[C13/W] Enhancing Multiple Reliability Measures via Nuisance-extended Information Bottleneck

  Code   Talk   Slides   Poster

tl;dr: Modeling nuisance information properly can improve out-of-distribution generalization.

• A preliminary version appeared at ECCV OOD-CV Workshop 2022

Additional information

Abstract

In practical scenarios where training data is limited, many predictive signals in the data can be rather from some biases in data acquisition (i.e., less generalizable), so that one cannot prevent a model from co-adapting on such (so-called) "shortcut" signals: this makes the model fragile in various distribution shifts. To bypass such failure modes, we consider an adversarial threat model under a mutual information constraint to cover a wider class of perturbations in training. This motivates us to extend the standard information bottleneck to additionally model the nuisance information. We propose an autoencoder-based training to implement the objective, as well as practical encoder designs to facilitate the proposed hybrid discriminative-generative training concerning both convolutional- and Transformer-based architectures. Our experimental results show that the proposed scheme improves robustness of learned representations (remarkably without using any domain-specific knowledge), with respect to multiple challenging reliability measures including novelty detection, corruption (or natural) robustness and certified adversarial robustness. For example, our model could advance the state-of-the-art on a recent challenging OBJECTS benchmark in novelty detection by 78.4% → 87.2% in AUROC, while simultaneously enjoying improved corruption and certified robustness.

BibTeX

@InProceedings{Jeong_2023_CVPR,
    author    = {Jeong, Jongheon and Yu, Sihyun and Lee, Hankook and Shin, Jinwoo},
    title     = {Enhancing Multiple Reliability Measures via Nuisance-Extended Information Bottleneck},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
    month     = {June},
    year      = {2023},
    pages     = {16206-16218}
}