Convolutional dictionary learning based auto-encoders for natural exponential-family distributions.
Bahareh TolooshamsAndrew H. SongSimona TemereancaDemba E. BaPublished in: ICML (2020)
Keyphrases
- exponential family
- dictionary learning
- sparse coding
- sparse representation
- maximum likelihood
- density estimation
- log likelihood
- graphical models
- closed form
- statistical models
- mixture model
- missing values
- generative model
- natural images
- linear combination
- unsupervised learning
- image classification
- markov chain monte carlo
- order statistics
- variational methods
- feature space
- probability density function
- visual words
- hidden variables
- information theoretic
- random variables
- missing data
- image representation
- image patches
- multiscale
- higher order
- dirichlet process mixture models