Using Pre-Training Can Improve Model Robustness and Uncertainty

By Dan Hendrycks, Kimin Lee, Mantas Mazeika. 2019

836 citations recorded by the source corpus.

He et al. (2018) have called into question the utility of pre-training by showing that training from scratch can often yield similar performance to pre-training. We show that although pre-training may not improve performance on traditional classification metrics, it improves model robustness and uncertainty estimates. Through extensive experiments on adversarial examples, label corruption, class imbalance, out-of-distribution detection, and confidence calibration, we demonstrate large gains from pre-training and complementary effects with task-specific methods. We introduce adversarial pre-training and show approximately a 10% absolute improvement over the previous state-of-the-art in adversarial robustness. In some cases, using pre-training without task-specific methods also surpasses the state-of-the-art, highlighting the need for pre-training when evaluating future methods on robustness and uncertainty tasks.

Subfields: Control and Systems Engineering, Representation & Self-Supervised Learning

How researchers see this paper

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

In Control and Systems Engineering

In Representation & Self-Supervised Learning

Citations in the corpus

Pretrained Transformers Improve Out-of-Distribution Robustness (2020, methodology): “Similar to computer vision (Orhan, 2019; Xie et al., 2020; Hendrycks et al., 2019a), pretraining on larger and more diverse datasets can improve robustness.”
The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization (2020, background): “• Pretraining: pretraining on larger and more diverse datasets improves robustness [25, 13].”
Large-Scale Adversarial Training for Vision-and-Language Representation Learning (2020, background): “Due to growing dominance of large-scale pre-training, very recent work has started to explore adversarial training in the pre-training stage [19, 10, 37].”
Making Convolutional Networks Shift-Invariant Again (2019, methodology): “. OUT-OF-DISTRIBUTION ROBUSTNESS We have shown increased stability (to shifts), as well as accuracy. Next, we test the generalization capability the classiﬁer in these two aspects, using datasets fromHendrycks et al. (2019). We test stability to perturbations other than shifts. We then test accuracy”
Unsolved Problems in ML Safety (2021, background): “Researchers could enhance data for adversarial robustness by simulating more data [208], augmenting data [149], repurposing existing real data [30, 80], and extracting more information from available data [82].”
Strategies for Pre-training Graph Neural Networks (2019, background): “ralized pre-training to graph data. Pre-training has the potential to provide an attractive solution to the following two fundamental challenges with learning on graph datasets (Pan & Yang, 2009; Hendrycks et al., 2019): First, task-speciﬁc labeled data can be extremely scarce. This problem is e”
Simple and Principled Uncertainty Estimation with Deterministic Deep Learning via Distance Awareness (2020, background): “, data augmentation or unsupervised pretraining, that are known to also improve a network’s uncertainty performance [35, 36].”
Verified Uncertainty Calibration (2019, methodology): “Prior work in machine learning [7, 9, 15, 16, 17] directly estimates each term in the calibration error from samples (Definition 5.”
Unlabeled Data Improves Adversarial Robustness (2019, methodology): “[18] is based on ImageNet adversarial pretraining and is less directly comparable to ours due to the difference in external data and training method.”
Are Labels Required for Improving Adversarial Robustness? (2019, methodology): “‘*’ indicates it is from [21] using 100 PGD steps with 1000 random restarts, an attack that we have found to be weaker than the MultiTargeted attack.”
Improving Robustness Without Sacrificing Accuracy with Patch Gaussian Augmentation (2019, background): “Another line of work has attempted to increase model robustness performance, either by directly projecting out superﬁcial statistics [48], via architectural improvements [5], pre-training schemes [25], or through the use of data augmentations.”
Provably Robust Deep Learning via Adversarially Trained Smoothed Classifiers (2019, methodology): “Specifically, we pursue two ideas: 1) pre-training similar to [17], and 2) semi-supervised learning as in [5].”
An Empirical Study on Robustness to Spurious Correlations using Pre-trained Language Models (2020, background): “Nevertheless, recent empirical results have suggested that self-supervised pre-training improves robust accuracy, while not using any task-speciﬁc knowledge nor incurring in-distribution accuracy drop (Hendrycks et al., 2019, 2020).”
Post-hoc Concept Bottleneck Models (2022, background): “In addition, these fine-tuned models are more robust to distribution shifts (Hendrycks et al., 2020; Mathis et al., 2021; Hendrycks et al., 2019).”
Uncovering the Limits of Adversarial Training against Norm-Bounded Adversarial Examples (2020, methodology): “Most notably, without additional data, we improve on the results of four methods that used additional data (Hendrycks et al., 2019a; Uesato et al., 2019; Wang et al., 2020; Sehwag et al., 2020).”
Orthogonal Projection Loss (2021, neutral): “11% , [18] by 0 .”