#49 - Meta-Gradients in RL - Dr. Tom Zahavy (DeepMind)
Machine Learning Street Talk (MLST)
00:00
Understanding Meta-Gradients in Reinforcement Learning
This chapter explores the role of meta-gradients as an optimization technique in reinforcement learning, particularly in non-stationary environments. It highlights the complexities of balancing exploration and exploitation, discussing both theoretical implications and practical applications such as hyperparameter tuning. The conversation also contrasts single lifetime and multi-lifetime learning approaches, examining the challenges of adapting algorithms to varying contexts.
Transcript
Play full episode
