Robust and Accurate - Compositional Architectures for Randomized Smoothing


Loading...

Date

2022-04-29

Publication Type

Conference Paper

ETH Bibliography

yes

Citations

Altmetric

Data

Abstract

Randomized Smoothing (RS) is considered the state-of-the-art approach to obtain certifiably robust models for challenging tasks. However, current RS approaches drastically decrease standard accuracy on unperturbed data, severely limiting their real-world utility. To address this limitation, we propose a compositional architecture, ACES, which certifiably decides on a per-sample basis whether to use a smoothed model yielding predictions with guarantees or a more accurate standard model without guarantees. This, in contrast to prior approaches, enables both high standard accuracies and significant provable robustness. On challenging tasks such as ImageNet, we obtain, e.g., 80.0% natural accuracy and 28.2% certifiable accuracy against l2 perturbations with r = 1.0. We release our code and models at https://github.com/eth-sri/aces.

Publication status

published

Editor

Book title

Journal / series

Volume

Pages / Article No.

2204.00487

Publisher

Cornell University

Event

Workshop on Socially Responsible Machine Learning (SRML 2022), co-located with ICRL 2022

Edition / version

Methods

Software

Geographic location

Date collected

Date created

Subject

Adversarial Robustness; Certified Robustness; Randomized Smoothing

Organisational unit

03948 - Vechev, Martin / Vechev, Martin check_circle

Notes

Poster presentation on April 29, 2022.

Funding

Related publications and datasets