Interview
summary
Argues that modern deep learning already trains networks equal to humans at detecting cars and that researchers still do not know why those models work so well; focuses on building a hybrid stack for self-driving vehicles where motion planning for steering braking and throttle uses proven classical algorithms and relies on classical algorithms aided by machine learning for perception and upstream tasks; reports that the hybrid design is practical in deployment and avoids throwing away reliable nonlearning components.
AI-written summary · about
related talks
31mDeep learning for self-driving vehicle technology
51mHow Autonomous Vehicles Work
08mHuman–AI Collab for Reliable Automotive Control Systems
38mDesigning Machine Learning for Humans
13mArtificial General Intelligence: Why aren't we there yet?
28mChallenges & future trends of Deep Learning
29mDeep Learning, Neuroscience and the future of AI
29mThe Present & Future of AI and Games