• View
• Edit
• History
• Print

Reinforcement Learning

Issues:

• State Space explosion or having to interpolate / grow states dynamically.
• State descriptions and their adaptability
• The markov assumption
• The assumption of stationary preference.
• The linear computation time of policies. How do we handle hierarchical learning & developmental learning in RL / MDPs?

Questions:

• What do we want to do with these agents? What is the ultimate goal?
• What are some realistic and stretch scenarios for RL to handle?
• What is the most complex RL system so far?