Facebook AI Research (FAIR) has released new features and research results, including a completely retrained model of ELF OpenGo using reinforcement learning. The new bot outperforms the previous bot, which defeated several top 30 professional Go players 20-0 and has been widely adopted by the AI research community to benchmark future research.
FAIR first released ELF: an Extensive, Lightweight, and Flexible platform for game research in 2017. ELF allowed researchers to test algorithms in various game environments, including board games, arcade games, and real-time strategy games like Go. Last year, FAIR open-sourced ELF OpenGo, based on the ELF platform. AI researchers used the game-playing bot to better understand how AI systems learn, and Go enthusiasts tested their skills against it as a new state-of-the-art artificial sparring partner.
Today’s model is the result of 20 million self-play games, all of which FAIR is sharing (along with 1,500 intermediate models). The team released a Windows executable version of the bot, which can be used as a training aid by Go players, along with a new data archive and analysis of 87,000 professional Go games over 300 years. An interactive tool provides a visualization of the archive and analysis.
The open-sourcing of these models provides an important benchmark for future research. FAIR is sharing their insights in a new paper, which could help researchers better understand the underlying mechanism and apply it to situations beyond Go.
Read the full story on our new Facebook AI blog.