MIC: Maximizing Informational Capacity in Adaptive Representations via Isotropic Subspace Alignment 事件

MIC: Maximizing Informational Capacity in Adaptive Representations via Isotropic Subspace Alignment arXiv:2605.29987v1 Announce Type: cross Abstract: Although multi-scales representation learning enables elastic-dimension embeddings, nested subspaces often suffer from dimensional redundancy and spectral collapse. To address this, we introduce MIC, a framework that optimizes the geometric landscape of multi-granular embeddings through isotropic subspace alignment. MIC employs Soft Collapse Regul