A Survey on Uncertainty Quantification Methods for Deep Neural Networks: An Uncertainty Source Perspective

Abstract

Deep neural networks (DNNs) have achieved tremendous success in making accurate predictions for computer vision, natural language processing, as well as science and engineering domains. However, it is also well-recognized that DNNs sometimes make unexpected, incorrect, but overconfident predictions. This can cause serious consequences in high-stake applications, such as autonomous driving, medical diagnosis, and disaster response. Uncertainty quantification (UQ) aims to estimate the confidence of DNN predictions beyond prediction accuracy. In recent years, many UQ methods have been developed for DNNs. It is of great practical value to systematically categorize these UQ methods and compare their advantages and disadvantages. However, existing surveys mostly focus on categorizing UQ methodologies from a neural network architecture's perspective or a Bayesian perspective and ignore the source of uncertainty that each methodology can incorporate, making it difficult to select an appropriate UQ method in practice. To fill the gap, this paper presents a systematic taxonomy of UQ methods for DNNs based on the types of uncertainty sources (data uncertainty versus model uncertainty). We summarize the advantages and disadvantages of methods in each category. We show how our taxonomy of UQ methodologies can potentially help guide the choice of UQ method in different machine learning problems (e.g., active learning, robustness, and reinforcement learning). We also identify current research gaps and propose several future research directions.

Multi-scale image recognition strategy based on convolutional neural network

“…For example, RoI Pooling [12] and RoI Align [13], which are widely used in target detection, can carry out dynamic pooling according to the specific size of the target. In addition, pyramid pooling [14][15][16][17] (such as SPP-net and ASPP) allows multi-scale features to be extracted at the same level, thus strengthening the model's ability to identify objects at different scales.…”

Quantifying the Impact of Single-Event Upsets on Neural Networks with Uncertainty Analysis

“…Prediction uncertainty in deep neural networks is generally categorized into aleatory and epistemic uncertainties. The former is due to noise in the data, while the latter arises from imbalances in the distribution of the training data [12] [13]. Uncertainty quantification as a priority for future scientific research in NASA's CFD Vision 2030 [14].…”

Deep Reinforcement Learning for Mobile Robot Path Planning

“…In terms of robots, deep reinforcement learning is also widely used in robot localization [5], perception [6], decision-making [7], planning [8], hardware [9], etc. In the field of exploring intelligent systems, deep reinforcement learning (DRL) has become a powerful tool for achieving complex decision-making and path planning tasks [10]. The core advantage of DRL is its ability to learn optimal policies through interaction with the environment, without the need for pre-defined rules or models [11].…”