AI Paper Atlas

Research map / Reinforcement learning / MDPs and planning

Learning and Planning for Time-Varying MDPs Using Maximum Likelihood Estimation

Melkior Ornik, Ufuk Topcu

JMLR 2021 · Reinforcement learning › MDPs and planning · approach: Robust MDP policy gradient · 9 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use CCMLE instead of classical estimation (slow change, exp a)transition probability estimator · Gridworld patrolling exp (a)mean error: 0.11 -> 0.03Improved
use CCMLE instead of classical estimation when change is faster than the assumed bound (0.02 per step)transition probability estimator · Gridworld patrolling exp (c)mean error: 0.12 -> 0.06Improved
use CCMLE instead of classical estimation for a sudden switchtransition probability estimator · Gridworld patrolling exp (d)mean error: 0.16 -> 0.1Improved
add forgetfulness to CCMLE (ablation)CCMLE · Two-state periodic TVMDPmean error: 0.06 -> 0.05Improved
add forgetfulness (sliding window) to classical estimation (ablation)classical estimation · Two-state periodic TVMDPmean error: 0.28 -> 0.14Improved
use CCMLE-based planning instead of classical estimation planningpatrolling agent · Gridworld patrolling planning, t=200walls visited: 18 -> 38Improved
add uncertainty-based learning bonus (beta>0) to CCMLE planning (ablation)CCMLE planner · Two-state MDP planningasymptotic average reward: 1 -> 1.15Improved
increase beta to a very high value (ablation)CCMLE planner · Two-state MDP planningasymptotic average reward: 1.15 -> 0.88Hurt
use forgetful CCMLE with uncertainty bonus instead of classical estimation in 5-arm banditbandit agent · Time-varying 5-armed bandit, 10000 stepsaverage collected reward: 25% higherImproved
use CCMLE instead of classical estimation with misspecified bound epsilontransition probability estimator · Wind flow estimationaverage of maximal error: 0.4 -> 0.21Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy