Identifying Incorrect Classifications with Balanced Uncertainty

Abstract

Uncertainty estimation is critical for costsensitive deep-learning applications (i.e. disease diagnosis).It is very challenging partly due to the inaccessibility of uncertainty groundtruth in most datasets. Previous works proposed to estimate the uncertainty from softmax calibration, Monte Carlo sampling, subjective logic and so on. However, these existing methods tend to be overconfident about their predictions with unreasonably low overall uncertainty, which originates from the imbalance between positive (correct classifications) and negative (incorrect classifications) samples. For this issue, we firstly propose the distributional imbalance to model the imbalance in uncertainty estimation as two kinds of distribution biases, and secondly propose Balanced True Class Probability (BT CP ) framework, which learns an uncertainty estimator with a novel Distributional Focal Loss (DF L) objective. Finally, we evaluate the BT CP in terms of failure prediction and out-of-distribution (OOD) detection on multiple datasets. The experimental results show that BT CP outperforms other uncertainty estimation methods especially in identifying incorrect classifications.

Improving the Reliability for Confidence Estimation

“…Confidence Estimation. Being an important task that helps determine whether a deep predictor's predictions can be trusted, confidence estimation has been studied extensively across various computer vision tasks [14,11,19,5,34,36,32,44,4,35,26,43,47]. At the beginning, Hendrycks and Gimpel [14] proposed Maximum Class Probability utilizing the classifier softmax distribution, Gal and Ghahramani [11] proposed MCDropout from the perspective of uncertainty estimation, and Jiang et al [19] proposed Trust Score to calculate the agreement between the classifier and a modified nearest-neighbor classifier in the testing set.…”

“…correct and incorrect task model predictions from each other. Afterwards, Li et al [26] proposed an extension to True Class Probability [5] that uses a Distributional Focal Loss to focus more on predictions with higher uncertainty. Unlike previous methods that design strategies to handle a specific imbalanced distribution of correct and incorrect labels, we adopt a novel perspective, and tackle the label imbalance problem through meta-learning, which allows our confidence estimator to learn distribution-generalizable knowledge to tackle a variety of diverse label distributions.…”

“…1. Many previous works [5,32,26,47] have proposed to train such a confidence estimator to conduct confidence estimation more reliably. Some recent works [32,26] have noted that a reliable confidence estimator should perform well under label imbalance.…”

“…Many previous works [5,32,26,47] have proposed to train such a confidence estimator to conduct confidence estimation more reliably. Some recent works [32,26] have noted that a reliable confidence estimator should perform well under label imbalance. This is because confidence estimators use the correctness of task model predictions (C) as labels, which are often imbalanced.…”

“…Thus, since many task models have achieved good performance on computer vision tasks (e.g., Small ConvNet [19] achieves > 99% for MNIST [24] and VGG-16 [41,31] achieves > 93% for CIFAR-10 [21] in image classification), there are often many more correct predictions (where C = 1) than incorrect ones (where C = 0), which leads to label imbalance for the confidence estimation task. If this label imbalance is not accounted for during training, the confidence estimator is likely to be overly confident [32,26] for incorrect predictions (where C = 0), which is undesirable.…”