Active Learning of Linear Embeddings for Gaussian Processes
By Roman Garnett, Michael A. Osborne, Philipp Hennig. 2013
104 citations recorded by the source corpus.
We propose an active learning method for discovering low-dimensional structure in high-dimensional Gaussian process (GP) tasks. Such problems are increasingly frequent and important, but have hitherto presented severe practical difficulties. We further introduce a novel technique for approximately marginalizing GP hyperparameters, yielding marginal predictions robust to hyperparameter mis-specification. Our method offers an efficient means of performing GP regression, quadrature, or Bayesian optimization in high-dimensional spaces.
Subfields: Artificial Intelligence
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Frank Hutter
- José Miguel Hernández-Lobato
- Zoubin Ghahramani
- Sergey Levine
- Ryan P. Adams
- Nando de Freitas
- Ziyun Wang
- Quoc V. Le
Citations in the corpus
- Taking the Human Out of the Loop: A Review of Bayesian Optimization (2016, background): “We could do away with samples entirely and approximately integrate out the hyperparameters as shown in [53].”
- Control functionals for Monte Carlo integration (2014, methodology): “this correlated sample setting (Sun and Wu, 2010). Moreover, state-of-the-art techniques for high-dimensional functional approximation should be applicable in the context of control functionals (e.g. Garnett et al., 2014). Applications: Our methods were motivated by complex models dened by un-normal”
- Bayesian Optimization for Synthetic Gene Design (2015, methodology): “This will require, however, the development of scalable Bayesian optimization methods able to work well in high dimensions in the line of some recent works [17, 5, 2, 7].”
- Sample-Efficient Optimization in the Latent Space of Deep Generative Models via Weighted Retraining (2020, methodology): “Early formulations of LSO were motivated by scaling Gaussian processes to high dimensional problems with simple linear manifolds, using either random projections [60] or a learned transformation matrix [12].”
- Scalable Global Optimization via Local Bayesian Optimization (2019, unknown): “Other methods exist that rely on a mapping between the high-dimensional space and an unknown low-dimensional subspace to scale to large numbers of observations [51, 31, 17].”