Stochastic Security: Adversarial Defense Using Long-Run Dynamics of Energy-Based Models

May 27, 2020 Β· Entered Twilight Β· πŸ› International Conference on Learning Representations

πŸŒ… TWILIGHT: Old Age
Predates the code-sharing era β€” a pioneer of its time

"Last commit was 5.0 years ago (β‰₯5 year threshold)"

Evidence collected by the PWNC Scanner

Repo contents: LICENSE, README.md, bpda_eot_attack.py, config_attack, config_train_clf, config_train_ebm, nets.py, requirements.txt, train_clf.py, train_ebm.py, utils.py

Authors Mitch Hill, Jonathan Mitchell, Song-Chun Zhu arXiv ID 2005.13525 Category stat.ML: Machine Learning (Stat) Cross-listed cs.CR, cs.LG Citations 87 Venue International Conference on Learning Representations Repository https://github.com/point0bar1/ebm-defense ⭐ 26 Last Checked 1 month ago
Abstract
The vulnerability of deep networks to adversarial attacks is a central problem for deep learning from the perspective of both cognition and security. The current most successful defense method is to train a classifier using adversarial images created during learning. Another defense approach involves transformation or purification of the original input to remove adversarial signals before the image is classified. We focus on defending naturally-trained classifiers using Markov Chain Monte Carlo (MCMC) sampling with an Energy-Based Model (EBM) for adversarial purification. In contrast to adversarial training, our approach is intended to secure pre-existing and highly vulnerable classifiers. The memoryless behavior of long-run MCMC sampling will eventually remove adversarial signals, while metastable behavior preserves consistent appearance of MCMC samples after many steps to allow accurate long-run prediction. Balancing these factors can lead to effective purification and robust classification. We evaluate adversarial defense with an EBM using the strongest known attacks against purification. Our contributions are 1) an improved method for training EBM's with realistic long-run MCMC samples, 2) an Expectation-Over-Transformation (EOT) defense that resolves theoretical ambiguities for stochastic defenses and from which the EOT attack naturally follows, and 3) state-of-the-art adversarial defense for naturally-trained classifiers and competitive defense compared to adversarially-trained classifiers on Cifar-10, SVHN, and Cifar-100. Code and pre-trained models are available at https://github.com/point0bar1/ebm-defense.
Community shame:
Not yet rated
Community Contributions

Found the code? Know the venue? Think something is wrong? Let us know!

πŸ“œ Similar Papers

In the same crypt β€” Machine Learning (Stat)

R.I.P. πŸ‘» Ghosted

Graph Attention Networks

Petar VeličkoviΔ‡, Guillem Cucurull, ... (+4 more)

stat.ML πŸ› ICLR πŸ“š 24.7K cites 8 years ago
R.I.P. πŸ‘» Ghosted

Layer Normalization

Jimmy Lei Ba, Jamie Ryan Kiros, Geoffrey E. Hinton

stat.ML πŸ› arXiv πŸ“š 12.0K cites 9 years ago