Facebook Artificial Intelligence Research has released new capabilities and research outcomes to its ELF OpenGo bot. With the new features including an updated model, a Windows executable version of the bot, and a unique archive analyzing 87k professional Go games, ELF OpenGo bot is an AI bot that has outperformed world champion professional GO players based on its existing ELF platform for Reinforcement Learning Research.
Introduced in May last year, ELF OpenGo is an open-source reimplementation of the AlphaZero algorithm, is the first open-source Go AI which has convincingly manifested superhuman performance, achieving a (20:0) record against global top professionals. The team at Facebook AI Research has updated that ELF OpenGo model, by re-training the model from scratch. The team has also offered a data set of 20 million self-play games and the 1,500 intermediate models used to produce them. As a result, it further lessens the need for computing resources.
The Facebook AI Research team utilized the new model to examine the professional human players’ games; however, they observed that its ability to foresee the players’ moves went down very early during its learning process, after less than 10 percent of the total training time. But, as the model undergone for training, researchers found that its skills levels kept on improving. In due course, this updated feature led to it outperforming the team’s earlier prototype ELF OpenGo model by 60 percent of the time.