A Simple Approach to Improve Single-Model Deep Uncertainty via Distance-Awareness

Abstract

Accurate uncertainty quantification is a major challenge in deep learning, as neural networks can make overconfident errors and assign high confidence predictions to out-of-distribution (OOD) inputs. The most popular approaches to estimate predictive uncertainty in deep learning are methods that combine predictions from multiple neural networks, such as Bayesian neural networks (BNNs) and deep ensembles. However their practicality in real-time, industrial-scale applications are limited due to the high memory and computational cost. Furthermore, ensembles and BNNs do not necessarily fix all the issues with the underlying member networks. In this work, we study principled approaches to improve uncertainty property of a single network, based on a single, deterministic representation. By formalizing the uncertainty quantification as a minimax learning problem, we first identify distance awareness, i.e., the model's ability to quantify the distance of a testing example from the training data, as a necessary condition for a DNN to achieve highquality (i.e., minimax optimal) uncertainty estimation. We then propose Spectral-normalized Neural Gaussian Process (SNGP), a simple method that improves the distance-awareness ability of modern DNNs with two simple changes: (1) applying spectral normalization to hidden weights to enforce bi-Lipschitz smoothness in representations and (2) replacing the last output layer with a Gaussian process layer. On a suite of vision and language understanding benchmarks and on modern architectures (Wide-ResNet and BERT), SNGP consistently outperforms other single-model approaches in prediction, calibration and out-of-domain detection. Furthermore, SNGP provides complementary benefits to popular techniques such as deep ensembles and data augmentation, making it a simple and scalable building block for probabilistic deep learning. Code is open-sourced at https://github.com/google/uncertainty-baselines.

Assaying Out-Of-Distribution Generalization in Transfer Learning

“…Orthogonally, several (less scalable) works advocated for leveraging the compositional (perhaps causal [44]) structure in the underlying data-generative process to introduce suitable inductive biases [45][46][47][48][49][50][51]. (4) Simultaneously, Bayesian approaches for uncertainty predictions have been proposed to improve model calibration [52][53][54][55][56][57][58][59][60] and robustness on new distributions [61][62][63]. Recent work, however, found that larger models were natively better calibrated [64].…”

Can Land Cover Classification Models Benefit From Distance-Aware Architectures?

“…We refer the interested reader to [24]. Note that the SNGP approach proved to effectively increase the model's predictive uncertainty quality in an OOD detection setting for remote sensing image classification [19].…”