The main working principle of AlphaGo is "deep learning". "Deep Learning" refers to the method of multi -
This layers and how to train it. A layer of neural networks will use a large number of matrix numbers as input, the weight of the non -linear activation method, and then generate another data set as the output. This is just like the working mechanism of the biological nerve brain. Through the appropriate number of matrix, the multi -layer tissue link together forms a neural network "brain" for accurate and complex processing, just like people identify objects to label pictures.
mainly includes 4 parts: 1. 2. Fast rollout, the goal is the same as 1, but under the condition of proper sacrificing chess quality, the speed is 1000 times faster than 1. 3. Value Network, given the current situation, is estimated to be Bai Sheng or Black Sheng. 4. Monte Carlo Tree Search (MCTS), connect the above three parts to form a complete system.
AlphaGo (AlphaGo) is to improve chess through two different neural network "brain" cooperation. These brains are multi -layered neural networks are similar to those Google picture search engine recognition pictures. They start with multi -layer inspiration two -dimensional filters to process the positioning of Go board, just like the picture classifier network processing pictures. After filtering, 13 completely connected neural network layers generate the situation judgment they see. These layers can be classified and logical reasonable.
The main working principle of AlphaGo is "deep learning". "Deep Learning" refers to the method of multi -
This layers and how to train it. A layer of neural networks will use a large number of matrix numbers as input, the weight of the non -linear activation method, and then generate another data set as the output. This is just like the working mechanism of the biological nerve brain. Through the appropriate number of matrix, the multi -layer tissue link together forms a neural network "brain" for accurate and complex processing, just like people identify objects to label pictures.
mainly includes 4 parts:
1.
2. Fast rollout, the goal is the same as 1, but under the condition of proper sacrificing chess quality, the speed is 1000 times faster than 1.
3. Value Network, given the current situation, is estimated to be Bai Sheng or Black Sheng.
4. Monte Carlo Tree Search (MCTS), connect the above three parts to form a complete system.
AlphaGo (AlphaGo) is to improve chess through two different neural network "brain" cooperation. These brains are multi -layered neural networks are similar to those Google picture search engine recognition pictures. They start with multi -layer inspiration two -dimensional filters to process the positioning of Go board, just like the picture classifier network processing pictures. After filtering, 13 completely connected neural network layers generate the situation judgment they see. These layers can be classified and logical reasonable.