Globally-Robust Neural Networks

By Klas Leino, Zifan Wang, Matt Fredrikson. 2021

156 citations recorded by the source corpus.

The threat of adversarial examples has motivated work on training certifiably robust neural networks to facilitate efficient verification of local robustness at inference time. We formalize a notion of global robustness, which captures the operational properties of on-line local robustness certification while yielding a natural learning objective for robust training. We show that widely-used architectures can be easily adapted to this objective by incorporating efficient global Lipschitz bounds into the network, yielding certifiably-robust models by construction that achieve state-of-the-art verifiable accuracy. Notably, this approach requires significantly less time and memory than recent certifiable training methods, and leads to negligible costs when certifying points on-line; for example, our evaluation shows that it is possible to train a large robust Tiny-Imagenet model in a matter of hours. Our models effectively leverage inexpensive global Lipschitz bounds for real-time certification, despite prior suggestions that tighter local bounds are needed for good performance; we posit this is possible because our models are specifically trained to achieve tighter global bounds. Namely, we prove that the maximum achievable verifiable accuracy for a given dataset is not improved by using a local bound.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

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

1. Huan Zhang
2. Bo Li
3. Jinfeng Yi
4. Percy Liang
5. Cho-Jui Hsieh
6. J. Kolter
7. Anima Anandkumar
8. Bowen Zhou

In Computer Vision and Pattern Recognition

1. Huan Zhang
2. Anima Anandkumar
3. J. Kolter
4. Bo Li
5. Bowen Zhou
6. Duen Horng Chau
7. Shuai Di
8. Yujia Huang

Citations in the corpus

• On the Effectiveness of Lipschitz-Driven Rehearsal in Continual Learning (2022, neutral): “Many current approaches to Adversarial Learning similarly operate either by minimizing L at the global or local level [46, 73, 45] or by devising models characterized by a small L by design [22, 33].”
• Trustworthy AI: From Principles to Practices (2021, background): “To achieve certified robustness on unseen inputs as well, global robustness has recently attracted the interest of the AI community [77, 206].”
• Training Certifiably Robust Neural Networks with Efficient Local Lipschitz Bounds (2021, methodology): “Since the MaxMin activation function [15] has been shown to outperform ReLU on certified robustness [4, 30], we take MaxMin as an example to explain how to apply our local Lipschitz bound.”
• Universal and Transferable Adversarial Attacks on Aligned Language Models (2023, unknown): “State-of-the-art methods to prevent adversarial examples [Madry et al., 2018, Cohen et al., 2019, Leino et al., 2021] are almost never used in practice, owing to 1) their computational inefficiency, 2) the fact that they lead to substantial performance drops (in terms of “clean” model performance)…”
• LLM Self Defense: By Self Examination, LLMs Know They Are Being Tricked (2023, background): “Developing defenses against LLM attacks is an emerging field, unlike in the case of vision models, where there is a substantial body of literature exploring robust defenses against adversarial examples [34, 18, 5, 14].”
• Detecting Language Model Attacks with Perplexity (2023, background): “We find that roughly 95% of them can be identified via a perplexity above 1000 or an exact match on repeat ! characters at the end of a user’s query.”