Chapter 229： [One night mysterious master (2 in 1)]

, The latest chapter in the latest global monopoly of technology!

As the game enters the confrontation phase, Luo Sheng looks forward to watching Nafu's first perspective full and seeing how it performs.

This confrontation is [Human vs. Zerg], the human opponent is a Zerg player, and Nave chooses the human race.

Just after the game started, Luo Sheng watching Nafu's first perspective felt awkward, because "she" controlled the four farmers to wander around the map, and then turned back to mining for about half a minute.

Well, the start is bad.

It also hits peat, which is really mentally handicapped.

This scene directly drunk Luo Sheng.

I didn't understand that she thought she was full of curiosity about the world.

Later, Nuff built two houses in a row. Luo Sheng thought that "she" would put down the barracks building, but "she" did not build the barracks.

I saved some money to open the second mine ...

From the first perspective, Luo Sheng suddenly felt that these behaviors of Nafu's operation are just like a pure newcomer's first contact with the game of StarCraft, without any difference.

As the minutes passed, the Zerg players who came against it came to scout and saw the opponent's start.

Zerg players also look silly.

What the hell?

What kind of play is this?

So a wave of puppies came over, and at this time Nave had just built the first machine gunman and hadn't come out yet.

With the adversary's tentative attack puppies entering the mining area, Nave resisted for a while and was unable to return to the sky, eventually playing GG to end the game.

The first confrontation was killed by a small group of reconnaissance troops.

In fact, the human player on the opposite side is not high, just an ordinary player. To meet the real top professional players, Nave needs to constantly fight on the ladder to defeat a personal opponent. Each victory will match more Strong opponents until you reach the top of the pyramid to confront the real top professional players.

Of course, with Luo Sheng's technology, the algorithm of the matching system can be forcibly modified to let Nafu directly match the top players, but he didn't do that.

Currently, StarCraft is a popular e-sports game nowadays, and there are many people playing it, especially in South Korea.

There is even an exaggerated statement that, in South Korea, a 10-40 year-old man would be ridiculed if he said he would not play interstellar.

...

Soon, the second round of confrontation began, this time it was the TVT confrontation round, which was [Human vs Human].

In the second round of confrontation, Nave made a significant improvement. Although this match was also lost, at least the soldiers were pushed to the opponent's house.

After all the main forces were sent away, they were pushed back by the opponent's mechanized tactics.

Hit GG again.

The next one quickly reopened.

...

When it came to the fifth round of confrontation, it was TVZ again, which was [Human vs. Zerg].

Nave, who lost the game one after another, finally showed the extraordinary side of "her", and her strong learning ability allowed her to adapt to this level of "TVZ" confrontation, and easily defeated her opponent with operation and crush.

Although this match against Nave defeated the opponent with an operation level comparable to that of top professional players, one of the details was that "she" would never make a second mistake in the same place.

In other words, Nave's learning progress is too scary, and every confrontation is learning for "her".

Luo Sheng also saw a defect exposed by Nafu in these five games against human players. In addition to the specific division tactics, "She" could not form a flexible force distribution.

In the lost fourth game, a high-level human player constantly harassed Nafu's base, and once she returned to the opponent, she immediately ran away, waiting for her soldiers to go out and start. He turned back on the offensive and continued to harass.

In this confrontation, faced with the flexible troop dispatch strategy of human players, Nave could only passively respond to the challenge, unable to form an effective attack on human players, and also caused the final snowball effect.

Lost in the end.

After the end of the fifth game, Luo Sheng told Nafu to stop matching opponents, and then re-optimized it a lot.

He felt that such learning efficiency still seemed too slow.

Nafu's agent uses a deep learning algorithm. After Luo Sheng's careful upgrade, it includes neural networks, self-playing games through reinforcement learning, multi-agent learning, imitation learning and other algorithms or technologies.

The so-called self-playing game is simply learning on its own, and through continuous trial and error, find the way to maximize your winning rate.

And through self-playing and reinforcement learning, it constitutes the basic paradigm of open-end learning algorithms, learning human strategies, and ensuring that Nave's agent continuously explores these strategies throughout the game, thereby allowing "her" to get a greater upgrade. And progress.

In addition, due to the huge action space of StarCraft, many previous reinforcement learning methods will be invalidated.

Luo Sheng therefore also developed a new algorithm for Nafu for non-strategic reinforcement learning, so that "she" can effectively update the decision from the old historical decision game.

The ultimate goal of Luo Sheng is not to make Nafu an "IMBA" StarCraft AI player, but to use it as a starting point to make "she" capable of other complex and varied application scenarios.

If "She" just becomes a super master in the StarCraft game, changing a scene, such as in Go, language or other fields, will not work, then it is still a pure artificial mental retardation, and there is a huge gap with true artificial intelligence.

...

