Training Discrete Deep Generative Models via Gapped Straight-Through Estimator.
Ting-Han FanTa-Chung ChiAlexander I. RudnickyPeter J. RamadgePublished in: CoRR (2022)
Keyphrases
- generative model
- deep belief networks
- probabilistic model
- discriminatively trained
- mixture model
- discriminative learning
- discriminative models
- em algorithm
- prior knowledge
- discriminative methods
- hierarchical hidden markov models
- unsupervised learning
- conditional random fields
- semi supervised
- maximum entropy principle
- deep learning
- training set
- hidden variables
- object categories
- maximum likelihood
- similarity measure
- bag of words
- hierarchical models
- representational power
- expectation maximization
- object detection