A Dimensional Structure based Knowledge Distillation Method for Cross-Modal Learning

Lingyu Si

doi:10.1016/j.neunet.2026.108672

A Dimensional Structure based Knowledge Distillation Method for Cross-Modal Learning

2026-02-03

0

PRE

AI

DOI：10.1016/j.neunet.2026.108672

Original

Original request for help

Save

Abstract

En 中文

• Discovered the correlation between feature discriminability and dimensional structure (DS), revealing superior DS in high-quality modality features. • Proposed a DS-based knowledge distillation method, enhancing low-quality modality performance by optimizing feature distribution and independence. • Introduced IIS-ISCAS, a new cross-modal learning (CML) dataset for marine target recognition, with experiments validating the method across several datasets.

A Dimensional Structure based Knowledge Distillation Method for Cross-Modal Learning

Abstract

Journal

Organization