RWTH Talent Talks

As part of this year's RWTH Talent Talks, the master's theses funded by the Hans Hermann Voss Foundation were presented last week. The theses presented in the following were funded within the framework of the so-called "Speed Funds". The purpose of the Speed Funds is to support research work on solving application-oriented and interdisciplinary problems by students in scientific and technical subjects.

Rofaida Daabis

In her work "Reinforcement Learning for Pick-and-Place Tasks in the RoboCup Logistics League" Daabis uses Rainforcement Learning to realize manipulation tasks in the RoboCup Logistics League (RCLL for short). In RCLL, on the one hand, cylindrical base products have to be picked up and set down and, on the other hand, colored rings have to be stacked on the lids of the base products. For the realization robotic arms are used which are operated with the sensor data of a 3D camera which is statically located at the edge of the work surface. Their work especially investigates the pick-and-place problem with robotic arms based on the RCLL problem. In addition, multi-arm coordination is also important so that collision-free motions can occur in confined spaces.

Mohammed Hossein

Due to the limitations caused by the von Neumann bottleneck in the data flow between CPUs and memory, conventional computer systems are reaching their limits. To overcome the challenge, a computing-in-memory (CIM for short) computing paradigm has been proposed that integrates the computational unit directly with memory. In his paper, "Highly lntegrated Computing-in-Memory Platform for Energy-Efficient Signal Processing," Hossein presents the development and implementation of the NeuroIoT platform. The NeuroIoT platform is a printed circuit board that is a CIM system designed specifically for use in the Internet of Things. NeuroIoT represents a significant advance in hardware platforms for memristive crossbars and enables the realization of machine learning applications in the Internet of Things domain. The platform shows tremendous potential in applications that require precision and accuracy, as it can accurately measure resistance values over a wide range with minimal variability during an experiment.

Dominik Kus

In his paper, "Ensemble Learning to the Rescue: Improving Machine Learning-based Industrial Intrusion Detection," Kus focuses on the potential and challenges of industrial intrusion detection systems (IIDS for short) as a means of securing industrial control systems. Due to the rapid digitization of industrial processes, information technology has penetrated much of the critical infrastructure. Systems that our society depends on are becoming vulnerable to cyberattacks. An IIDS passively monitors network traffic for anomalies and attacks and triggers an alert to the operator if there is potentially malicious activity. Initially, Ku's work involved evaluating the capability of IIDS against new attacks and developing a methodology. To increase the overall detection performance of potential attacks, several independent IIDS were combined into one system. Various ensemble approaches were used to ensure detection of previously unknown attacks.

Tabib Ibne Mazhar

Due to the complexity and control flow variations of processes in healthcare, optimizing healthcare services is a challenging task. In his work, "Comparative Process Mining in Healthcare," Mazhar explores the use of process mining to improve patient treatment pathways and increase operational efficiency. Process mining is a technique that transforms recorded event data into insights and measures, enabling optimization of processes and prediction of future actions. Comparative process mining techniques provide insight into healthcare processes, enabling optimization of patient care, increased productivity, and reduced costs. The method compares healthcare processes by analyzing the probabilities associated with a process model and event log to assess their degree of compliance. The joint project between RWTH Aachen University and Queensland University of Technology aims to improve patient outcomes and increase the overall quality of hospital care.

Asad Tariq

In the project "Dealing with Missing Data for Process Mining in Healthcare", Tariq developed techniques that enable the statically or causally based comparison of healthcare processes with each other or with healthcare outcomes. Healthcare systems in Germany and other countries face significant challenges in managing the impact of an aging population with complex and costly healthcare needs. Process mining involves process analysis techniques and tools for effective process management that can provide detailed analysis of the behavior of operational processes. Since process mining techniques are always based on data, models have been created for large groups depending on the accuracy and scale of the data. Previous techniques do not deal with the problem of missing data, which can lead to misleading process models. That is why a framework has been developed that provides a consistent estimate of the desired relationship when the conditions are met.