Low-rank weight factorization creates singular posteriors in Bayesian neural networks that scale as sqrt(r(m+n)) in complexity and use up to 33x fewer parameters than ensembles.
When din ≫d out and C2 0 C4 1 R2 is large, PAC-Bayes can still be favorable despite thedin term, because it avoids the spectral norm constants entirely
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ML 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Singular Bayesian Neural Networks
Low-rank weight factorization creates singular posteriors in Bayesian neural networks that scale as sqrt(r(m+n)) in complexity and use up to 33x fewer parameters than ensembles.