On tight bounds for function approximation error in risk-sensitive reinforcement learning

DOI10.1016/J.SYSCONLE.2021.104899MaRDI QIDQ2243003zbMATH OpenOpenAlexFDO

Authors Prasenjit Karmakar, Shalabh Bhatnagar

Publication date 10 November 2021

Published in Systems \& Control Letters (Search for Journal in Brave)

Full work available at URL https://doi.org/10.1016/j.sysconle.2021.104899

zbMATH Keywords

stochastic optimal controlstochastic systemsfunction approximationeigenvalue perturbationPerron-Frobenius eigenvaluerisk-sensitive reinforcement learning

Mathematics Subject Classification ID

Learning and adaptive systems in artificial intelligence (68T05) Decision theory (91B06) Optimal stochastic control (93E20)

Recommendations

• A learning algorithm for risk-sensitive cost
• Advances in Artificial Intelligence
• Risk-sensitive reinforcement learning
• Performance Loss Bounds for Approximate Value Iteration with State Aggregation
• Error bounds for approximations from projected linear equations

Cites work

• A quantitative comparison of risk measures
• A sensitivity formula for risk-sensitive cost and the actor-critic algorithm
• Comparing the Spectral Radii of Two Nonnegative Matrices
• Inequalities. Theorems, techniques and selected problems
• Markov decision processes with average-value-at-risk criteria
• Multiplicative ergodicity and large deviations for an irreducible Markov chain.
• On comparison of the Perron-Frobenius eigenvalues of two ML-matrices
• On the Maximal Eigenvector of a Positive Matrix
• Optimization of risk measures
• Q-Learning for Risk-Sensitive Control
• Risk-Sensitive Optimal Control for Markov Decision Processes with Monotone Cost
• Risk-averse dynamic programming for Markov decision processes
• Simulation-based optimization of Markov reward processes
• Stochastic approximation with `controlled Markov' noise

Cited in

(5)

• Estimation and approximation bounds for gradient-based reinforcement learning
• Robust bounds on risk-sensitive functionals via Rényi divergence
• On the sample complexity of actor-critic method for reinforcement learning with function approximation
• Advances in Artificial Intelligence
• Data-driven direct adaptive risk-sensitive control of stochastic systems

This page was built for publication: On tight bounds for function approximation error in risk-sensitive reinforcement learning

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2243003)