(Deep) Reinforcement Learning for Dynamic Routing Problems / Introducing and collaborating with the BDSi

Title: (Deep) Reinforcement Learning for Dynamic Routing Problems - given by Fabian Akkerman

• Abstract: Suppose you control two agents (let us call them “Pacmen”) on a rectangular 6x6 grid. Each Pacman may take 1 step vertically or horizontally each round. Every round, with certain probability, tokens (“cherries”) appear on the grid that can be eaten by the two Pacmen. The objective is to eat cherries as quickly as possible, or equivalently, to have as few cherries as possibly on average on the grid. How would you control the Pacmen?
  For simple problem structures (1 Pacman on a small grid) it is easy to obtain fast and (close-to-) optimal solutions using exact models or heuristics. However, when the complexity increases (>1 Pacman on a larger grid) it becomes harder to obtain good solutions, and more complex heuristics are needed. Since the multiple-Pacman problem requires cooperation of the different agents, a reinforcement learning approach might be beneficial. Although our approach is tested on the pacman “game”, this game can be easily transformed to relevant problems in practice, e.g. pick-up and delivery problems. We show how our reinforcement learning algorithm yields, without too much tuning, considerable improvement compared to heuristics.
• Bio: Fabian Akkerman is a PhD candidate within the section of Industrial Engineering and Business Information Systems (IEBIS. He holds a master in Industrial Engineering and Management from the University of Twente (2021). His research involves the development of DynaPlex; an artificial intelligence toolbox containing various (deep) reinforcement learning algorithms that can be applied to a variety of dynamic data-driven logistics challenges. The DynaPlex project is in collaboration with the Eindhoven University of Technology and a wide range of companies.

Title: Introducing and collaborating with the BDSi - given by Karel Kroeze, Anna Machens, and Abhishta Abhishta 

• Abstract: The goal for this meeting is for it to be a conversation, so we’ll try to keep the boilerplate to a minimum, and get straight to the point – how can we help you use data science to improve your research and education. This depends a lot on what you’re currently doing, and what you would like to be doing, so it would helpful if you could think about the following questions before the meeting;
        - What statistical/methodological/research techniques are you currently using?
        - Do you feel have mastered these techniques?
        - Do you know where to go if you have questions?
        - What problems do you have using these techniques?
        - What new ‘data science’ techniques you would like to be using?
        - What is stopping you from trying these techniques?
        - Do you have problems you don’t know how to solve, or questions you cannot
          answer using your current research toolbox?
  During the meeting we’ll ask you these questions, and try to answer any questions you might have. By the end of the meeting, we hope that you have a better idea of type of data science applications that BDSi can help you with.
  At the same time, we hope to have a better idea of the methods you’re currently using and would like to use, so that we can help build expertise and communities around these methods for the benefit of us all! 
• Bio: You can check the speakers' bio at: https://bdsi.bms.utwente.nl/team/