Fast Learning of Small Strategies - Jan Křetínský, Technical University of Munich
Alan Turing Institute via YouTube
Overview
Explore a 39-minute conference talk on merging verification and machine learning approaches to analyze Markov decision processes. Learn how to quickly develop ε-optimal strategies and represent them concisely for better understanding and debugging. Discover the speaker's method for combining the precision of verification algorithms with the scalability of machine learning techniques. Gain insights into the application of this approach to complex systems analysis, formal methods development, and practical problem-solving. Understand the structure of the talk, which covers topics such as degrees of freedom, machine learning, reinforcement learning, and defining important decisions, concluding with an example and final thoughts.
Syllabus
Intro
Degrees of freedom
Hero
Machine Learning
Reinforcement Learning
Markov Decision Processes
Compute
Frequent visits
Scheduling
Defining important decisions
Example
Conclusion
Taught by
Alan Turing Institute