Research Associate in "Hybrid model based reinforcement learning" (m/w/d)
Research Framework
Hybrid model-based reinforcement learning (RL) is a type of RL that combines model-based RL with model-free RL. Hybrid model-based RL algorithms (HMRLAs) typically learn a model of the environment and then use that model to plan the agent's actions. However, they also use model-free RL techniques to learn from experience and to adapt to changes in the environment. Hybrid model-based RL algorithms have a number of advantages over both model-based RL and model-free RL. However designing HMRLAs can be challenging since it is often difficult to find a balance between efficiency and robustness. In this regard, with the aim of increasing robustness, resiliency and trust in RL algorithms it is quite advantageous to incorporate a dynamical system formulation in the design of model based RLs. This is specifically important in case of certain applications that adhere to certain well known physical/chemical/biological laws.
In this context, we are seeking motivated researchers to join our team in exploring the interesting field of "Hybrid model-based reinforcement learning". This research endeavor represents a critical step towards developing trustable, resilient and efficient RL algorithms that have widespread impact in various domains of applications such as robotics, autonomous driving, cancer biology, epidemiology etc.
Task Description
The research compiles from the following list of tasks.
- Developing novel model based RL methods and comparing them with the state of the art.
- Mathematically investigate how the above developed model based RL can be made less sensitive to errors in model specification by considering more dynamic updates in the value function.
- Derive mathematical guarantees for robustness and resiliency. Special emphasis must be provided for the case of ‘catastrophic forgetting’ during online-learning.
- Develop connections to stochastic optimal control problems and associated HJB equations.
- Collaborating closely with academic researchers who are specialized in one or more areas such as stochastic control and machine learning.
- Apply the above developed schemes to specific problems in the domains of Biology, Processes engineering and Autonomous driving.
Qualification
- Above average university degree in mathematics and control
- Knowledge of at least one programming language: Matlab, Python, C++ is expected
- Knowledge in dynamical systems and probability theory
- Proficiency in English or / and German is essential
- Highly motivated, eager to work within a team or independently.
Application procedure and deadline:
- Applications must include the following elements (as a single PDF file):
- Cover letter with a brief description of why you want to pursue research studies, about what your academic interests are, and how they relate to your previous studies and future goals
- CV including your relevant professional experience and knowledge
- Copies of diplomas and grades from previous university studies
- Two references
- List of publications
Send an email with the required documents to the address: @mec-apps@mv.uni-kl.de. The application deadline is 28. February 2025.
Interested candidates are encouraged to apply promptly, as applications will be processed as received, and the position may be filled before the deadline.
For more information, visit our webpage: https://mv.rptu.de/fgs/mec/research