📄 中文摘要
MIDAS是一种新颖的方法，通过自注意力机制将静态架构参数替换为动态的输入特定参数，从而现代化了DARTS。为提高鲁棒性，MIDAS分别为激活图的每个空间块计算架构选择，并引入了一种无参数的拓扑感知搜索空间，以建模节点连接性并简化每个节点选择两个输入边的过程。MIDAS在DARTS、NAS-Bench-201和RDARTS搜索空间上的评估结果显示，在DARTS中达到了97.42%的准确率。
MIDAS: 马赛克输入特定可微分架构搜索
出处: MIDAS: Mosaic Input-Specific Differentiable Architecture Search
发布: 2026年2月23日
📄 English Summary
MIDAS: Mosaic Input-Specific Differentiable Architecture Search
MIDAS is a novel approach that modernizes DARTS by replacing static architecture parameters with dynamic, input-specific parameters computed via self-attention. To enhance robustness, MIDAS localizes architecture selection by computing it separately for each spatial patch of the activation map and introduces a parameter-free, topology-aware search space that models node connectivity, simplifying the selection of two incoming edges per node. Evaluations of MIDAS on the DARTS, NAS-Bench-201, and RDARTS search spaces demonstrate a performance of 97.42% accuracy in DARTS.