EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
2344. Online Learning Linear Programming Heuristic for Weakly Coupled MDP
Invited abstract in session TB-40: Reinforcement Learning: Methods and Applications , stream Advances in Stochastic Modelling and Learning Methods.
Tuesday, 10:30-12:00Room: 96 (building: 306)
Authors (first author is the speaker)
| 1. | Alexandre Reiffers
|
Abstract
In this talk, we present novel algorithms for online learning of linear programming heuristics for weakly coupled Markov Decision Processes (MDPs). Leveraging two-timescale stochastic approximation theory, we design simple and efficient algorithms with theoretical guarantees. Furthermore, we show that we can extend our approach to learn policies for continuous state space weakly coupled MDPs, while also demonstrating that certain theoretical guarantees are preserved. Finally, we demonstrate the efficiency of our algorithms on real-world applications, including load-balancing coupled with autoscaling in parallel queues, and monitoring strategy of slowly varying signals in networks.
Keywords
- Optimal Control
- Machine Learning
- Queuing Systems
Status: accepted
Back to the list of papers