Skip to content Skip to navigation

New models and algorithms for addressing limitations in deep reinforcement learning - Timothy Lillicrap

Timothy Lillicrap, Google DeepMind
April 1, 2019 - 5:10pm
Sloan Hall, Math Building 380, Room 380-C

Timothy Lillicrap

Google DeepMind

 

Abstract

There has been rapid progress in the field of deep reinforcement learning, leading to solutions to difficult control problems such as: playing video games from raw-pixels, controlling high-dimensional motor systems, and winning at the games of Go, Chess and StarCraft. Nevertheless, animal and human brains remain capable of behaviors that outstrip our best artificial agents, particularly in those capacities that require data efficiency, memory, long-term credit assignment, and planning in unknown environments.  I will describe new models and algorithms that work towards solving these limitations.  Several of these new models are inspired by the continued interplay between machine learning and neuroscience, and may offer powerful tools for understanding the brain.

Curiculum vitae

Related Papers

[1] Wayne, G., Hung, C., Amos, D., Mirza, M., Ahuja, A., Grabska-Barwin ́ska, A., Rae, J., Mirowski, P., Leibo, J.Z., Santoro, A., Gemici, M., Reynolds, M., Harley, T., Abramson, J., Mohamed, S., Rezende, D., Saxton, D., Cain, A., Hillier, C., Silver, D, Kavukcuoglu, K., Botvinick, M., Hassabis, D., and Lillicrap, T. (2018). Unsupervised Predictive Memory in a Goal-Directed Agent, arXiv preprint arXiv:1803.10760

[2] Hafner, D., Lillicrap, T., Fischer, I., Villegas, R., Ha, D., Lee, H., & Davidson, J. (2018). Learning Latent Dynamics for Planning from Pixels. arXiv preprint arXiv:1811.04551.

[3] Barth-Maron, G., Hoffman, M. W., Budden, D., Dabney, W., Horgan, D., Muldal, A., ... & Lillicrap, T. (2018). Distributed distributional deterministic policy gradients. arXiv preprint arXiv:1804.08617.