Closest decision-boundary (or 'flip') points are computed for a CIFAR-10 network, showing Taylor approximations underestimate distances and that robustness measures should use these exact points.
With friends like these, who needs adversaries? In Advances in Neural Information Processing Systems (NeurIPS 2018), pages 10749--10759, 2018
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2019 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Investigating Decision Boundaries of Trained Neural Networks
Closest decision-boundary (or 'flip') points are computed for a CIFAR-10 network, showing Taylor approximations underestimate distances and that robustness measures should use these exact points.