Maximizing a lower bound and minimizing an upper bound of the EBM negative log-likelihood stabilizes training and improves density estimation and sample quality.
Bounds all around: training energy-based models with bidirectional bounds.Advances in Neural Information Processing Systems, 34:19808– 19821, 2021
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
extension 1
citation-polarity summary
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1roles
extension 1polarities
extend 1representative citing papers
citing papers explorer
-
Exploring bidirectional bounds for minimax-training of Energy-based models
Maximizing a lower bound and minimizing an upper bound of the EBM negative log-likelihood stabilizes training and improves density estimation and sample quality.