies in Economics, Xiamen University（厦门大学王亚南经济研究所） ； Department of Statistics and Data Science, University of California, Los Angeles（加州大学洛杉矶分校统计与数据科学系）
AI总结 针对递归生成模型训练中的模型崩溃问题，提出基于加权的训练策略，在混合真实与合成数据场景下，理论推导出最优加权方案的统一表达式，揭示合成数据利用与模型性能间的权衡。
Comments This article has been accepted for publication in Journal of the Royal Statistical Society: Series B, published by Oxford University Press. The Version of Record is available at https://doi.org/10.1093/jrsssb/qkag099
