Bayesian Linear Regression on Deep Representations
This work addresses uncertainty modeling in neural networks for regression, particularly in reinforcement learning, but is incremental as it builds on existing Bayesian linear regression approaches.
The paper tackled the limitation of Bayesian linear regression on deep representations to homoscedastic noise by proposing a novel variation for heteroscedastic noise, achieving competitive performance with standard ensembling and outperforming compared methods in benchmarks and model-based reinforcement learning.
A simple approach to obtaining uncertainty-aware neural networks for regression is to do Bayesian linear regression (BLR) on the representation from the last hidden layer. Recent work [Riquelme et al., 2018, Azizzadenesheli et al., 2018] indicates that the method is promising, though it has been limited to homoscedastic noise. In this paper, we propose a novel variation that enables the method to flexibly model heteroscedastic noise. The method is benchmarked against two prominent alternative methods on a set of standard datasets, and finally evaluated as an uncertainty-aware model in model-based reinforcement learning. Our experiments indicate that the method is competitive with standard ensembling, and ensembles of BLR outperforms the methods we compared to.