US time on October 18, "Nature" magazine published in the official website of a "Mastering the game of Go without human knowledge" of the paper.I see PDF ) Details the characteristics of DeepMind's artificial intelligence algorithm AlphaGo Zero.
AlphaGo Zero is the most powerful place, it does not require the guidance of human experts, will be able to self-intensive learning to carry out training. Straightforward to say that before defeating Go Master Shi Shishi AlphaGo may need to provide it with a variety of master chess Training, and Zero is a completely white, humans do not have to teach it anything, just tell it the rules, it can completely through the self-learning method to master Go.
Compared with the previous generation version, Zero has a significant improvement in efficiency, and now only need four TPU (by Google to build a dedicated AI processor) and 72 hours of training, can be better than a few months before the training time Of AlphaGo.
According to Demis Hassabis, co-founder of DeepMind, the emergence of AlphaGo Zero means that they can now provide an artificial intelligence engine similar to scientific research to society, which, by striving from human dependence, makes artificial intelligence in other scientific fields Application becomes possible.
US time on October 18, "Nature" magazine published in the official website of a "Mastering the game of Go without human knowledge" of the paper.I see PDF ) Details the characteristics of DeepMind's artificial intelligence algorithm AlphaGo Zero.
AlphaGo Zero is the most powerful place, it does not require the guidance of human experts, will be able to self-intensive learning to carry out training. Straightforward to say that before defeating Go Master Shi Shishi AlphaGo may need to provide it with a variety of master chess Training, and Zero is a completely white, humans do not have to teach it anything, just tell it the rules, it can completely through the self-learning method to master Go.
Compared with the previous generation version, Zero has a significant improvement in efficiency, and now only need four TPU (by Google to build a dedicated AI processor) and 72 hours of training, can be better than a few months before the training time Of AlphaGo.
According to Demis Hassabis, co-founder of DeepMind, the emergence of AlphaGo Zero means that they can now provide an artificial intelligence engine similar to scientific research to society, which, by striving from human dependence, makes artificial intelligence in other scientific fields Application becomes possible.
