The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Trends in Deep Reinforcement Learning with Kamyar Azizzadenesheli - #560
Feb 21, 2022
Kamyar Azizzadenesheli, an Assistant Professor at Purdue University and an expert in deep reinforcement learning, dives deep into the evolution of the field. He discusses the interplay between reinforcement learning, robotics, and control theory, and highlights the importance of stable controllers for real-world applications. Kamyar predicts trends like self-supervised learning's rise and emphasizes the need for specialized algorithms. The conversation touches on risk-sensitive reinforcement learning and the innovations transforming decision-making in high-stakes environments.
AI Snips
Chapters
Transcript
Episode notes
Deep RL Progress
- Deep reinforcement learning (RL) has seen significant theoretical and practical advancements.
- These advancements haven't received as much media attention as NLP progress.
RL in Robotics
- RL methods are being adopted in robotics, enabling advancements like guaranteed drone flight.
- This adoption allows drones to perform complex maneuvers in extreme, unknown conditions.
Guarantees in Robotics
- Guarantees in robotics algorithms ensure reliability in extreme scenarios, like drone flight in unknown winds.
- Guaranteed algorithms eliminate extensive fine-tuning, saving time and resources.