Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore the challenges and innovations in scaling Inverse Reinforcement Learning (IRL) for planetary-scale problems in this colloquium talk by Matt Barnes from Google Research. Delve into the development of Receding Horizon Inverse Planning (RHIP), a novel method that bridges classic IRL algorithms and offers fine-tuned control over performance trade-offs. Discover how these advancements led to a significant 16-24% improvement in Google Maps route quality, marking a milestone in real-world IRL application. Gain insights into the trade-offs between deterministic planners and stochastic policies in IRL, and learn about the practical challenges encountered when implementing these algorithms at scale. Understand the potential impact of these techniques on large-scale systems at the intersection of learning and planning, particularly in the context of serving billions of users in real-world applications.
