Model-level mixtures of experts for semantic segmentation are, in most tested settings, more robust to white-box adversarial attacks than fixed ensembles, but the advantage disappears under universal attacks.
Robustness to adversarial examples through an ensemble of specialists,
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.CV 1years
2024 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Towards Adversarial Robustness of Model-Level Mixture-of-Experts Architectures for Semantic Segmentation
Model-level mixtures of experts for semantic segmentation are, in most tested settings, more robust to white-box adversarial attacks than fixed ensembles, but the advantage disappears under universal attacks.