PixelDefend: Leveraging Generative Models to Understand and Defend against Adversarial Examples.
Yang SongTaesup KimSebastian NowozinStefano ErmonNate KushmanPublished in: ICLR (Poster) (2018)
Keyphrases
- generative model
- probabilistic model
- discriminative learning
- mixture model
- hierarchical hidden markov models
- discriminative models
- em algorithm
- prior knowledge
- hidden variables
- object categories
- conditional random fields
- topic models
- training examples
- semi supervised
- maximum entropy principle
- active learning
- generative and discriminative models
- discriminative methods