paper-with-me

홈 › Papers

Maximum Expected Hitting Cost of a Markov Decision Process and Informativeness of Rewards

2019-12-01 · NeurIPS 2019 12 · Falcon Dai, Matthew Walter

We propose a new complexity measure for Markov decision processes (MDPs), the maximum expected hitting cost (MEHC). This measure tightens the closely related notion of diameter [JOA10] by accounting for the reward structure. We show that this parameter replaces diameter in the upper bound on the optimal value span of an extended MDP, thus refining the associated upper bounds on the regret of several UCRL2-like algorithms. Furthermore, we show that potential-based reward shaping [NHR99] can induce equivalent reward functions with varying informativeness, as measured by MEHC. By analyzing the change in the maximum expected hitting cost, this work presents a formal understanding of the effect of potential-based reward shaping on regret (and sample complexity) in the undiscounted average reward setting. We further establish that shaping can reduce or increase MEHC by at most a factor of two in a large class of MDPs with finite MEHC and unsaturated optimal average rewards.

📄 PDF Abstract BibTeX

Code (0)

등록된 구현이 없습니다.

Tasks

Informativeness

Similar Papers 제목 키워드 기반

Maximum Expected Hitting Cost of a Markov Decision Process and Informativeness of Rewards

2019-07-03 · Falcon Z. Dai, Matthew R. Walter

We propose a new complexity measure for Markov decision processes (MDPs), the maximum expected hitting cost (MEHC). This measure tightens the closely related notion of diameter [JOA10] by accounting for the reward struct…

Informativeness

On Reward Structures of Markov Decision Processes

2023-08-28 · Falcon Z. Dai

A Markov decision process can be parameterized by a transition kernel and a reward function. Both play essential roles in the study of reinforcement learning as evidenced by their presence in the Bellman equations. In ou…

reinforcement-learningReinforcement LearningSafe Reinforcement Learning

Hitting time for Markov decision process

2022-05-06 · Ruichao Jiang, Javad Tavakoli, Yiqinag Zhao

We define the hitting time for a Markov decision process (MDP). We do not use the hitting time of the Markov process induced by the MDP because the induced chain may not have a stationary distribution. Even it has a stat…

Imitation Learning

A Unified Markov Chain Approach to Analysing Randomised Search Heuristics

2013-12-09 · Jun He, Feidun He, Xin Yao

The convergence, convergence rate and expected hitting time play fundamental roles in the analysis of randomised search heuristics. This paper presents a unified Markov chain approach to studying them. Using the approach…

Hitting Time Isomorphism for Multi-Stage Planning with Foundation Policies

2026-05-07 · Magnus Victor Boock, Abdullah Akgül, Mustafa Mert Çelikok, Melih Kandemir arxiv

We present a new operator-theoretic representation learning framework for offline reinforcement learning that recovers the directed temporal geometry of a controlled Markov process from hitting time observations. While p…

Representation LearningReinforcement Learning