Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore the concept of generative neural network models for reinforcement learning environments in this 19-minute video by Yannic Kilcher. Dive into the research by David Ha and Jürgen Schmidhuber on building world models that can quickly learn compressed spatial and temporal representations of environments through unsupervised training. Discover how features extracted from these world models can be used as inputs to train compact and efficient agent policies. Learn about the fascinating possibility of training agents entirely within their own hallucinated dreams generated by their world models, and how these policies can be successfully transferred back to actual environments. Gain insights into this innovative approach that combines unsupervised learning, reinforcement learning, and generative models to create more efficient and adaptable AI agents.
