Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

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.

Dang Nguyen Hong, Nhi Ngoc-Yen Nguyen, Huy-Hieu Pham• 2026

Related benchmarks

TaskDatasetResultRank
Word Sense DisambiguationWiC
Avg Accuracy64.92
261
Intent ClassificationBanking77
Accuracy894.8
260
Semantic Textual SimilaritySTSB
Spearman Correlation73.2
112
Text ClassificationTweetEval
Accuracy72.69
112
Semantic Textual SimilaritySTS-12
Spearman Correlation (rho)0.6754
91
Natural Language InferenceSICK
Accuracy70.84
85
Paraphrase DetectionMRPC
Accuracy73.96
70
Semantic Textual SimilaritySTS 2014
Spearman Correlation66.93
39
Natural Language InferenceSciTail
Accuracy74.5
26
Text ClassificationEmotion
Accuracy61.31
22
Showing 10 of 15 rows

Other info

Follow for update