Optimal Policies Tend To Seek Power.
Alexander Matt TurnerLogan SmithRohin ShahAndrew CritchPrasad TadepalliPublished in: NeurIPS (2021)
Keyphrases
- optimal policy
- markov decision processes
- decision problems
- dynamic programming
- finite horizon
- reinforcement learning
- infinite horizon
- long run
- sufficient conditions
- average reward
- state space
- finite state
- state dependent
- multistage
- bayesian reinforcement learning
- dynamic programming algorithms
- average cost
- partially observable markov decision processes
- initial state
- markov decision process
- policy iteration
- serial inventory systems
- average reward reinforcement learning
- machine learning
- control policies
- markov decision problems
- expected cost
- semi markov decision processes
- reinforcement learning algorithms
- model free
- computational complexity