MIC: Maximizing Informational Capacity in Adaptive Representations via Isotropic Subspace Alignment
About
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 Regularization (SCR) to mitigate redundancy between prefix and residual subspaces via cross-correlation penalties, alongside Spectral Isotropy Regularization (SIR) to ensure hyper-spherical uniformity in low-dimensional prefixes. By unifying these strategies through a self-distillation objective, MIC generates semantically dense representations that maintain high discriminative power. Our experiments demonstrate that MIC significantly outperforms standard baselines, particularly in high-compression scenarios where maintaining informational capacity is most critical.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Word Sense Disambiguation | WiC | Avg Accuracy64.92 | 261 | |
| Intent Classification | Banking77 | Accuracy894.8 | 260 | |
| Semantic Textual Similarity | STSB | Spearman Correlation73.2 | 112 | |
| Text Classification | TweetEval | Accuracy72.69 | 112 | |
| Semantic Textual Similarity | STS-12 | Spearman Correlation (rho)0.6754 | 91 | |
| Natural Language Inference | SICK | Accuracy70.84 | 85 | |
| Paraphrase Detection | MRPC | Accuracy73.96 | 70 | |
| Semantic Textual Similarity | STS 2014 | Spearman Correlation66.93 | 39 | |
| Natural Language Inference | SciTail | Accuracy74.5 | 26 | |
| Text Classification | Emotion | Accuracy61.31 | 22 |