Thursday , September 24 2020

[표지로 읽는 과학] StarCraft II: conquered AI



[ad_1]

The nature provided

International journal Nature published a study on the 14th of this month that the AI ​​game "Alphastar" developed by Google's research team, Deep Mind Google, such as researcher Orion Binals, reached the level of a grandmaster at Star Craft II. . The game's cover. The nature provided

Artificial Intelligence (AI) won the & # 39; Starcroft & # 39; after the chess and Go. StarCraft is the longest running science-fiction (RTS) real-time game.

International journal Nature published a study on the 14th of this month that the AI ​​game "Alphastar" developed by Google's research team, Deep Mind Google, such as researcher Orion Binals, reached the level of a grandmaster at Star Craft II. . The game's cover. Grandmaster levels are among the top 0.2%.

Chess and Gambling are played by certain rules. So when AI learns a lot of existing patterns, it predicts and responds to the number of cases the opponent will put in the right game. But games are more complicated than Go or Chess, and more complicated than two or more players. In particular, a strategy game like StarCraft is not easy, as it foresees the opponent's strategy, and at the same time, devises my strategy so that the opponent does not notice it.

AlphaStar is composed of a synthetic artificial neural network (CNN) that analyzes visual images and a circular artificial neural network (LSTM) that can be predicted using past data. Just as one looks at the game screen, he realizes the situation, anticipates his opponent's strategy, and develops his own tactics. We've learned more than 800,000 game play data.

In addition, DepMind's team trained AlfaStar with 16 different AI algorithms and 14-day iterations without human intervention. In a human sense, it's like playing a game for about 200 years.

In January, Alfastar hit a 10-1 victory over British pro gamers Darius Dance (TLO) and Gregory Cominis (Mana). At the time, experts said Alfastar was a little more favorable because it played games with different interfaces.

DeepMind has since upgraded AlphaStar to allow users to play games with the same interface. And at the end of October, Alfastar reached the level of Grand Master in all three races, including Teran, Proto and Zerg.

[ad_2]
Source link