"Machine learning and computer vision are currently among the most exciting subject areas and fields of technology with an incredibly large potential for application and innovation in a wide range of areas of technology, business and culture."
A. Schreyer: Congratulations on your appointment as Professor of Computer Vision and Machine Learning at the Faculty of Engineering at HTWK Leipzig. Can you briefly explain the areas of responsibility associated with your professorship?
Prof Fuchs: First of all, thank you very much for the congratulations - even though I have already held the professorship for a year and am therefore already somewhat immersed in everyday life, the appointment naturally marks something very special for me. If only because I can now work as a professor in exactly the same place where I myself immersed myself in engineering and engineering science problems as a student almost exactly 20 years ago. I am now responsible for subject areas that played almost no role for graduates like me back then, but which - I am pretty sure - many future graduates will encounter in one way or another in their professional lives.
Whether it's news on television, podcasts, social media, radio programmes or magazine articles - whenever the often overused term artificial intelligence (AI) comes up today, the topic of machine learning is usually not far behind. In principle, it's quite simple: machine learning is about teaching a machine or computer to solve a problem without explicitly formulating the necessary rules for the solution, i.e. without having to programme them. Instead, the computer is presented with as much data as possible and given a strategy for finding the solution on its own. What sounds so simple is of course not so easy to implement in practice and has therefore only been used in a few fields of application for a long time. However, in recent years in particular, ML methods have opened up completely new possibilities for data analysis and led to impressive applications.
There is particular potential for modern machine learning methods in applications with very complex solution spaces, where classic rule-based programming quickly reaches its limits. This applies to human language, for example, but above all to processing and analysing image data. And that brings us to the topic of computer vision. This is about giving computers eyes through cameras, so to speak, and extracting information from the images recorded in the process. Certain cameras can capture images extremely quickly compared to the human eye. Others convert not only the range of visible light into digital signals, but also areas of the electromagnetic spectrum that are completely inaccessible to the human eye. This results in a wide range of potential applications for computer vision systems, but first and foremost a huge flood of data and, consequently, a very complex decision space from which the desired information must be extracted. Nowadays, it is generally the interplay of classic algorithms and deep learning methods developed by engineers that leads to the solution of the problem in the respective application.
So I don't think it's an exaggeration to say that both machine learning and computer vision are currently among the most exciting subject areas and fields of technology with an incredibly large potential for application and innovation in the most diverse areas of technology, business and culture. In this respect, I am delighted to be able to represent precisely these areas in both teaching and research at our University of Applied Sciences.
For me, the challenge, or rather my aspiration, is above all to systematise the diverse, complex and sometimes rapidly changing topics and to convey them with modern tools in an appropriate balance of theory and practical implementation. I also hope that I will be able to get the students at least a little enthusiastic about my own research work so that one or two of them will join my research team in the future. I teach the relevant topics in the compulsory subjects Machine Learning I and Computer Vision I as well as in the compulsory elective module Machine Learning II in the Bachelor's programme. In the Master's programme, these are the compulsory subjects Computer Vision II and CV/ML Applications in Embedded Systems and the compulsory elective subjects CV/ML Advanced and Camera-based Applications.
A. Schreyer: How did you decide to specialise in this area of research and teaching? Did you already know what path you wanted to take before you started your studies?
Prof Fuchs: I really enjoyed playing computers when I was young. In practice, however, in the mid-1990s I had to spend at least 80% of my time trying to get the PC to work at all. Sometimes the hardware was the problem, sometimes the operating system. I think the decisive factor here was that this was still in a time before Google & Co. and I therefore had to deal very intensively with the respective problems. This was often tedious, but not boring, because the intensive involvement with the technology and the interaction between hardware and software was of course fun. This was certainly one aspect that led me to study electrical engineering here at the HTWK. At the same time, I didn't yet know where my journey would take me after graduation. During my studies, I was particularly interested in topics relating to microcontrollers, signal processing and communications engineering. Because here I was suddenly back in my world of the interplay of hardware, software and, of course, algorithms.
