An initialization that maximizes a semi-orthogonal weight matrix's alignment with the all-ones vector prevents dying ReLU and keeps 100-layer ReLU networks trainable.
Resurrecting the sigmoid in deep learning through dynamical isometry: theory and practice,
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.LG 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
support 1representative citing papers
citing papers explorer
-
Optimized Weight Initialization on the Stiefel Manifold for Deep ReLU Neural Networks
An initialization that maximizes a semi-orthogonal weight matrix's alignment with the all-ones vector prevents dying ReLU and keeps 100-layer ReLU networks trainable.