MLDR-KD improves heterogeneous knowledge distillation by aligning decoupled class-wise and sample-wise relations at logit and feature levels, with a multiscale dynamic fusion module.
Distillation- based training for multi-exit architectures
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Multi-Level Decoupled Relational Distillation for Heterogeneous Architectures
MLDR-KD improves heterogeneous knowledge distillation by aligning decoupled class-wise and sample-wise relations at logit and feature levels, with a multiscale dynamic fusion module.