Benefits of Overparameterization in Single-Layer Latent Variable Generative Models.
Rares-Darius BuhaiAndrej RisteskiYoni HalpernDavid A. SontagPublished in: CoRR (2019)
Keyphrases
- latent variables
- generative model
- single layer
- probabilistic model
- hidden layer
- topic models
- prior knowledge
- discriminative methods
- hidden variables
- multi layer
- neural nets
- latent structure
- neural network
- latent variable models
- graphical models
- conditional random fields
- gaussian process
- activation function
- discriminative learning
- scalable video coding
- structured prediction
- feed forward neural networks
- latent dirichlet allocation
- random variables
- bayesian networks
- semi supervised
- markov networks
- computer vision
- object categories
- em algorithm
- maximum likelihood
- probability distribution
- pairwise
- data mining