Scalable Gradient-Based Tuning of Continuous Regularization Hyperparameters.
Jelena LuketinaMathias BerglundKlaus GreffTapani RaikoPublished in: CoRR (2015)
Keyphrases
- hyperparameters
- regularization parameter
- prior information
- cross validation
- model selection
- closed form
- bayesian framework
- parameter settings
- bayesian inference
- random sampling
- support vector
- em algorithm
- posterior distribution
- maximum likelihood
- gaussian process
- maximum a posteriori
- gaussian processes
- sample size
- noise level
- incremental learning
- image restoration
- prior knowledge
- incomplete data
- missing values
- grid search
- parameter space
- expectation maximization
- parameter optimization
- active learning
- parameter values
- level set
- high dimensional
- feature selection
- computer vision
- gaussian process regression
- machine learning