SDP-CROWN's deep-network integration is unsound: the per-layer L2 ball is centered at the linear network's preactivation instead of the actual forward preactivation, allowing invalid robustness certificates.
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
REJECT 1representative citing papers
citing papers explorer
-
SDP-CROWN: Efficient Bound Propagation for Neural Network Verification with Tightness of Semidefinite Programming
SDP-CROWN's deep-network integration is unsound: the per-layer L2 ball is centered at the linear network's preactivation instead of the actual forward preactivation, allowing invalid robustness certificates.