Towards Trustworthy Predictions from Deep Neural Networks with Fast Adversarial Calibration

TL;DR

This paper introduces a new training method that combines entropy and adversarial calibration losses to produce neural network predictions that remain well-calibrated and trustworthy even when faced with domain shifts, enhancing AI reliability.

Contribution

The paper presents a novel, efficient training strategy that improves the calibration of neural network predictions under domain shift, outperforming existing methods.

Findings

Outperforms state-of-the-art calibration methods across multiple datasets.

Produces well-calibrated predictions under various domain drifts.

Effective across different data modalities and neural network architectures.

Abstract

To facilitate a wide-spread acceptance of AI systems guiding decision making in real-world applications, trustworthiness of deployed models is key. That is, it is crucial for predictive models to be uncertainty-aware and yield well-calibrated (and thus trustworthy) predictions for both in-domain samples as well as under domain shift. Recent efforts to account for predictive uncertainty include post-processing steps for trained neural networks, Bayesian neural networks as well as alternative non-Bayesian approaches such as ensemble approaches and evidential deep learning. Here, we propose an efficient yet general modelling approach for obtaining well-calibrated, trustworthy probabilities for samples obtained after a domain shift. We introduce a new training strategy combining an entropy-encouraging loss term with an adversarial calibration loss term and demonstrate that this results in well-calibrated and technically trustworthy predictions for a wide range of domain drifts. We comprehensively evaluate previously proposed approaches on different data modalities, a large range of data sets including sequence data, network architectures and perturbation strategies. We observe that our modelling approach substantially outperforms existing state-of-the-art approaches, yielding well-calibrated predictions under domain drift.