AI Paper Atlas

Research map / Reinforcement learning / Policy gradient and actor-critic

Risk-Averse Learning by Temporal Difference Methods with Markov Risk Measures

Umit Köse, Andrzej Ruszczyński

JMLR 2021 · Reinforcement learning › Policy gradient and actor-critic · approach: Temporal-difference and discounting theory

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 risk-averse TD(lambda) (mean-semideviation) instead of risk-neutral TD(lambda)TD(lambda) for fleet-assignment MDP · Fleet management, 20,000 stages, alpha 0.95/0.8/0.6, lambda 0/0.5/0.9average profit per stageMixed
use a large lambda (0.9) (ablation)optimistic risk-averse TD(lambda) · Fleet management, 20,000 stagesaverage profit per stageHurt
use risk-averse instead of risk-neutral TD at early training (t=200)TD(lambda) · Fleet management, 207 trajectories, alpha 0.95empirical distribution of profit per stage at t=200Improved
use risk-averse instead of risk-neutral linear policy evaluationlinear TD policy evaluation · Small fleet (K=12, M=3, 91 states), myopic policymean-square error vs full evaluation: risk-neutral final error twice as large as risk-averseImproved
use the policy from risk-averse linear value approximation instead of the myopic policymyopic policy · Small fleet, one policy improvement stepaverage profitImproved
use the policy from risk-neutral linear value approximation instead of the myopic policymyopic policy · Small fleet, one policy improvement stepaverage profitHurt

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