World Simulation
Meta-learning
To train complex models, parameter space needs to be explored effectively and key features recognized. How can we best learn how to learn? We research adaptive strategies for adjusting the training path or the attribution of rewards.