Certifying Geometric Robustness of Neural Networks

Balunović, Mislav; Baader, Maximilian; Singh, Gagandeep; Gehr, Timon; Vechev, Martin

doi:10.3929/ethz-b-000395340

Cited by 9 publications

(3 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While also based on metamorphic properties, and more specifically on geometric transformations, the work in [1], specific to neural networks, tries to formally verify a neural network against these metamorphic transformations. By computing linear relaxations for these transformations, they are able to prove, by using external formal verification tools, the robustness of the model around a set of selected inputs.…”

Section: Related Workmentioning

confidence: 99%

“…In a sense, AIMOS allows to rapidly test a much larger set of inputs in a much shorter time, which offers a simple and rapid early problem detection mechanism. In layman's terms, our testing is useful to cover a much larger ground surface, detecting unwanted mines, so to speak, with little cost and in little time, whereas the formal verification techniques such that of [1] can verify at a much deeper level a much smaller space.…”

Section: Related Workmentioning

confidence: 99%

“…f i (x) = blur(x, k) where blur is a convolution by a kernel with k ∈ [1,20] the kernel size chosen for the blur with a step of 1.…”

Section: Renault Welding Use Casementioning

confidence: 99%

See 2 more Smart Citations

AIMOS: Metamorphic Testing of AI - An Industrial Application

Lemesle,

Varasse,

Chihani

et al. 2023

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Despite initial fears of a renewed AI winter, the current summer seems to have no end in sight. AI in general and Deep Learning in particular are permeating a growing number of our daily applications with an undeniable added value. As with Computer Science itself, this increasing entanglement with our lives calls for a special attention to safety and demands adequate verification and validation methods and tools to assist in the development of reliable AI. In this paper, we present the AIMOS tool as well as the results of its application to industrial use cases. Relying on the widely used Metamorphic testing paradigm, we show how the process of verification and validation can benefit from the early testing of models' robustness to perturbations stemming from the intended operational domain.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

AIMOS: Metamorphic Testing of AI - An Industrial Application

Lemesle,

Varasse,

Chihani

et al. 2023

Lecture Notes in Computer Science

View full text Add to dashboard Cite

show abstract

Efficient Certification of Spatial Robustness

Ruoss

Baader

Balunović

et al. 2021

AAAI

View full text Add to dashboard Cite

Recent work has exposed the vulnerability of computer vision models to vector field attacks. Due to the widespread usage of such models in safety-critical applications, it is crucial to quantify their robustness against such spatial transformations. However, existing work only provides empirical robustness quantification against vector field deformations via adversarial attacks, which lack provable guarantees. In this work, we propose novel convex relaxations, enabling us, for the first time, to provide a certificate of robustness against vector field transformations. Our relaxations are model-agnostic and can be leveraged by a wide range of neural network verifiers. Experiments on various network architectures and different datasets demonstrate the effectiveness and scalability of our method.

show abstract

DeformRS: Certifying Input Deformations with Randomized Smoothing

Alfarra

Bibi

Khan

et al. 2022

AAAI

View full text Add to dashboard Cite

Deep neural networks are vulnerable to input deformations in the form of vector fields of pixel displacements and to other parameterized geometric deformations e.g. translations, rotations, etc. Current input deformation certification methods either (i) do not scale to deep networks on large input datasets, or (ii) can only certify a specific class of deformations, e.g. only rotations. We reformulate certification in randomized smoothing setting for both general vector field and parameterized deformations and propose DeformRS-VF and DeformRS-Par, respectively. Our new formulation scales to large networks on large input datasets. For instance, DeformRS-Par certifies rich deformations, covering translations, rotations, scaling, affine deformations, and other visually aligned deformations such as ones parameterized by Discrete-Cosine-Transform basis. Extensive experiments on MNIST, CIFAR10, and ImageNet show competitive performance of DeformRS-Par achieving a certified accuracy of 39\% against perturbed rotations in the set [-10 degree, 10 degree] on ImageNet.

show abstract

Certifying Geometric Robustness of Neural Networks

Cited by 9 publications

References 11 publications

AIMOS: Metamorphic Testing of AI - An Industrial Application

AIMOS: Metamorphic Testing of AI - An Industrial Application

Efficient Certification of Spatial Robustness

DeformRS: Certifying Input Deformations with Randomized Smoothing

Contact Info

Product

Resources

About