As time passed by minute by minute, Luo Shengma spent more than three hours in his office unconsciously, and completed the second optimization and upgrade for Nafu before leaving work. It was too late.

Before Luo Sheng left the office, he put Nafu into the ladder of Blizzard Star Wars Network, so that "she" and the human players kept fighting.

I'm not in charge of all this, and I will go to the Cote d'Azur company in the afternoon to handle some work and a meeting.

The stocked Nuff is doing one thing, constantly fighting against human players.

Nafu's backing "her" is almost the entire Sioux City data center, which consumes computing power and electricity costs.

In fact, if there is no money, it is impossible to support an artificial intelligence, let alone the development stage.

If it is an artificial intelligence research and development institution like IBM, it may be the developer who restricts its development, but Luo Sheng, who has witnessed artificial intelligence, is the easiest for development.

But there are still problems that restrict him.

That's the hardware. Nave's growth is limited by the current level of hardware. If there is a better hardware foundation, it will usher in a qualitative leap.

Deep networks require high-end GPUs to be trained within a reasonable amount of data. These hardware are very expensive, of course, there is no such thing as Luo Sheng here.

But the problem is that there is no hardware that meets his requirements, and you ca n’t buy it if you have money. In the future, you can only start the stove to make the hardware.

It is not feasible to train deep networks to achieve high performance without advanced hardware foundation.

Not only using high-end GPUs, but also fast CPUs, storage, and fast and very large amounts of RAM.

...

Over time, through the continuous battle, Nave's win rate from the first inning of her career was horrible, and gradually reversed, and the win rate is rapidly increasing.

At the operational level, Nauf is undoubtedly the number one in the world. No human player can compete with "her" at this level, even if it is restricted by APM and EPM.

But this is not the point. The focus is on the "consciousness", "behavior" and "decision" that Nave showed in the game. This is the embodiment of intelligence.

...

Unconsciously, Nave has already played 27 games on the ladder, and the winning percentage has climbed to about 70%, which is a very high winning percentage.

In order to improve the efficiency of Nafu ’s "evolution", Luo Sheng entered the massive database of Bluestar Technology Group and some websites of StarCraft to access the "Battle Report" videos of some top players while fighting against human beings. The player's game playback video is trained in parallel, and then confronted with other opponents to train.

A script running in the background will allow "her" to iteratively upgrade at regular intervals, update through reinforcement learning, and then conduct new confrontations.

Each iteration matches a new opponent, and at the same time freezes the original defeated opponent. The probability of matching the opponent determines the learning objective function adopted by the agent, so that it maintains diversity and increases difficulty.

With the progress of self-game, Nave gradually developed more and more mature tactics, or relatively optimal solutions. This process is similar to the process of human players discovering tactics. New tactical systems continue to defeat the old tactical system. Then replace the old tactical system.

...

In the 32nd round of confrontation, just like climbing the ladder, Nafu continued to defeat his opponents, constantly iterating on himself, becoming more and more powerful, and the human players he encountered were getting stronger and stronger, but 90% have become "her "Your defeat.

At this moment, another round of confrontation has begun, and the players matched by Nave have reached the level of confrontation of professional players.

All human players who were defeated by "her" didn't know that their opponent only started to contact this game today, let alone that their opponent was not a human, but an AI.

This round of confrontation is [Human vs Zerg], and the IDs of both sides are [Nov] and [= Y.J = F91].

At the beginning of the battle, the two sides normally started operations and reconnaissance. The human player whose ID is [= Y.J = F91] is located in a room in Hangcheng. He has left the "Ideology" team.

However, he still retains this ID, and he is now a real interstellar professional player. He is a first-class top player in the domestic interstellar world. Jianghu people sent the nickname "Jiangnan Dog King". As a Zerg player, his puppy operates Five minutes before the start of the game, it was called world-class.

In five minutes ...

Well, if F91 goes to South Korea, his strength is a dishwasher there. Anyone who can squat a water dispenser can hang him, and if he pulls out a Korean professional player, he may hang him.

"Nov ...? Whose vest is this ID? It's a good operation and a bit of strength." At this time, the confrontation has entered the middle stage, and F91 did not take the opponent lightly to win the game. On the contrary, it was quite difficult to play, and he had even clearly felt the pressure brought by his opponent.

The point is that he is very confused now. He has never seen this ID. The players in the domestic interstellar circle who have this strength are basically the cards, and they are basically known.

But the ID of Nov was the first time I saw it, and the style of play was also the first time I saw it, which directly allowed F91 to rule out those familiar domestic star players one by one.

"No matter which way the gods come, they will have to die when they meet the brother." F91 meditated in his heart, and then his main force began to go out, preparing to launch an attack.

