An abstract domain for certifying neural networks


Loading...

Date

2019-01

Publication Type

Conference Paper

ETH Bibliography

yes

Citations

Altmetric

Data

Abstract

We present a novel method for scalable and precise certification of deep neural networks. The key technical insight behind our approach is a new abstract domain which combines floating point polyhedra with intervals and is equipped with abstract transformers specifically tailored to the setting of neural networks. Concretely, we introduce new transformers for affine transforms, the rectified linear unit (ReLU), sigmoid, tanh, and maxpool functions. We implemented our method in a system called DeepPoly and evaluated it extensively on a range of datasets, neural architectures (including defended networks), and specifications. Our experimental results indicate that DeepPoly is more precise than prior work while scaling to large networks. We also show how to combine DeepPoly with a form of abstraction refinement based on trace partitioning. This enables us to prove, for the first time, the robustness of the network when the input image is subjected to complex perturbations such as rotations that employ linear interpolation.

Publication status

published

Editor

Book title

Volume

3 (POPL)

Pages / Article No.

41

Publisher

Association for Computing Machinery

Event

46th ACM SIGPLAN Symposium on Principles of Programming Languages (POPL 2019)

Edition / version

Methods

Software

Geographic location

Date collected

Date created

Subject

Abstract interpretation; Deep Learning; Adversarial attacks

Organisational unit

03893 - Püschel, Markus / Püschel, Markus check_circle
03948 - Vechev, Martin / Vechev, Martin check_circle

Notes

Funding

Related publications and datasets

Is part of: