Addressing Failure Prediction by Learning Model Confidence

By Charles Corbière, Nicolas Thomé, Avner Bar-Hen, Matthieu Cord, Patrick Pérez. Neural Information Processing Systems, 2019

369 citations recorded by the source corpus.

Assessing reliably the confidence of a deep neural net and predicting its failures is of primary importance for the practical deployment of these models. In this paper, we propose a new target criterion for model confidence, corresponding to the True Class Probability (TCP). We show how using the TCP is more suited than relying on the classic Maximum Class Probability (MCP). We provide in addition theoretical guarantees for TCP in the context of failure prediction. Since the true class is by essence unknown at test time, we propose to learn TCP criterion on the training set, introducing a specific learning scheme adapted to this context. Extensive experiments are conducted for validating the relevance of the proposed approach. We study various network architectures, small and large scale datasets for image classification and semantic segmentation. We show that our approach consistently outperforms several strong methods, from MCP to Bayesian uncertainty, as well as recent approaches specifically designed for failure prediction.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/5a05dd1e5e2859ecb845ee75e4ede22805fb22a6

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Solving Long-tailed Recognition with Deep Realistic Taxonomic Classifier (2020, methodology): “The confidence score s(·) can be implemented in various ways [11,25,57].”
A taxonomy, data set, and benchmark for detecting and classifying malevolent dialogue responses (2021, methodology): “The true class probability (TCP) confidence is estimated using a learning-based method; the original TCP method is designed for image classification (Corbière et al., 2019).”
Disentangling Physical Dynamics From Unknown Factors for Unsupervised Video Prediction (2020, background): “RungeKutta [15], and extension to probabilistic forecasts with uncertainty estimation [18, 9], e.”
Generating with Confidence: Uncertainty Quantification for Black-box Large Language Models (2023, background): “Specifically, the confidence scores are generally associated with both the input and the prediction and can be expressed as C(x, y) [4, 6, 17, 28].”
A Review of Uncertainty Quantification in Deep Learning: Techniques, Applications and Challenges (2020, methodology): “A general view of the proposed model in [164] is illustrated by Fig.”