Blog

Unlike chess and Jeopardy, the game of Go has proved elusive for researchers seeking to craft a machine that can beat a human opponent. MIT Technology Review reports on advances by Facebook research, where a pair of scientists have recently created a pretty darn good Go-playing machine. Their approach combines a traditional search-based approach with neural network-based machine learning and trounces earlier Go machines over 90% of the time.

Read more at MIT Technology Review