[Chichi is under attack! ]

However, just a few moments after his main force went out, the sound of the headset madly prompted the base to be attacked. F91 subconsciously cut the screen to his main mine, and then saw a picture that made him extremely shocked.

"Baking, lying, what's up? What is this? Spicy buttocks ... lying down!"

It turned out that at the same time that F91 was preparing to launch an attack, Nave had sent five boatmen and launched an attack, and also intentionally detoured at six o'clock on the map to avoid the vision of the opponent in the middle of the map. Use radar to do anti-reconnaissance in advance to ensure that your airdrop troops are not found by the other party.

If Luo Sheng saw it, he would be very satisfied. This seems to be a very common decision for human players, but if it is an AI, it will be different.

This progress is not small.

It is worth mentioning that although Nafu is AI, "she" does not have "open maps", there is also the fog of war, and the operation also needs to cut the screen like a human player. In these parameters, ensure that it is the same as a human player, otherwise it is Bullying.

F91's troops had just gone out, and Nave's troopers then successfully reached the opponent's main mine heights.

Then, F91 saw the operation screen that made him extremely shocked.

The opponent's five transport planes actually dropped airdrops while flying, making the combat troops tiled on the map like a carpet, thereby achieving the maximum firepower output in the shortest time.

This wave of airdrops directly crashed the F91's main mine, but for a moment, another wave of three-boat rug-type airdrops completely destroyed the F91 sub-mine base and a small number of troops left behind.

Seeing the two in the family blossom, the main force has advanced more than half. At this time, the return of support is meaningless and it must be lost.

So F91 decided to fight one shot.

Change home with your opponent!

The main force does not return support, only in this way may still have a chance of winning.

Therefore, regardless of the flowering situation in the family, the Zerg army rushed to the mining area where the human race was located.

"Architecture is a bit good ..."

F91's troops pushed to the opponent's sub-mine, and some construction cards were a little uncomfortable in the middle, but after a while, they exclaimed again: "What is this hot buttock? This ... How can there be so many in his family? Troops? Troops like diarrhea? "

The shocked F91 was struggling with the strength of the enemy and failed to replace the opponent ’s mine, while the main force has lost two-thirds and the family has blossomed again.

Already unable to return to the sky, he took all the remaining troops A up to send light, giving himself a reason for GG.

After exiting, F91 did not qualify. He immediately opened the video and watched the game. From the beginning to the end, he watched the opponent ’s entire first-degree playback. The opponent ’s operations, violent forces, climbing technology, and offense included defense. and many more.

This ID has never seen before, and the operations and operations shown by him have made him extremely shocked, which is definitely the only level of top Korean masters.

Especially the carpet-type airdrop unit when the opponent attacked. At the moment of seeing that scene, he felt that this ID was the operation of the interstellar "person emperor" Boxer, and the carpet-type airdrop was the extreme operation he created.

In fact, F91 knew how to operate the carpet airdrop, but he couldn't do it at all, which was a flaw.

The carpet airdrop of the opponent's five ships just now requires the numbering of each transport plane separately, and then when landing the troops, they need to switch formations in turn with their left hand, and at the same time operate the mouse with their right hand to click on the soldiers in the transport plane. Complete all airdrops.

It's simple to say, it's actually extremely difficult, and ordinary players can't do it at all. To complete such an operation means that the mouse and keyboard need to be accurately and coherently operated with left and right hands 40 times in about three seconds. A sequence error has occurred.

If you just do that, then StarCraft will not be so difficult to play, and you have to run your home while doing so.

At this moment, F91's mind is full of question marks: When did domestic star players have such bad practices?

This level of operation has definitely reached the world's top level, including smooth operation and horrible violent abilities. If you do n’t play GG yourself, you are in confrontation, and you have watched the video yourself, F91 will definitely not believe that this is a domestic player, it will absolutely This is the vest of a top Korean expert.

After the shock, he continued to qualify.

I hope to be able to touch it again. Even the abused dish is very happy. For professional players, they are more than willing to fight against a master who is better than themselves, even if they are hanged by the opponent, because they can also improve.

However, he can no longer rank next. Any human player defeated by Nafu will not be ranked second ~ www.novelhall.com ~ After Nafu reaches the top in the country, "she" will enter the Korean battle network. , And fight against the world's top professional star players.

In this short day, Luo Sheng didn't know that the ID had suddenly caught fire in the domestic interstellar space.

The great **** players who became "victims" in various ways began to post in the community forum to ask which fairy was this sudden pop-out Nov, and actually blasted each master.

GG before the hand was returned.

The "victims" also gathered together to "complain" how they were blown up. At the same time, some videos of the battle with Nafu were also uploaded to the Blue Star video. Watch the video.

The ID called "Nov" ignited in the domestic interstellar circle almost overnight.

The topics in the circle are all discussed around this mysterious and powerful Terran player, and they are all trying to find a real person with this ID.

...