Bias-Reduced Uncertainty Estimation for Deep Neural Classifiers
This addresses a specific issue in uncertainty estimation for deep learning, offering an incremental improvement for practitioners in fields like safety-critical applications.
The paper tackles the problem of biased uncertainty estimates in deep neural classifiers, particularly for highly confident predictions, and proposes an algorithm that uses earlier model snapshots to reduce this bias, achieving consistently better uncertainty estimates than existing methods.
We consider the problem of uncertainty estimation in the context of (non-Bayesian) deep neural classification. In this context, all known methods are based on extracting uncertainty signals from a trained network optimized to solve the classification problem at hand. We demonstrate that such techniques tend to introduce biased estimates for instances whose predictions are supposed to be highly confident. We argue that this deficiency is an artifact of the dynamics of training with SGD-like optimizers, and it has some properties similar to overfitting. Based on this observation, we develop an uncertainty estimation algorithm that selectively estimates the uncertainty of highly confident points, using earlier snapshots of the trained model, before their estimates are jittered (and way before they are ready for actual classification). We present extensive experiments indicating that the proposed algorithm provides uncertainty estimates that are consistently better than all known methods.