Artificial intelligence for logistics and its interfaces

1CM240

About this course

In this challenge-based course, you will learn how and when to use machine-learning techniques to support operational decisions in application areas such as transportation, inventory and production management, i.e., real-time operational decision making environments. The main focus lies on teaching you how to apply reinforcement learning to operational problems. That means you will learn the fundamental building blocks of modern-day reinforcement learning such as Neural Nets (Deep Learning) and Markov Decision Processes, but on-top we will provide you with state-of-the art applications and real-world examples of the learned technologies.

In the course, you will be confronted with a company facing challenges in the domains of transportation and inventory management. Using known techniques from operations management, domain knowledge, and the new AI techniques (in particular Reinforcement Learning), it is your goal to improve upon the company’s current performance. In this course, we guide you through all the steps of achieving this.
The company's current approach will be provided to you as a working computer program in Python. We provide you the environment, called ‘Gym’, which is a simulation of the challenge under consideration and a basic implementation of the ‘Trainer’, which takes decisions based on the state of the Gym. In addition, the structure for the communication between Gym and Trainer (e.g., reward function, termination criteria) is provided. This ensures that you will be handed a completely working program imitating the company’s current practice.

In this course, the focus will be on improving the performance of the trainer and interpreting the results in the context of the operational challenge. In other words, it is your job to better describe, predict, and prescribe, the uncertainty and the actions (or decisions) that should be taken. The challenges combine modern dynamic and stochastic transportation management (e.g. city logistics, same-day delivery), and relevant inventory problems (global sourcing, re-shoring, capacitated supply chains). The challenges will be of a stochastic and dynamic nature and will be addressed from an problem perspective.

The course is performed in small teams. You will be graded based on two parts: an exam and a report. The exam evaluates your capabilities to model stylized operational problem and to solve them with AI methods. For the report, you need to solve the company’s problem where you need to decide yourselve on the solution approach. Evaluation of the report is based on the program developed, the report, and potentially discussions.

Learning outcomes

In this course, you will learn how to address real-time decision making in applications in Operations Management (for instance Transportation, Inventory, and Production Management). Such so-called stochastic sequential decision making problems arise everywhere in practice, and have been subject of study in the Industrial Engineering domain for decades. Recently, classical optimization methods are making increasingly use of data, opening up the wide area of Artificial Intelligence (AI). In this course, we teach you when and how these AI techniques can help decision making in real-time operational environments. To achieve this, we provide you with lectures, tutorials and online resources that guide you on the when’, why’, and ‘how’ of hybrid optimization techniques encapsulating AI and optimization.
Upon successful completion of the course, you will be able to:

  • Model stochastic sequential optimization problems in operations management as Markov Decision Processes
  • Trade off the usefulness of different data-driven optimization techniques (supervised, unsupervised and reinforcement learning) with traditional optimization methods.
  • Decide which data-driven optimization method to use dependent on the problem
  • Independently implement problem-specific data-driven optimization techniques (e.g. reinforcement learning) to operational settings
  • Analyze and critically interpret output generated by data-driven optimization techniques

Prior knowledge

You must meet one of the following collections of requirements

  • Collection 1
  • Completed Final examination Bsc program
  • Collection 2
  • Completed Pre-Master

Resources

  • Materials will be provided at the beginning of the course.

Additional information

  • Credits
    ECTS 5
  • Level
    master
If anything remains unclear, please check the FAQ of TU Eindhoven.

Offering(s)

  • Start date

    3 February 2025

    • Ends
      6 April 2025
    • Term *
      Block GS3
    • Location
      Eindhoven
    • Instruction language
      English
    Enrolment starts in 36 days
For guests registration, this course is handled by TU Eindhoven