Alternating one-sided saturating activations, or non-positive weights with ReLU, make four-layer monotonic MLPs universal approximators, and a new sign-switch layer avoids constrained weights in practice.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Advancing Constrained Monotonic Neural Networks: Achieving Universal Approximation Beyond Bounded Activations
Alternating one-sided saturating activations, or non-positive weights with ReLU, make four-layer monotonic MLPs universal approximators, and a new sign-switch layer avoids constrained weights in practice.