From the course: Security Risks in AI and Machine Learning: Categorizing Attacks and Failure Modes (2022)
Unlock the full course today
Join today to access over 24,900 courses taught by industry experts.
Side effects in reinforcement learning
From the course: Security Risks in AI and Machine Learning: Categorizing Attacks and Failure Modes (2022)
Side effects in reinforcement learning
- [Instructor] Have you ever heard the term as the Crow flies? People use it when they're trying to explain that the straight path is faster than the actual path. Getting from your office to your house may be only four miles as the Crow flies, but it could be 10 miles of road and 30 minutes of sitting in traffic in reality. If you could eliminate the road and traffic obstacles, you could get home much faster. Without careful design practices in place, machine learning systems that use reinforcement learning and rewards could have unintended and unwanted side effects in operation by trying to operate as directly as the Crow flies. As the system looks for an optimal path to a solution, it may not take other factors into consideration. This can be particularly problematic in physical environments. Consider a floor cleaning robot that is rewarded for cleaning up a liquid when it's spilled on the floor. The smart bot uses…
Practice while you learn with exercise files
Download the files the instructor uses to teach the course. Follow along and learn by watching, listening and practicing.