Learning Hyperparameters via a Data-Emphasized Variational Objective

Ethan Harvey, Mikhail Petrov, Michael C Hughes
Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, PMLR 300:2143-2151, 2026.

Abstract

When training large models on limited data, avoiding overfitting is paramount. Common grid search or smarter search methods rely on expensive separate runs for each candidate hyperparameter, while carving out a validation set that reduces available training data. In this paper, we study gradient-based learning of hyperparameters via the evidence lower bound (ELBO) objective from Bayesian variational methods. This avoids the need for any validation set. We focus on scenarios where the model is over-parameterized for flexibility and the approximate posterior is chosen to be Gaussian with isotropic covariance for tractability, even though it cannot match the true posterior. In such scenarios, we find the ELBO prioritizes posteriors that match the prior, leading to severe underfitting. Instead, we recommend a data-emphasized ELBO that upweights the likelihood but not the prior. In Bayesian transfer learning of image and text classifiers, our method reduces the 88+ hour grid search of past work to under 3 hours while delivering comparable accuracy. We further demonstrate how our approach enables efficient yet accurate approximations of Gaussian processes with learnable lengthscale kernels.

Cite this Paper

BibTeX
@InProceedings{pmlr-v300-harvey26a, title = { Learning Hyperparameters via a Data-Emphasized Variational Objective }, author = {Harvey, Ethan and Petrov, Mikhail and Hughes, Michael C}, booktitle = {Proceedings of The 29th International Conference on Artificial Intelligence and Statistics}, pages = {2143--2151}, year = {2026}, editor = {Khan, Emtiyaz and Li, Yingzhen and Solin, Arno and Ramdas, Aaditya}, volume = {300}, series = {Proceedings of Machine Learning Research}, month = {02--05 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v300/main/assets/harvey26a/harvey26a.pdf}, url = {https://proceedings.mlr.press/v300/harvey26a.html}, abstract = { When training large models on limited data, avoiding overfitting is paramount. Common grid search or smarter search methods rely on expensive separate runs for each candidate hyperparameter, while carving out a validation set that reduces available training data. In this paper, we study gradient-based learning of hyperparameters via the evidence lower bound (ELBO) objective from Bayesian variational methods. This avoids the need for any validation set. We focus on scenarios where the model is over-parameterized for flexibility and the approximate posterior is chosen to be Gaussian with isotropic covariance for tractability, even though it cannot match the true posterior. In such scenarios, we find the ELBO prioritizes posteriors that match the prior, leading to severe underfitting. Instead, we recommend a data-emphasized ELBO that upweights the likelihood but not the prior. In Bayesian transfer learning of image and text classifiers, our method reduces the 88+ hour grid search of past work to under 3 hours while delivering comparable accuracy. We further demonstrate how our approach enables efficient yet accurate approximations of Gaussian processes with learnable lengthscale kernels. } }
Endnote
%0 Conference Paper %T Learning Hyperparameters via a Data-Emphasized Variational Objective %A Ethan Harvey %A Mikhail Petrov %A Michael C Hughes %B Proceedings of The 29th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2026 %E Emtiyaz Khan %E Yingzhen Li %E Arno Solin %E Aaditya Ramdas %F pmlr-v300-harvey26a %I PMLR %P 2143--2151 %U https://proceedings.mlr.press/v300/harvey26a.html %V 300 %X When training large models on limited data, avoiding overfitting is paramount. Common grid search or smarter search methods rely on expensive separate runs for each candidate hyperparameter, while carving out a validation set that reduces available training data. In this paper, we study gradient-based learning of hyperparameters via the evidence lower bound (ELBO) objective from Bayesian variational methods. This avoids the need for any validation set. We focus on scenarios where the model is over-parameterized for flexibility and the approximate posterior is chosen to be Gaussian with isotropic covariance for tractability, even though it cannot match the true posterior. In such scenarios, we find the ELBO prioritizes posteriors that match the prior, leading to severe underfitting. Instead, we recommend a data-emphasized ELBO that upweights the likelihood but not the prior. In Bayesian transfer learning of image and text classifiers, our method reduces the 88+ hour grid search of past work to under 3 hours while delivering comparable accuracy. We further demonstrate how our approach enables efficient yet accurate approximations of Gaussian processes with learnable lengthscale kernels.
APA
Harvey, E., Petrov, M. & Hughes, M.C.. (2026). Learning Hyperparameters via a Data-Emphasized Variational Objective . Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 300:2143-2151 Available from https://proceedings.mlr.press/v300/harvey26a.html.

Related Material