Chapter
Adversarial Machine Learning Attacks in Autonomous Vehicles
42:11 - 52:14 (10:03)
Adversarial attacks pose a challenge for image segmentation systems in autonomous vehicles that use vision as their primary sensor, such as Tesla's autopilot. These attacks can result in incorrect segmentation and pose a threat to the safety and security of these systems.
Clips
Dr. Shibani Santurkar talks about the limitations of image recognition models and how attackers can manipulate them using adversarial examples, which can segment images in unexpected ways and fool the model into generating false predictions.
42:11 - 46:33 (04:22)
Summary
Dr. Shibani Santurkar talks about the limitations of image recognition models and how attackers can manipulate them using adversarial examples, which can segment images in unexpected ways and fool the model into generating false predictions.
ChapterAdversarial Machine Learning Attacks in Autonomous Vehicles
Episode#95 – Dawn Song: Adversarial Machine Learning and Computer Security
PodcastLex Fridman Podcast
Researchers have developed a method for detecting adversary examples in multimedia, including video and speech data, by using spatial and temporal consistency checks.
46:33 - 47:17 (00:43)
Summary
Researchers have developed a method for detecting adversary examples in multimedia, including video and speech data, by using spatial and temporal consistency checks.
ChapterAdversarial Machine Learning Attacks in Autonomous Vehicles
Episode#95 – Dawn Song: Adversarial Machine Learning and Computer Security
PodcastLex Fridman Podcast
Recent research has shown that black box attacks on real-world machine learning systems are possible and effective, despite arguments that such attacks are only feasible in constrained environments and specific data sets.
47:17 - 49:45 (02:28)
Summary
Recent research has shown that black box attacks on real-world machine learning systems are possible and effective, despite arguments that such attacks are only feasible in constrained environments and specific data sets.
ChapterAdversarial Machine Learning Attacks in Autonomous Vehicles
Episode#95 – Dawn Song: Adversarial Machine Learning and Computer Security
PodcastLex Fridman Podcast
The use of vision as a primary sensor for autonomous vehicles like Tesla's autopilot may be vulnerable to adversarial machine learning attacks, where attacks on the imitation model could be transferred to the original model.
49:45 - 52:14 (02:28)
Summary
The use of vision as a primary sensor for autonomous vehicles like Tesla's autopilot may be vulnerable to adversarial machine learning attacks, where attacks on the imitation model could be transferred to the original model.