Citation Context

...nce of hyperparameters that are tested. To address these limitations, there has been an increasing amount of work on automatic hyperparameter optimization [Bergstra et al., 2011, Hutter et al., 2011, =-=Snoek et al., 2012-=-, Thornton et al., 2013]. Most rely on an approach called sequential model based optimization (SMBO). The idea consists in treating RS(hγ) def= f(γ) as a learnable function of γ, which we can learn fr...

Citation Context

... results of the sequence of hyperparameters that are tested. To address these limitations, there has been an increasing amount of work on automatic hyperparameter optimization [Bergstra et al., 2011, =-=Hutter et al., 2011-=-, Snoek et al., 2012, Thornton et al., 2013]. Most rely on an approach called sequential model based optimization (SMBO). The idea consists in treating RS(hγ) def= f(γ) as a learnable function of γ, w...

Citation Context

...rs that are tested. To address these limitations, there has been an increasing amount of work on automatic hyperparameter optimization [Bergstra et al., 2011, Hutter et al., 2011, Snoek et al., 2012, =-=Thornton et al., 2013-=-]. Most rely on an approach called sequential model based optimization (SMBO). The idea consists in treating RS(hγ) def= f(γ) as a learnable function of γ, which we can learn from the observations {(γ...

Citation Context

...st importantly, the normal assumption of Section 3.3 in Lacoste et al. [2014] suggests that methods based on the covariance of the predictions such as ensemble pruning [Zhang et al., 2006] and MinCq [=-=Roy et al., 2011-=-] are simply different approximations of this idea. This connection allows us to be confident that the fast and simple algorithm we propose in this paper is at least equivalent in generalization perfo...

Citation Context

...ormance is often obtained by, instead, combining several (good) models into an ensemble. This was best illustrated by the winning entry of the Netflix competition, which combined a variety of models [=-=Bell et al., 2007-=-]. Even if one concentrates on a single learning algorithm, combining models produced by using different hyperparameters is also helpful. Intuitively, models with comparable performances are still lik...

Citation Context

...pose a method for leveraging the recent research on SMBO in order to generate an ensemble of models, as opposed to the single best model. The proposed approach builds on the Agnostic Bayes framework [=-=Lacoste et al., 2014-=-], which provides a successful strategy for weighting a predetermined and finite set of models (already trained) into an ensemble. Using a successful SMBO method, we show how we can effectively genera...