<ul data-eligibleForWebStory="true"><li>Deep learning has revealed scaling laws that predict model performance based on dataset and model sizes.</li><li>Researchers are exploring whether similar scaling laws govern predictive uncertainties in deep learning.</li><li>In identifiable parametric models, scaling laws for uncertainty can be derived by treating model parameters in a Bayesian way.</li><li>Guarantees on uncertainty contraction rates do not hold in over-parameterized models.</li><li>Empirical evidence shows scaling laws for predictive uncertainty with respect to dataset and model sizes.</li><li>Experiments on vision and language tasks confirm scaling laws for predictive uncertainty using Bayesian inference and ensemble methods.</li><li>This research challenges skepticism towards Bayesian approaches in deep learning.</li><li>Having a large amount of data is not sufficient to eliminate epistemic uncertainty.</li></ul>

Scaling Laws for Uncertainty in Deep Learning

Discover more