Skip to content
conftrace_
2018 ICML ICML 2018

Differentiable Abstract Interpretation for Provably Robust Neural Networks

Abstract

We introduce a scalable method for training robust neural networks based on abstract interpretation. We present several abstract transformers which balance efficiency with precision and show these can be used to train large neural networks that are certifiably robust to adversarial perturbations.

🐣 Hot Topic Early Bird - adversarial perturbation

Authors

Matthew Mirman, Timon Gehr, Martin Vechev

Topics

Artificial Intelligence > Core AI > AI Safety Deep Learning > Architectures > Neural Networks

Keywords

adversarial robustness adversarial perturbation abstract interpretation certifiable robustness neural network
Download PDF

More from ICML 2018

All ICML 2018 papers →
One-Shot Segmentation in Clutter
Differentiable plasticity: training plastic neural networks with backpropagation
Training Neural Machines with Trace-Based Supervision
A Delay-tolerant Proximal-Gradient Algorithm for Distributed Learning
Data Summarization at Scale: A Two-Stage Submodular Approach
The Hierarchical Adaptive Forgetting Variational Filter