1-2-3-Go! Policy Synthesis for Parameterized Markov Decision Processes via(…)
summary
Generalizes optimal policies from model-checked small instances to huge parameterized MDPs by training and pruning decision-tree learning models that predict actions from state features, avoiding explicit state-space exploration. Uses Storm and standard model checkers to generate training labels, then evaluates on the quantitative verification benchmark set; learned policies achieve near-optimal reachability probabilities while scaling to models orders of magnitude larger than what explicit analysis fits in memory. Experiments report successful generalization and practical runtimes rather than exact optimality.
AI-written summary · about
related talks
57mFundamentals of Markov Decision Processes
55mOptimality and Approximation with Policy Gradient Methods in Markov Decision Processes
43mModel Checking and Strategy Synthesis for Stochastic Games: From Theory to Practice
17mProgrammatic Strategy Synthesis: Resolving Nondeterminism in Probabilistic Progr...
42mReinforcement Learning using Generative Models for Continuous State and Action Space Systems
14mSequential Monte Carlo Program Synthesis with Refinement Proposals
53mOff-policy Policy Optimization
15mSequential Decision Making Problems in Multi-Agent Environments