A training-free framework merges heterogeneous neural networks of different depth and width by segmenting deeper models and elastically zipping neurons, reaching accuracy close to homogeneous merging.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Training-free Heterogeneous Model Merging
A training-free framework merges heterogeneous neural networks of different depth and width by segmenting deeper models and elastically zipping neurons, reaching accuracy close to homogeneous merging.