IEEE CoG 2019 - Industry Day - Katja Hofmann (Microsoft Research)
summary
Microsoft-built Project Malmo platform exposes Minecraft as an open-source 3D RL lab (Windows/Linux/Mac, many languages) and already appears in 100+ papers; it runs simple tabular Q-learning (agent learns a cliff-walk policy after ~10 minutes) up to deep visual policies. A competition stack centers on MINERL: obtain a diamond with agents retrained from scratch, limited to four days of training on one machine and encouraged to use 60 million state-action pairs of human demonstrations; baselines and Docker submission flow are provided.
AI-written summary · about
chapters (9)
- 0:00Introduction and motivation
- 2:33Project Malmo platform
- 8:12Reinforcement learning demo
- 16:01Human-AI collaboration study
- 27:52AI competitions overview
- 31:48Mineral competition details
- 40:01Meta learning research
- 47:46Future research directions
- 49:51Closing and acknowledgments
related talks
01h 09mMinecraft as AI Playground and Laboratory
28mAI and Minecraft Lars
41mReinforcement Learning and Minecraft
32mAI and Minecraft
05mInterview with
49mA Raspberry Pi and DeepLearning4J Brain That Plays Minecraft
50mA Raspberry Pi and DeepLearning4J Brain That Plays Minecraft
26mChatGPT and Minecraft's diamonds playing agents: Reinforcement Learning in a nutshell with sheeprl
