Deep mindhealth blockchain

1. Alphago zero is a new app for deepmind, a Google subsidiary
learning from the blank state, alphago zero can quickly learn go by itself without any human input, and beat the "predecessors" with a 100:0 record
History
editor
in a research paper published in the international academic journal Nature on the early morning of October 19, 2017, Google's deepmind reported that alphago zero, a new version of the program, started from a blank state, could quickly teach itself go without any human input, and beat the "predecessors" with a 100:0 record. As soon as deepmind's paper is published, TPU sales are likely to increase dramatically. His 100:0 record is suspected to be true[ 1] After three days of training, it beat his brother alphogo Lee 100:0, and after 40 days of training, it beat its other brother alphogo master [1]

working principle
editor
abandoning human experience "and" self training "are not the biggest highlights of alphago zero, The key lies in the adoption of a new reinforcement learning (reinforcement learning) algorithm, which brings new development to the algorithm[ 1]

achievements
editor
alphago zero has only four TPUs and zero human experience. Its self-training time is only three days, and the number of self playing chess games is 4.9 million. But it beat its predecessors 100-0[ 1]

2. Deepmind, co founded by demis hassabis, an AI programmer and neuroscientist, is a cutting-edge AI enterprise. It combines the most advanced technology of machine learning and system neuroscience to establish a powerful general learning algorithm
the initial achievements are mainly used in simulation, e-commerce, game development and other commercial fields
alphago, a program developed by deepmind
in March 2016, alphago defeated Lee se dol, the Korean go champion, 4:1, becoming a rare milestone in the field of artificial intelligence in recent years.

3. Unknown_Error

4. Netone accelerator

5. Before deepmind, torch was chosen as the main research framework because torch has better flexibility and faster speed. Deep mind's research based on torch has also led many people who want to follow their work to use torch. Similarly, TF as a framework very similar to theano, its flexibility will not be too bad, and with so many people in Google to maintain and update, the implementation speed of the new algorithm on it will not be very slow. When deepmind turns to TF, we can imagine that the status of TF in the research field will be improved, and deepmind can make a great contribution to the progress of TF. This is a good news for those who choose TF as a research tool in the future.

6. There are two reasons why the probability is low:
1. There are no Chinese employees in deepmind in the public data (update: a close friend in the comments area says there are Chinese employees, called aja Huang). Even if there are, they don't necessarily know the users (Update 2: for example, the Chinese students mentioned by Xiong Chenyan in the comments area)
2. The supply of AI talents is in short supply, especially those with deepmind work experience, Let out the wind will be crazy dig angle; For example, when openai was founded, it first visited a big bull, got a list, and then went to dig one by one. So I believe deepmind must have relevant PR strategies to prevent employees from exposing themselves publicly

7. On March 9, 2016, Google's deepmind Go program "alpha... This problem is not groundless. In a sense, AI has occupied the logo of IBM Watson. Image source: IBM, then, the technology of alphago

8. For example, when openai was founded, it first visited a big bull, got a list, and then went to dig one by one. Therefore, I believe deepmind has two reasons for its low probability of PR strategy:
1. There are no Chinese employees in deepmind in the public data (update: a friend in the comment area said that there are Chinese employees. The supply of AI direction bulls is in short supply, especially those with deepmind work experience, If you let out the news, you will be poached crazily: for example, in the comment area @ Chinese students mentioned by Xiong Chenyan)
2, you are not necessarily a Zhihu user (Update 2, even if there is one, it's called aja Huang)

9.

There are two reasons why the probability is small:

1. there are no Chinese employees in deepmind in the public information (update: there is a close friend in the comment area who says there are Chinese employees, called aja Huang). Even if there are, it is not necessarily known to users (Update 2: for example, the Chinese students mentioned by Xiong Chenyan in the comment area)

2. 2, In particular, Daniel, who has deep mind working experience, is bound to be crazily poached when he releases the wind; For example, when openai was founded, it first visited a big bull, got a list, and then went to dig one by one. Therefore, I believe deepmind must have relevant PR strategies to prevent employees from exposing themselves publicly
   

< / OL >