1 Hit in 2.7 sec

Constraining Logits by Bounded Function for Adversarial Robustness [article]

Sekitoshi Kanai, Masanori Yamada, Shin'ya Yamaguchi, Hiroshi Takahashi, Yasutoshi Ida
2020 arXiv   pre-print
The addition of our function improves adversarial robustness because it makes logit and pre-logit vectors have small norms.  ...  Following this hypothesis, we analyze norms of logit vectors at the optimal point under the assumption of universal approximation and explore new methods for constraining logits by addition of a bounded  ...  Batch-wise Logit-Similarity: Generalizing Logit-Squeezing and Label-Smoothing. In Proc. BMVC. Summers, C.; and Dinneen, M. J. 2019.  ... 
arXiv:2010.02558v1 fatcat:q5sru4ziojf73b4fho2qfmvh2i