New Q-learning algorithm offers model-free reachability in MDPs

Researchers have developed Quasar, a novel model-free algorithm that uses Q-learning to achieve asymptotic convergence for reachability specifications in Markov Decision Processes (MDPs) that are free of non-terminal maximal end components (MECs). This approach eliminates the need to explicitly estimate transition probabilities, a requirement of previous model-based methods. Quasar significantly reduces memory footprint and demonstrates faster convergence on the Quantitative Verification Benchmark Set compared to existing state-of-the-art model-based techniques, marking a practical step towards specification-guided reinforcement learning. AI

IMPACT Introduces a more memory-efficient and sample-efficient method for specification-guided reinforcement learning, potentially enabling broader applications.

RANK_REASON This is a research paper detailing a new algorithm for reinforcement learning in Markov Decision Processes. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Markov decision process
• MEC-Free MDPs
• Q-learning
• Quantitative Verification Benchmark Set
• Quasar
• reinforcement learning

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →