The Rebirth of the Male God Starts From Playing Games Chapter 598

Chapter 598 598. Please be a master who does not talk about martial arts

In previous versions, Bomi Go Master used a "strategy network" to choose the next move, and a "value network" to predict the winner after each move. In the new version, these two neural networks are combined into one, allowing it to be trained and evaluated more efficiently.

Bomi Go AI does not use fast and random moves. In previous versions, Bomi Go Master used the fast move method to predict which player would win the game from the current situation.

Instead, the new version relies heavily on its high-quality neural network to evaluate chess situations.

Now the "Bomi Go Master" improves the game by working together with two different neural network "brains".

These "brains" are multi-layered neural networks, similar in structure to those used by image search engines to identify images.

Liu Guangran and Cao Yang stood in the observation room and looked at the chess players opposite who were struggling in pain. In the small black room, there were many monitors, which were demonstrating the game situation between the chess players and the computer.

Cao Yang took a look and said that he couldn't understand. Anyway, in the end, as long as he could see who lost and who won, it would be fine.

"Our initial Go AI algorithm started with neural networks and deep learning, and then input a large number of game situations into the database, and then learned from these games to find more suitable methods." Liu Guangran told Cao Yang , “This set of plans at that time achieved good results, and at least it was able to defeat some professional second- and third-level players.”

Before doing this project, Liu Guangran actually did not play Go himself, but after completing this project, Liu Guangran was not only an AI expert, but his chess skills also improved rapidly.

I have received guidance from many professional chess players before, and I even had a 5th-dan professional player as my teacher.

Later, when Liu Guangran discovered that he could not defeat the AI he developed no matter what, he fell into deep despair. The reason why Liu Guangran was so happy recently was because he had completely given up on becoming an [Go master]'s thoughts.

It turns out that Liu Guangran was a fan of Jin Yong's martial arts when he was in college, especially the Zhenlong chess game in it. Now he feels -

This is all **** on his part.

If there is anything in this world that can surpass the limits of human beings, Liu Guangran has no doubt that it is artificial intelligence.

In a sense, Liu Guangran is happy.

Because he can focus on doing what he is good at instead of wasting time on unnecessary daydreams. After understanding that he could not beat the computer in playing Go, his whole heart became pure.

"The current Bomi Go masters have abandoned human chess records and only relied on deep learning to grow and challenge the limits of Go."

"Why!?"

Boss Cao turned around in surprise and looked at Liu Guangran in disbelief.

What the **** is this? !

In other words, you no longer let it learn chess records?

So how can it beat those professional chess players? !

I really want our Go master to have a good battle with the world champion and establish his authority in one battle.

Liu Guangran explained patiently, "Didn't I say it before? Our new improved plan is to give the Bomi Go master two brains.

They start with multiple layers of heuristic 2D filters to handle the localization of the Go board, much like a picture classifier network handles images. After filtering, 13 fully connected neural network layers produce judgments about the situations they see. These layers are capable of classification and logical reasoning.

The first brain: move selector

Bomi Go Master's first neural network brain is a "supervised learning policy network" that observes the board layout in an attempt to find the best next move.

In fact, it predicts the best probability of each legal next step, so the first guess is the one with the highest probability. This can be understood as a "move selector".

The second brain is called the chess game evaluator. The second brain of the Bomi Go master answers another question compared to the move selector. It does not guess the specific next step, but predicts the probability of each player winning given the position of the chess piece.

This "situation evaluator" is the "value network", which assists the move selector through overall situation judgment. This judgment is only approximate, but it is very helpful for improving reading speed.

By analyzing and classifying potential future positions as "good" or "bad," a Bomi Go master can decide whether to read further through a particular variant.

If the position evaluator says this particular variant won't work, the AI skips reading.

These networks check the results through repeated training, and then proofread and adjust the parameters to make the next execution better. This processor has a large element of randomness, so it's impossible to know exactly how the network "thinks", but more training can make it evolve better.

In order to cope with the complexity of Go, Bomi Go Master combines the advantages of supervised learning and reinforcement learning.

It forms a strategy network through training, takes the situation on the chessboard as input information, and generates a probability distribution for all feasible positions.

Then, a value network is trained to predict the self-playing game, and predict the results of all feasible positions on a scale from -1 (the opponent's absolute victory) to 1 (the Bomi Go master's absolute victory).

Both networks are powerful on their own, but Bomi Go master achieved its true advantage by integrating them into a probability-based Monte Carlo tree search.

The new version of Bomi Go Master generates a large number of self-playing games, providing training data for the next generation version, and the process repeats itself.

After obtaining the chess game information, the Bomi Go master will explore which position has both high potential value and high probability based on the strategy network, and then determine the best position.

At the end of the allotted search time, the position most frequently visited by the system during the simulation will become the final choice of the Bomi Go master.

After comprehensive exploration in advance and constant speculation on the best moves during the process, Bomi Go Master's search algorithm can add near-human intuitive judgment to its computing power. "

"Hmm...very good." Cao Yang nodded, "I'm already confused."

Oh my god, this guy actually wants to teach me.

I'm sorry, Guangran, I let you down.

Although Boss Cao graduated from Tsinghua University in Computer Science, he was still confused.

The main reason is that I haven't done research work for many years. After leaving this industry for a long time, it is inevitable that there will be some regression.

Liu Guangran:…

"Uh... in human terms, we didn't let the computer learn how to play Go, but it can play left and right, and at the same time it can understand the principle of winning without any move, just like in the movie "The Heaven and the Dragon Sword", Zhang Sanfeng teaches Zhang Wuji how to learn Tai Chi Boxing, when he has forgotten all the moves, he has already mastered this magical skill. "

"Oh I got it."

Cao Yang nodded seriously, I understand now.

In the past few years, it can be said that the professional players who participated in the Bomi Go AI test have experienced a process from heaven to hell.

When I first played chess with the Bomi Go master, the AI seemed very stupid. Sometimes it would often freeze, or it would suddenly go into fits and play chess pieces randomly.

Sometimes when someone surrounds you, the dragon is still trying to escape.

Sometimes the Bomi Go master is particularly obsessed with robbing.

According to their previous thoughts, do you think this stupid thing developed by Bomi deserves to be called artificial intelligence?

It's better to spend your money elsewhere where the returns may be higher.

However, they were later slapped in the face by the speed of light.

Because they found that the opponents they faced were constantly improving and growing every day.

This feeling is very scary, and it grows much faster than people imagine.

As time goes by, Bomi Go masters make fewer mistakes.

Then some professional players who could have won 100% of the games started to lose to AI, but the Go community didn't take it seriously at the time, mainly because the previous Go master Bomi was too stupid.

Later, these professional players couldn't help but take it seriously, and every time they played chess, they regarded each other as a truly powerful chess player.

Gradually, he put away his contempt.

But even in this situation, as time goes by, he becomes increasingly invincible.

Until recently, 9 out of 10 players lost, and these days it is completely impossible to beat Bomi's Go master.

What's even more frightening is that these 5-dan chess players sitting in the small dark room simply can't understand the chess routines of Bomi Go master.

During the game, the random move it made seemed very strange at first glance. You thought the AI had suddenly gone crazy like before, but when you chased it, you realized that you had fallen into the enemy's trap.

In the past, when playing chess with the Bomi Go masters, these professional chess players would either win a big victory or a big defeat, and the gap could be as high as ten or twenty goals.

But recently they discovered that Bomi Go Master has evolved into a real [Master], the kind that completely crushes you in terms of strength, but also takes care of your self-esteem, and in the end can barely win you by a glance. .

But every time you play chess, you have a feeling of being tied up. This feeling is very uncomfortable.

Facing a Go master who talks about [Martial Arts].

Sometimes they even hope that it will be less about martial arts and better, so that it will not give you some vain hopes, and then continue to humiliate you by winning half an eye in the next game.

Cao Yang observed silently, looked at the battle situation on the screen, and then looked at the changes in expressions on the faces of the players in the small black room.

In fact, it mainly depends on the changes in their expressions. After all, Cao Yang doesn't understand Go.

From the hard and frowned brows on their faces, it can be seen that the current Bomi Go masters have indeed reached a very powerful level. (End of this chapter)

Chapter end

Report