Summary #900

AlphaStar: Grandmaster level in StarCraft II using multi-agent reinforcement learning

deepmind.com
  • AI agent AlphaStar has been trained, using novel techniques, to be better than 99.8% Starcraft II players, and achieve Grandmaster level for all 3 races in the game.
  • The AI uses a camera interface, it’s limited to receiving similar input information, and playing at a similar speed to human players, and plays anonymously on Battle.net.
  • One of the key innovations was using a League self-play system, which involved flaw-exposing agents alongside win-oriented agents.
  • Other techniques included neural networks, multi-agent learning and imitation learning.
  • The study helped improve those methods by adapting them to a complex, dynamic environment, contributing to the safety and robustness of AI systems in general.

Categories

  • Science Science (53) AI Copernicus ‘discovers’ that Earth orbits the Sun
  • Psychology Psychology (53) Uncertainty and Surprise Jointly Predict Musical Pleasure and Amygdala, Hippocampus, and Auditory Cortex Activity
  • Cosmos Cosmos (25) Kilometer-Long Space Tether Tests Fuel-Free Propulsion
  • History History (14) Humans survived off rodents in the mountains during the last ice age, study says
  • Society Society (59) Majority of anti-vaxx ads on Facebook are funded by just two organizations
  • Medicine Medicine (53) Antibody injection stops peanut allergy for 2 to 6 weeks, study shows
  • Technology Technology (61) Hologram-like device animates objects using ultrasound waves
  • Nature Nature (39) Is a long-dormant Russian volcano waking up? It’s complicated

Newsletter

Newsletter

Fame 🙌

Bullets.tech - Articles for science lovers shortened to five bullet points. | Product Hunt Embed

We were featured on Hacker News, O'REILLY® Ideas, and Boing Boing.