Towards Automated Testing and Robustification by Semantic Adversarial Data Generation

Shetty, Rakshith; Fritz, Mario; Schiele, Bernt

doi:10.1007/978-3-030-58536-5_29

Cited by 5 publications

(5 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, large datasets also contain substantial contextual biases, in background/context, rotation, viewpoint, etc. and have insufficient controls to ensure networks do not exploit trivial correlations in the data [2,60,69,48,47]. For instance, in [60] authors find that changing backgrounds in ImageNet significantly decreases average performance, and that choosing backgrounds in an adversarial manner can lead to misclassifying 87.5% of the images.…”

Section: Related Workmentioning

confidence: 99%

A biologically inspired architecture with switching units can learn to generalize across backgrounds

Voina,

Shea-Brown,

Mihalas

2023

Neural Networks

View full text Add to dashboard Cite

Section: Related Workmentioning

confidence: 99%

A biologically inspired architecture with switching units can learn to generalize across backgrounds

Voina,

Shea-Brown,

Mihalas

2023

Neural Networks

View full text Add to dashboard Cite

“…A recent strand of work is concerned with data augmentation for improving invariance against spurious correlations. Shetty et al (2020) propose to train object detectors on generated semantic adversarial data, effectively reducing the texture dependency of their model. Their finding is in line with (Geirhos et al, 2018) that proposes to transfer the style of paintings onto images and use them for data augmentation.…”

Section: Disentangled Representation Learningmentioning

confidence: 99%

Counterfactual Generative Networks

Sauer,

Geiger

2021

Preprint

View full text Add to dashboard Cite

Neural networks are prone to learning shortcuts -they often model simple correlations, ignoring more complex ones that potentially generalize better. Prior works on image classification show that instead of learning a connection to object shape, deep classifiers tend to exploit spurious correlations with low-level texture or the background for solving the classification task. In this work, we take a step towards more robust and interpretable classifiers that explicitly expose the task's causal structure. Building on current advances in deep generative modeling, we propose to decompose the image generation process into independent causal mechanisms that we train without direct supervision. By exploiting appropriate inductive biases, these mechanisms disentangle object shape, object texture, and background; hence, they allow for generating counterfactual images. We demonstrate the ability of our model to generate such images on MNIST and ImageNet. Further, we show that the counterfactual images can improve out-of-distribution robustness with a marginal drop in performance on the original classification task, despite being synthetic. Lastly, our generative model can be trained efficiently on a single GPU, exploiting common pre-trained models as inductive biases.

show abstract

“…semantic adversarial examples. These works often rely on synthetic data, using differentiable rendering or other optimization methods to find adversarial images by modifying scene parameters [5,20,21,22,23,24,25,26]. These include a custom differentiable renderer to perturb the camera, lighting, or object mesh vertices [20], and using a neural renderer where light is represented by network activations [21].…”

Section: Semantic Adversarial Attacks and In-distribution Brittlenessmentioning

confidence: 99%

Adversarial examples within the training distribution: A widespread challenge

Madan¹,

Sasaki²,

Pfister³

et al. 2021

Preprint

View full text Add to dashboard Cite

Neural networks are susceptible to small transformations including 2D rotations and shifts, image crops, and even changes in object colors. This is often attributed to biases in the training dataset, and the lack of 2D shift-invariance due to not respecting the sampling theorem. In this paper, we challenge this hypothesis by training and testing on unbiased datasets, and showing that networks are brittle to both small 3D perspective changes and lighting variations which cannot be explained by dataset bias or lack of shift-invariance. To find these in-distribution errors, we introduce an evolution strategies (ES) based approach, which we call CMA-Search. Despite training with a large-scale (∼ 0.5 million images), unbiased dataset of camera and light variations, in over 71% cases CMA-Search can find camera parameters in the vicinity of a correctly classified image which lead to in-distribution misclassifications with < 3.6% change in parameters. With lighting changes, CMA-Search finds misclassifications in 33% cases with < 11.6% change in parameters. Finally, we extend this method to find misclassifications in the vicinity of ImageNet images for both ResNet and OpenAI's CLIP model.

show abstract

Towards Automated Testing and Robustification by Semantic Adversarial Data Generation

Cited by 5 publications

References 19 publications

A biologically inspired architecture with switching units can learn to generalize across backgrounds

A biologically inspired architecture with switching units can learn to generalize across backgrounds

Counterfactual Generative Networks

Adversarial examples within the training distribution: A widespread challenge

Contact Info

Product

Resources

About