Royal Melbourne Institute of Technology researchers created new algorithms that learned to play Atari game like humans 10 times faster than Google’s DeepMind AI

Briefing

Royal Melbourne Institute of Technology researchers created new algorithms that learned to play Atari game like humans 10 times faster than Google’s DeepMind AI

February 12, 2019

Briefing

  • Atari-Playing AI – Royal Melbourne Institute of Technology (RMIT) researchers developed new algorithms that can autonomously play 1980s Atari game Montezuma’s Revenge
  • 10 Times Faster – Algorithms learned from mistakes and identified sub-goals, such as climbing ladder, jumping over pit, and picking up key, 10 times faster than Google’s DeepMind AI
  • Combination of Techniques – AI combines reinforcement learning (i.e. technique that provides rewards to AI for certain actions) with intrinsic motivation approach, which encourages it to explore environment
  • Other Applications – Include self-driving cars and robotic assistants with natural language recognition

Accelerator

Sector

Information Technology

Organization

RMIT University

Source

Original Publication Date

January 31, 2019

Leave a comment