Kies de Nederlandse taal
Course module: 202100226
Reinforcement learning in Engineering
Course info
Course module202100226
Credits (ECTS)5
Course typeCourse
Language of instructionEnglish
Contact B. Rosic
Lecturer B. Rosic
Contactperson for the course B. Rosic
Academic year2021
Starting block
Application procedureYou apply via OSIRIS Student
Registration using OSIRISYes
After this course the student is able:
  • to formalize engineering problems as Markov decision processes
  • to construct predictive models given data sets by using analytical functions or complex predictive models based on the deep learning concept
  • to implement dynamic programming to find optimal policy based on the value function
  • to analyze and validate learning approaches
  • to synthesize results of the analysis into a meaningful conclusion and to evaluate the quality of data
Industrial systems are becoming more complex, and hence more difficult to control or optimize in a physics-driven manner. Recent shift made towards the data-driven approach is requiring new types of skills from engineers. This course aims to bridge this gap and bring the data-driven expertise to a non-expert from both a theoretical and practical perspective. In particular, the focus will be on reinforcement learning as one of promising tools for automatic decision making. In this course one will study what reinforcement learning means and its qualitative relationship to the optimal control and stochastic optimization theory. Furthermore, the basic engineering problems will be formalized in a Markov Decision setting, and the concepts of value function, agent and policy will be explained. In this light, the difference and significance of exploration and exploitation strategies will be presented. The theory will be computationally embodied in different algorithmic approaches to the reinforcement learning. Finally, students will not only theoretically engage in the topic, but also gain practical experience in applying new techniques in different practical engineering situations spanning the range of applications such as mechanical, civil, and electrical application areas.

Assumed previous knowledge
Basic knowledge on machine and deep learning, e.g. Machine Learning in Engineering course (201900097) or equivalent.

Basic knowledge on probability theory, such as courses Probabilty Theory and Statistics (201400221), Statistics and Probability (191506103), or similar.
Participating study
Master Mechanical Engineering
Required materials
Course material
Lecture slides and videos
Recommended materials
Course material
Richard Sutton and Andrew Barto, Reinforcement Learning: An Introduction, MIT Press, Cambridge, MA, 2018
Instructional modes
Presence dutyYes

Presence dutyYes

Project unsupervised
Presence dutyYes

Presence dutyYes

Assignment(s) and oral exam

Kies de Nederlandse taal