Advancements in Markov Decision Processes and Bayesian Optimization

The field of Markov decision processes (MDPs) and Bayesian optimization is witnessing significant developments, with a focus on addressing complex problems with uncertain parameters and constraints. Researchers are exploring novel approaches to mitigate epistemic uncertainty, such as Bayesian-risk MDPs and coherent risk measures. Additionally, there is a growing interest in online and episodic settings, where algorithms must adapt to changing conditions and constraints. Noteworthy papers in this area include: Policy Gradient Optimzation for Bayesian-Risk MDPs with General Convex Losses, which proposes a policy gradient optimization method for Bayesian-risk MDPs. Beyond Slater's Condition in Online CMDPs with Stochastic and Adversarial Constraints, which provides a novel algorithm for online episodic Constrained MDPs with improved guarantees.

Sources

Policy Gradient Optimzation for Bayesian-Risk MDPs with General Convex Losses

Optimal Experimental Design of a Moving Sensor for Linear Bayesian Inverse Problems

Analysis of approximate linear programming solution to Markov decision problem with log barrier function

Beyond Slater's Condition in Online CMDPs with Stochastic and Adversarial Constraints

Efficient Multi-Objective Constrained Bayesian Optimization of Bridge Girder

Built with on top of