> ML_LITERATURE // SILVER-2016-MASTERING-GAME-OF-GO-DEEP-NEURAL-NETWORKS-TREE-SEARCH_v1.0
Mastering the game of Go with deep neural networks and tree search (AlphaGo)
David Silver, Aja Huang, Chris J. Maddison, Arthur Guez, Laurent Sifre, George van den Driessche, Julian Schrittwieser, Ioannis Antonoglou, Veda Panneershelvam, Marc Lanctot, Sander Dieleman, Dominik Grewe, John Nham, Nal Kalchbrenner, Ilya Sutskever, Timothy Lillicrap, Madeleine Leach, Koray Kavukcuoglu, Thore Graepel, Demis Hassabis · Nature (2016)
seminal-architecture2016foundationalthirdPartyReproduced
Principal Contribution
Combined value networks to evaluate board positions with policy networks to select moves, guided by Monte Carlo tree search, defeating 18-time world champion Lee Sedol.
Operational Relevance
Watershed moment in AI history demonstrating that intuitive planning and superhuman mastery in massive search spaces is possible.
Assumptions
- Combining deep neural position evaluation with rollouts reduces tree search depth and branching factor to tractable levels in Go
Limitations
- Required initial supervised bootstrapping from 30 million human expert moves before self-play reinforcement learning
Connected Algorithms, Architectures & Tools
Related Algorithms:
Related Architectures:
Implementing Libraries: