MIC: Maximizing Informational Capacity in Adaptive Representations via Isotropic Subspace Alignment
Nguyen Hong Dang, Nhi Ngoc-Yen Nguyen, Huy-Hieu Pham · 2026
Read MIC: Maximizing Informational Capacity in Adaptive Representations via Isotropic Subspace Alignment by Nguyen Hong Dang, Nhi Ngoc-Yen Nguyen, Huy-Hieu Pham (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-29987.txt.
Description
Dang Nguyen Hong, Nhi Ngoc-Yen Nguyen, and Huy-Hieu Pham fix a flaw in elastic-dimension embeddings, where nested subspaces waste dims and collapse spectrally. Their MIC frame reshapes mixed-grain embedding shapes through even subspace lining up. Soft collapse costs cut overlap between prefix and residual subspaces, while spectral evenness costs spread small prefixes evenly over a hypersphere. A self-teaching goal joins both costs, yielding dense, sharp codes. Tests beat standard rivals, above all where squeezing runs high and kept knowledge matters most.
Catalog record
| CREATOR | Nguyen Hong Dang, Nhi Ngoc-Yen Nguyen, Huy-Hieu Pham |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 43739 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.29987