https://padlet.com/jananjeet_bali/uo2gq3raydp6 From mastering chess with AlphaZero and then the extremely complex game of Go with AlphaGo, the Deepmind team now took the nearly impossible challenge of creating an AI that plays a real time strategy game and beat the top professional players. en-us 2019-03-20 12:51:40 UTC 2026-03-30 07:16:36 UTC hello@padlet.com https://padlet-assets.s3.amazonaws.com/icons/Playcontrol.png jananjeet_bali https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/344781921 Unlike games like chess or Go where players see everything, crucial information is hidden from a StarCraft player and must be actively discovered by “scouting”.
So the AI is capable of adapting to human behavior and unpredictability, which can have several applications in solving problems outside of just StarCraft II]]> 2019-03-25 12:38:25 UTC https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/344781921 jananjeet_bali https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/345501007 AlphaStar has to make a lot of decisions that may not have immediate effects. It has to make predictions of future moves that the player could make and adapt to said circumstances. This technology can have a lot of use outside of just SC2.

]]> 2019-03-27 00:10:00 UTC https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/345501007 jananjeet_bali https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/345528938 2019-03-27 02:40:57 UTC https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/345528938 jananjeet_bali https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/345533191 The neural network architecture of AlphaStar is capable of modelling long sequences of actions based on imperfect information. The fundamental problem of making complex predictions over very long sequences of data appears in many real world challenges, such as weather prediction, language interpretation and more. Significant advances can be made in these domains using the developments from the AlphaStar project.]]> 2019-03-27 03:11:39 UTC https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/345533191 jananjeet_bali https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/346896585 2019-03-31 14:42:48 UTC https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/346896585 jananjeet_bali https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/346896765 StarCraft II is a real time strategy game, usually played between two players. Players can play 3 different races, the human race Terran, the parasitic alien race Zerg and the technologically advanced alien race Protoss. Players have to train units at the start of every game to collect resources, train armies and destroy the opponent's base. ]]> 2019-03-31 14:44:26 UTC https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/346896765 jananjeet_bali https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/346897578 AlphaStar mastered SC2 using a deep learning network that was trained initially by observing human game replays, but the bulk of the training was done by playing a series of matches against itself and learning through a process similar to evolution in nature, where the best traits are passed onto the next generation.
The AlphaStar agent currently only plays Protoss vs Protoss matches, but in the future Deepmind is looking towards training an agent that is capable of handling all 9 match-ups.]]> 2019-03-31 14:51:32 UTC https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/346897578 jananjeet_bali https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/346899667 AlphaStar completed its training over a period of 2 weeks, and over those 2 weeks it played over 200 years worth of matches.
After those 2 weeks, this AI beat one of the best Protoss players MaNa in a Protoss vs. Protoss match.
The AI did not achieve this proficiency by being able to make more actions than the human players either. The average actions per minute (APM) of the AlphaStar agent was 280, which is significantly lower than the APM of professional players like MaNa who have an average APM of 400+.]]> 2019-03-31 15:08:49 UTC https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/346899667 jananjeet_bali https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/346903860 This process of training deep learning networks to play video games can be crucial to play testing video games and help with finding bugs in the game development life cycle.]]> 2019-03-31 15:42:11 UTC https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/346903860 jananjeet_bali https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/346905240 For learning more about the AlphaStar project and other projects by Deepmind visit their official website.]]> 2019-03-31 15:52:04 UTC https://padlet.com/jananjeet_bali/uo2gq3raydp6/wish/346905240