Thorsten ready for a deluge of data
A deluge of data confronts the world today. No wonder data science is now regarded as a very important field. Thorsten Schmidt-Dumont, a PhD graduate in Industrial Engineering, will have the world at his feet when he enters this exciting field next year as a postdoctoral fellow. From being uncertain about what to study at University, to getting a sought-after position took a few years, but everything fell into place for him.
Thorsten’s story begins in Namibia, his home country. “I was very uncertain about my future career and what to study,” he says. “I only knew I did not want to do accountancy. Although did well in it, I found it very boring. I did some aptitude tests and through Stellenbosch University I became aware of Industrial Engineering. I found this very interesting and was impressed by the wide variety of different fields and career paths within this discipline. I liked the fact that it had a business and an engineering side.
“My mother was a Matie, so Stellenbosch University was my first choice, although I applied to other universities as well. I was thrilled when I was accepted at Stellenbosch. The first couple of years were challenging, but not too difficult. In my third year, I was exposed to a module in Operations Research presented by Prof Jan van Vuuren. His passion for the subject ignited a passion for it in me as well. Already at the end of my third year I discussed a possibility to do my final-year project in this field with Prof van Vuuren as my supervisor. My love for this subject was cemented in my final year when I had another module in Operations Research. I realised I truly loved the ‘hard’ numerical side of Industrial Engineering instead of the ‘softer’ business side.
“After graduating, I continued with a master’s under the supervision of Prof Van Vuuren. The focus during my master’s degree was on controlling traffic flow along highways, focusing specifically on controlling the number of vehicles allowed onto the highway at on-ramps. Instead of formulating rules for controlling the traffic flow myself, I applied reinforcement learning, a machine learning approach. My master’s was upgraded to a PhD, for which Mrs Megan Bruwer acted as co-promoter. For this upgrade, I added another angle to my thesis, which involved a forward-looking approach, namely not performing on-ramp metering with traffic lights, but rather by sending direct instructions to autonomous vehicles. This means that you do not have to maintain the infrastructure such as traffic lights, but you can communicate with autonomous vehicles through existing mobile communication networks.
“During the year of my upgrade, I was appointed as a part-time junior lecturer helping with the supervision of final-year projects. I also presented Industrial Programming to second-year students.
“I have been appointed as a postdoc in machine learning applications at the Stellenbosch University Department of Industrial Engineering for three years, focusing in data science applications within the field of Industrial Engineering. I will work closely with the incumbent of the new Voigt Chair in Data Science, Prof Andries Engelbrecht. One major aspect of this postdoc will be to create visibility of machine learning and data science activity on campus, to prospective postgraduates and beyond,” he explains.
This young Namibian has certainly found his niche in Stellenbosch and at Stellenbosch University. “In my second year I started cycling, especially mountain biking. It’s fantastic! Stellenbosch has some of the best and most scenic routes. It is an ideal sport to keep fit and to keep a clear head. I also enjoy the wine farms in the area tremendously. Coming from Namibia, which has a very strong beer culture, I really enjoyed being introduced to the wine culture the first time when I came to Stellenbosch. I find it very interesting and thoroughly enjoy a glass of wine,” he concludes.
Title of Thesis: Reinforcement learning for the control of traffic flow on highways.
Traffic congestion has become a significant problem around the world. Due to spatial limitations, highway capacity expansion is not always a feasible solution. Ramp metering and variable speed limits are the best-known highway traffic control measures. A novel approach towards simultaneously solving the associated control problems by reinforcement learning was proposed. Furthermore, a novel method of ramp metering by autonomous vehicles, where instructions may be provided to autonomous vehicles, was developed. These approaches were evaluated in a simulated real-world scenario of a section of the N1 highway outbound out of Cape Town, and significant improvements in the traffic flow were observed.
Share this post: