A gradient-based bilevel optimization approach for tuning regularization hyperparameters.
Ankur SinhaTanmay KhandaitRaja MohantyPublished in: Optim. Lett. (2024)
Keyphrases
- hyperparameters
- regularization parameter
- prior information
- model selection
- cross validation
- grid search
- parameter optimization
- closed form
- bayesian inference
- bayesian framework
- support vector
- bilevel programming
- em algorithm
- gaussian process
- maximum a posteriori
- parameter settings
- random sampling
- incremental learning
- gaussian processes
- prior knowledge
- noise level
- maximum likelihood
- image restoration
- convex programming
- linear programming
- multiscale
- generalization ability
- missing values
- incomplete data
- sample size
- upper bound
- parameter space
- np hard
- support vector machine
- error rate
- generative model