HackerPapers

arXiv:1707.07397cs.CV

Synthesizing Robust Adversarial Examples

Anish Athalye, Logan Engstrom, Andrew Ilyas, Kevin Kwok

Abstract

Standard methods for generating adversarial examples for neural networks do not consistently fool neural network classifiers in the physical world due to a combination of viewpoint shifts, camera noise, and other natural transformations, limiting their relevance to real-world systems. We demonstrate the existence of robust 3D adversarial objects, and we present the first algorithm for synthesizing examples that are adversarial over a chosen distribution of transformations. We synthesize two-dimensional adversarial images that are robust to noise, distortion, and affine transformation. We apply our algorithm to complex three-dimensional objects, using 3D-printing to manufacture the first physical adversarial objects. Our results demonstrate the existence of 3D adversarial objects in the physical world.

arXiv comments
ICML 2018
Journal reference
DOI
Subjects
cs.CVComputer Vision and Pattern Recognition

On Hacker News

Linked by 7 identifiable commenters in 7 comments across 7 threads, from Oct 2017 to Sep 2024. Found in 1 HN submission from Nov 2017 to Nov 2017. Open submission discussions.

Comments linking this paper, per year.

Discussed in comments7 threads

  1. Big Tech Sees Like a State (2020) 1 comment

    • > But powerful humans are also gamed, and usually more easily than computers. It's the "more easily" that I'd dispute. A computer can be examined in detail, its weaknesses probed ruthlessly and systematically until a vulnerability is found. Any one human will have traits in…

      ben_wView on HNParent context

  2. Fawkes: Image “Cloaking” for Personal Privacy 1 comment

    • Applying some form of transformation is a weak countermeasure to adversarial examples. Especially simple ones such as low pass filters. https://arxiv.org/pdf/1707.07397.pdf Neural Networks do not learn what humans learn. They can learn completely different and sometimes much…

      dijksterhuisView on HNParent context

  3. Contrastive Self-Supervised Learning 1 comment

    • The problem that I see with supervised training of a linear classifier after unsupervised training is that if the unsupervised network is large enough, it allows the supervised trainer to choose the working components. As shown in [1] that can lead to randomly initialized…

      fxtentacleView on HNParent context

  4. Researchers trick Tesla autopilot into driving into opposing traffic 1 comment

    • I'm not sure why hackers prefer to hack large scale computer systems rather than individual humans, but they do. So we have to protect neural networks against adversarial examples for the same reason we have to protect databases against sql injections. > so what's the critical…

      bnegreveView on HNParent context

  5. Sikorsky’s autonomous helicopter tech is ready for takeoff 1 comment

    • I think they are using the wrong word here, I think by "determinism" they mean interpretable. You train your big neural network to classify images, it works fine on the test set but then one day in the wild you discover your network classified a turtle as a rifle [0]. Why did it…

      sgillenView on HNParent context

  6. AI and Deep Learning in 2017 – A Year in Review 1 comment

    • I think for most people that hurdle of understanding is immense, and isn't likely to get any better. ML/DL infrastructure will continue to improve and deploying AI systems will become more accessible as time goes on. But I don't see familiarity with (non) convex optimization…

      nealeratzlaffView on HNParent context

  7. The Neural Net Tank Urban Legend 1 comment

    • I think the author's conclusion- that this scenario is unrealistic and would never happen given today's understanding of machine learning techniques- is extremely optimistic. NNs are demonstrably[1] not robust image classifiers. In my opinion, it's far more dangerous to downplay…

      RodgerTheGreatView on HNParent context

Submitted to Hacker News

Connected through shared HN threads, not inferred similarity or a recommendation.