Decomposing Representations for Deterministic Uncertainty Estimation
This work addresses a key challenge in uncertainty estimation for deployed machine learning systems, offering a solution to enhance reliability in out-of-distribution detection.
The paper tackled the problem of inconsistent performance in out-of-distribution detection using feature density-based uncertainty estimators by proposing a method to decompose learned representations and integrate uncertainties separately, resulting in greatly improved performance and interpretability.
Uncertainty estimation is a key component in any deployed machine learning system. One way to evaluate uncertainty estimation is using "out-of-distribution" (OoD) detection, that is, distinguishing between the training data distribution and an unseen different data distribution using uncertainty. In this work, we show that current feature density based uncertainty estimators cannot perform well consistently across different OoD detection settings. To solve this, we propose to decompose the learned representations and integrate the uncertainties estimated on them separately. Through experiments, we demonstrate that we can greatly improve the performance and the interpretability of the uncertainty estimation.