Replicating models and routing their local losses via permutations from a mixing kernel Q enables structured message sharing that improves generalization.
Dropout: a simple way to prevent neural networks from overfitting.The journal of machine learning research, 15(1):1929–1958
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Improving Generalization by Permutation Routing Across Model Copies
Replicating models and routing their local losses via permutations from a mixing kernel Q enables structured message sharing that improves generalization.