AlphaStar – artificial intelligence from the company DeepMind. In January, he had already defeated players in Starcraft II From Team Liquid, and now make sure that the cars will soon capture humanity, you can also – AlphaStar appeared in the rating mode of strategy.
As Blizzard writes, fights with AI will help DeepMind In research. To try your hand at the fight with AlphaStar, just press the “participate” button in a pop -up window in the game. Using the “DeepMind Help button in the dueling mode menu, you can change the choice. Opponents of the program will determine the standard matchmaking match.
AI will participate in matches anonymously – you will not find out, a person is opposed to you or a computer. This is necessary in order to ensure balanced testing conditions. The results of such games will affect the MMR users, but it seems that you should not say goodbye to the rating in advance – DeepMind consulted with cyber ends and introduced a number of restrictions into the program.
Below – a little details from the list of answers to the main questions:
• AlphaStar will play for terraces, grains and protosss on all regime maps one on one.
• after completing the tests DeepMind Lay out the recording of games from AI.
• Several versions of AlphaStar will participate in the experiment, which can use different strategies.
• AlphaStar will not become stronger during the testing – DeepMind does not use these matches as part of training.
Recall that development DeepMind https://nonukcasinosites.co.uk/review/mr-q-casino/ play not only in Starcraft II – The programs have already won people in the board game and the shooter Quake III.
July 10, 2019 Rumor: For Oculus VR, exclusive Splinter Cell and Assassin’s Creed will be developed
July 10, 2019 The first major update for The Division 2 will start July 23
Starcraft II: Wings of Liberty
Starcraft II: Legacy of the Void
Starcraft II: Heart of the Swarm
The best comments
AI will participate anonymously, but it can be calculated by nickname:
Kill All Humans
Sarah Coner Suck
Kiss My Shyni Metal Ass
Neo is a le
Just Bender etc.
As a studying neural network (but I am not a specialist, but an amateur) and the one who completely watched all the matches with AI, as well as their analysis (both by bun and the old man) I can show several problems with him at once.
Yes, he wins the fig. That in Dota, that in SK2, in battles, he demonstrated the magnificent ability of all his units to attack enemy units in the most optimal way. Each frame is monitored all distances, all CP, all skills and their rollbacks. And this is monitored on the entire map at a time (except for the fog of war, obviously). Therefore, such AI will calmly fight, even if they attack him at least ten sides. Including invisible opponents, because a separate metasprait was made for them (in the UK, I don’t know about Dota), seeing which the studied bot will immediately begin to build counter-lawns.
To limit this, the developers should not give the bot, for example, each frame is so quickly switched between different units that are located on different parts of the map. How to do this adequately? What does it mean "so fast"? People seem to have hot keys, they also carry out something similar (although not at such speeds and optimization of damage on enemies/themselves). In short, a separate problem.
Consider the construction in the UK. A specific moment: in the UK Alfastar tried to build a building, it began to be built, he tried to build exactly the same building on top of it, could not, canceled the order and then built the same building again. What is this? The lack of a plan of orders, so to speak, the order of actions and accounting "What I did/I want to do".
In addition, the bot sometimes destroyed some objects on the map, which does not make sense. For what? The bot learned to repeat the actions of the players, because he studied on them, yes. But he even began to repeat Misclies.
By the way, running in Dota: all 5 heroes sometimes ran up to that giant monster-bass that gives a lot of money (Soryan, I do not rummage). They did not focus on the fact that next to the point there is already their ally, respectively, there seems to be information about the enemy’s movements in the right area, but its bot does not take into account.
In short, it is necessary to think of architecture, it is necessary to inhibit humanity as a “description of the tactics and strategies that are clear to the computer” so that you can feed the data in an adequate quantity, but it’s terrible to do these cheaters.
This is a whole phrase, you’ll come up with a format so that it is easily translated into numbers, block diagrams, optimization task, build a mat. Model, then we’ll talk. Or find a study, I even read it myself.
And what a camera he will have, ordinary or infinitely high? I remember when he was cut the camera, he lost about the player.
