Reinforcement Learning and Optimization in Stochastic Multi-objective Environments



In real-life, we are very often confronted with selecting one of several options where the resulting pay-off is stochastic and assumed to be unknown. However, in many applications, multiple objectives should be taken in account. We want to combine methods of machine learning (ML) and Multi-Objective Optimization (MOO) in order to learn and/or search difficult multi-objective environments that are possibly dynamic, uncertain and partially observable.

We are interested in exploring the potential synergies between reinforcement learning (RL), which is a well-established sequential decision Machine Learning (ML) problem, and Multi-objective optimization, which is an sub-area of multi-criteria decision making (MCDM). We consider the extension of RL to multi-criteria stochastic rewards (also called utilities in decision theory). The most successful hybrid algorithms are also motivated and validated on real-world problems.


Aim and scope.

The main goal of this special session is to start the process of unifying and streamlining research on learning and optimization in multi-objective stochastic environments which for time being seems to evolve independently and disconnected in Reinforcement learning and Multi-criteria decision making. We are considering machine-learning algorithms that are both theoretically and practically motivated. We want to bring together researchers from machine learning, optimization and artificial intelligence, interested learning and optimization in multi-objective stochastic environments. We also encourage submissions related to stochastic multi-objective optimization in other areas such as operation research, games and real-world applications.

Ideally, the special session will help researches with different background in Reinforcement Learning and Multi-objective Optimization to identify some common ground for their work.


Topics of interest.

Topics of interests include but are not limited to

- Multi-objective reinforcement learning

- Multi-objective optimization algorithms such as metaheuristics, evolutionary algorithms, etc. for stochastic environments

- Theoretical results on the learnability in multi-objective stochastic environments

- Novel algorithmic frameworks for multi-objective stochastic environments

- Multi-criteria aspects of robotics

- Multi-objective self-adapting systems

- Multi-objective automatic configuration systems

- Multi-objective games

- Real-world applications in engineering, business, computer science, biological sciences, scientific computation, etc. in Stochastic Multi-objective Environments

- Multi-criteria dynamic/reactive scheduling and planning


Important dates.

  Paper submission: 15 June 2014

  Notification to authors: 5 September 2014

  Final papers due: 5 October 2014

  Early registration: 5 October 2014

  Conference: 9-12 December 2014


Program Committee.

  • Lucian Busoniu
  • Richard Dazeley
  • Yann-Michaël De Hauwere
  • Madalina Drugan
  • Bernard Manderick
  • Diederik Roijers
  • Peter Vamplew
  • Peter Vrancx
  • Marco Wiering
  • Logan Yliniemi



Madalina M. Drugan

Artificial Intelligence Lab, Vrije Universiteit Brussels, Belgium.


Bernard Manderick

Artificial Intelligence Lab, Vrije Universiteit Brussels, Belgium.


Yann-Michaël De Hauwere

Artificial Intelligence Lab, Vrije Universiteit Brussels, Belgium.