# Solving Multi-Model MDPs by Coordinate Ascent and Dynamic Programming

arXiv: https://arxiv.org/abs/2407.06329 · HTML: https://mlsanity.com/paper/2407.06329/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 5 of its 41 references that are in the map.
- First posted on arXiv 8 Jul 2024.

## Builds on (most cited first, 5 of 5)

- [On Linear Convergence of Policy Gradient Methods for Finite MDPs](https://mlsanity.com/paper/2007.11120/) (2020)
- [Partial Policy Iteration for L1-Robust Markov Decision Processes](https://mlsanity.com/paper/2006.09484/) (2020)
- [Beyond Confidence Regions: Tight Bayesian Ambiguity Sets for Robust MDPs](https://mlsanity.com/paper/1902.07605/) (2019)
- [Soft-Robust Actor-Critic Policy-Gradient](https://mlsanity.com/paper/1803.04848/) (2018)
- [Thompson Sampling with a Mixture Prior](https://mlsanity.com/paper/2106.05608/) (2021)

## Built on by (most cited first, 2 of 2)

- [Risk-averse Total-reward MDPs with ERM and EVaR](https://mlsanity.com/paper/2408.17286/) (2024)
- [Risk-Averse Total-Reward Reinforcement Learning](https://mlsanity.com/paper/2506.21683/) (2025)

## Abstract (arXiv)

Multi-model Markov decision process (MMDP) is a promising framework for computing policies that are robust to parameter uncertainty in MDPs. MMDPs aim to find a policy that maximizes the expected return over a distribution of MDP models. Because MMDPs are NP-hard to solve, most methods resort to approximations. In this paper, we derive the policy gradient of MMDPs and propose CADP, which combines a coordinate ascent method and a dynamic programming algorithm for solving MMDPs. The main innovation of CADP compared with earlier algorithms is to take the coordinate ascent perspective to adjust model weights iteratively to guarantee monotone policy improvements to a local maximum. A theoretical analysis of CADP proves that it never performs worse than previous dynamic programming algorithms like WSU. Our numerical results indicate that CADP substantially outperforms existing methods on several benchmark problems.