Discounted MDP's: distribution functions and exponential utility maximization
SIAM Journal on Control and Optimization
Risk sensitive control of Markov processes in countable state space
Systems & Control Letters
Risk-Sensitive Control of Finite State Machines on an Infinite Horizon I
SIAM Journal on Control and Optimization
Risk-Sensitive Control of Finite State Machines on an Infinite Horizon II
SIAM Journal on Control and Optimization
Neural Computation
Risk-sensitive reinforcement learning applied to control under constraints
Journal of Artificial Intelligence Research
Brief On terminating Markov decision processes with a risk-averse objective function
Automatica (Journal of IFAC)
Safe exploration of state and action spaces in reinforcement learning
Journal of Artificial Intelligence Research
Hi-index | 22.15 |
This paper analyzes a connection between risk-sensitive and minimax criteria for discrete-time, finite-state Markov decision processes (MDPs). We synthesize optimal policies with respect to both criteria, both for the finite horizon and the discounted infinite horizon problem. A generalized decision-making framework is introduced, which includes as special cases a number of approaches that have been considered in the literature. The framework allows for discounted risk-sensitive and minimax formulations leading to stationary optimal policies on the infinite horizon. We illustrate our results with a simple machine replacement problem.