Chapter
Challenges of Creating Adversarial Examples in the Physical World
30:37 - 42:11 (11:33)
This podcast discusses the difficulties of creating successful adversarial examples in the physical world due to the need for perceptible changes in images that can cause a difference from the camera side, which can be more challenging than in the digital world where presentations can be added anywhere in the image.
Clips
Creating adversarial attacks in the physical world is much harder than the digital world, as it requires accounting for variations in viewing distance and angle.
30:37 - 33:04 (02:26)
Summary
Creating adversarial attacks in the physical world is much harder than the digital world, as it requires accounting for variations in viewing distance and angle. Adding preservation to digital images is a common tactic for generating adversarial attacks, but physical attacks must be more carefully crafted.
ChapterChallenges of Creating Adversarial Examples in the Physical World
Episode#95 – Dawn Song: Adversarial Machine Learning and Computer Security
PodcastLex Fridman Podcast
This podcast episode discusses how adversarial examples can be created in the physical world through empirical experiments and theoretical models.
33:04 - 35:21 (02:17)
Summary
This podcast episode discusses how adversarial examples can be created in the physical world through empirical experiments and theoretical models. However, in the physical world, changes made to an object must be perceptible to the camera to be effective.
ChapterChallenges of Creating Adversarial Examples in the Physical World
Episode#95 – Dawn Song: Adversarial Machine Learning and Computer Security
PodcastLex Fridman Podcast
This episode discusses the research about adversarial examples in machine learning, highlighting the importance of their existence in advancing scientific understanding of computing.
35:21 - 38:42 (03:20)
Summary
This episode discusses the research about adversarial examples in machine learning, highlighting the importance of their existence in advancing scientific understanding of computing.
ChapterChallenges of Creating Adversarial Examples in the Physical World
Episode#95 – Dawn Song: Adversarial Machine Learning and Computer Security
PodcastLex Fridman Podcast
The current state of AI image classification systems is limited and not reflective of the richness of human vision.
38:44 - 42:11 (03:26)
Summary
The current state of AI image classification systems is limited and not reflective of the richness of human vision. A 2018 paper discusses the need for characterizing adversarial examples based on spatial consistency information for semantic segmentation.