It was algorithms that have practically stayed with me ever since, in a wide variety of areas. Above all, I wanted and still want to understand very precisely how certain algorithms work, why algorithms are the way they are and what exactly is behind their formulas. This interest and a corresponding opportunity enabled me to work on my dissertation as a visiting scientist at the Max Planck Institute for Human Cognitive and Brain Sciences and ultimately complete my doctorate at Ilmenau University of Technology. I really enjoyed this scientific work - and it was here at the latest that I realised that I would like to build up my own research profile and actively shape and take responsibility for interdisciplinary projects with industry and other scientists.
In addition to co-founding the Laboratory for Biosignal Processing working group here at the HTWK Leipzig research centre as an important step towards building this profile, the main thing was to identify a suitable field of activity, a niche so to speak, with the opportunity to establish unique selling points with regard to research and development projects. I finally found what I was looking for in the non-contact, mostly camera-based recording of signals in a wide variety of application areas, e.g. in medicine and health, for monitoring tasks in industry, for process analyses and even sports technologies. This all happened at a time when topics relating to machine learning were experiencing immense technological progress, not only but above all in relation to image and video data processing - they were therefore also playing an increasingly important role in our problem solving. In this respect, this focus ultimately led me to machine learning methods.
A. Schreyer: What skills and interests do you think students who decide to study electrical engineering and information technology should have?
Prof. Fuchs: An affinity for understanding physical relationships and their description using mathematical methods is certainly of great importance for all of the specialisations that can be chosen during the course of this degree. Basic programming skills will also be a great help from the very first day of the degree programme and should therefore be acquired consistently by then at the latest. Students should definitely have an interest in familiarising themselves with the functional principles of established and, of course, new technologies. This not only helps during their studies, but also with a view to their later professional life. If you delve into a very specific problem, you can learn a lot and often draw conclusions about other areas. At the same time, curiosity about the inner workings of technical solutions develops the necessary creativity and an extensive portfolio of methods and tools to be able to solve even complex problems in a goal-oriented manner later on.
No less important, however, are the skills required to describe technical solutions and problems adequately and with the necessary care, both qualitatively and quantitatively - if this is difficult at the beginning, that is quite natural. But you should actively work on constantly improving your own skills in this respect. Because in the end, it's not just about developing great new solutions, but also about communicating them in a variety of ways.
A. Schreyer: What new projects would you like to realise in the future?
Prof Fuchs: Basically, I enjoy making new applications possible through technical solutions. In this respect, I always keep my eyes open to the fields of application in which machine learning and computer vision technologies can open up new possibilities. In recent years, for example, this has been achieved in the field of sports technologies, for health applications and also for special industrial sectors. All of these fields of application will be consolidated and deepened in the coming years, which will of course be accompanied by corresponding third-party funding projects. For example, I am delighted that we will be able to launch a new two-year project from January 2023 to develop a new technology for analysing competitions in canoe racing. It is also worth mentioning in this context that such projects regularly provide interesting opportunities for students to get involved in internships, bachelor's and master's theses.
In addition to subject areas that I have already addressed directly, I also see interesting starting points for interdisciplinary, application-orientated research projects, particularly within the HTWK. The first concrete preparatory work is already taking place in this regard as well, in order to hopefully soon conduct research across disciplines and faculties with third-party funding.
A no less important project is the forthcoming establishment of the Computer Vision and Machine Learning Lab. In the CVML lab, students will not only have the opportunity to try out algorithms and applications with common industrial cameras, but also to work with very modern technologies such as event cameras and hyperspectral cameras. Of course, the laboratory will also be equipped with modern computing technology so that students will be able to develop and train extremely complex artificial neural networks. With regard to camera technologies in particular, I am also hoping for synergies with other areas such as mechanical engineering or civil engineering - because these technologies allow us to visualise certain invisible things and, above all, analyse them automatically! I think that there is still great potential here to tackle a lot of new projects in the future.
